Posts in category: Google AI

Posted by Romina Stella, Product Manager, Google Translate

Advances on neural machine translation (NMT) have enabled more natural and fluid translations, but they still can reflect the societal biases and stereotypes of the data on which they’re trained. As such, it is an ongoing goal at Google to develop innovative techniques to reduce gender bias in machine translation, in alignment with our AI Principles.

One research area has been using context from surrounding sentences or passages to improve gender accuracy. This is a challenge because traditional NMT methods translate sentences individually, but gendered information is not always explicitly stated in each individual sentence. For example, in the following passage in Spanish (a language where subjects aren’t always explicitly mentioned), the first sentence refers explicitly to Marie Curie as the subject, but the second one doesn’t explicitly mention the subject. In isolation, this second sentence could refer to a person of any gender. When translating to English, however, a pronoun needs to be picked, and the information needed for an accurate translation is in the first sentence.

Advancing translation techniques beyond single sentences requires new metrics for measuring progress and new datasets with the most common context-related errors. Adding to this challenge is the fact that translation errors related to gender (such as picking the correct pronoun or having gender agreement) are particularly sensitive, because they may directly refer to people and how they self identify.

To help facilitate progress against the common challenges on contextual translation (e.g., pronoun drop, gender agreement and accurate possessives), we are releasing the Translated Wikipedia Biographies dataset, which can be used to evaluate the gender bias of translation models. Our intent with this release is to support long-term improvements on ML systems focused on pronouns and gender in translation by providing a benchmark in which translations’ accuracy can be measured pre- and post-model changes.

A Source of Common Translation Errors
Because they are well-written, geographically diverse, contain multiple sentences, and refer to subjects in the third person (and so contain plenty of pronouns), Wikipedia biographies offer a high potential for common translation errors associated with gender. These often occur when articles refer to a person explicitly in early sentences of a paragraph, but there is no explicit mention of the person in later sentences. Some examples:

Building the Dataset
The Translated Wikipedia Biographies dataset has been designed to analyze common gender errors in machine translation, such as those illustrated above. Each instance of the dataset represents a person (identified in the biographies as feminine or masculine), a rock band or a sports team (considered genderless). Each instance is represented by a long text translation of 8 to 15 connected sentences referring to that central subject (the person, rock band, or sports team). Articles are written in native English and have been professionally translated to Spanish and German. For Spanish, translations were optimized for pronoun-drop, so the same set could be used to analyze pro-drop (Spanish → English) and gender agreement (English → Spanish).

The dataset was built by selecting a group of instances that has equal representation across geographies and genders. To do this, we extracted biographies from Wikipedia according to occupation, profession, job and/or activity. To ensure an unbiased selection of occupations, we chose nine occupations that represented a range of stereotypical gender associations (either feminine, masculine, or neither) based on Wikipedia statistics. Then, to mitigate any geography-based bias, we divided all these instances based on geographical diversity. For each occupation category, we looked to have one candidate per region (using regions from census.gov as a proxy of geographical diversity). When an instance was associated with a region, we checked that the selected person had a relevant relationship with a country that belongs to a designated region (nationality, place of birth, lived for a big portion of their life, etc.). By using this criteria, the dataset contains entries about individuals from more than 90 countries and all regions of the world.

Although gender is non-binary, we focused on having equal representation of “feminine” and “masculine” entities. It’s worth mentioning that because the entities are represented as such on Wikipedia, the set doesn’t include individuals that identify as non-binary, as, unfortunately, there are not enough instances currently represented in Wikipedia to accurately reflect the non-binary community. To label each instance as “feminine” or “masculine” we relied on the biographical information from Wikipedia, which contained gender-specific references to the person (she, he, woman, son, father, etc.).

After applying all these filters, we randomly selected an instance for each occupation-region-gender triplet. For each occupation, there are two biographies (one masculine and one feminine), for each of the seven geographic regions.

Finally, we added 12 instances with no gender. We picked rock bands and sports teams because they are usually referred to by non-gendered third person pronouns (such as “it” or singular “they”). The purpose of including these instances is to study over triggering, i.e., when models learn that they are rewarded for producing gender-specific pronouns, leading them to produce these pronouns in cases where they shouldn’t.

Results and Applications
This dataset enables a new method of evaluation for gender bias reduction in machine translations (introduced in a previous post). Because each instance refers to a subject with a known gender, we can compute the accuracy of the gender-specific translations that refer to this subject. This computation is easier when translating into English (cases of languages with pro-drop or neutral pronouns) since computation is mainly based on gender-specific pronouns in English. In these cases, the gender datasets have resulted in a 67% reduction in errors on context-aware models vs. previous models. As mentioned before, the neutral entities have allowed us to discover cases of over triggering like the usage of feminine or masculine pronouns to refer to genderless entities. This new dataset also enables new research directions into the performance of different models across types of occupations or geographic regions.

As an example, the dataset allowed us to discover the following improvements in an excerpt of the translated biography of Marie Curie from Spanish.

Translation result with the previous NMT model.

Translation result with the new contextual model.

Conclusion
This Translated Wikipedia Biographies dataset is the result of our own studies and work on identifying biases associated with gender and machine translation. This set focuses on a specific problem related to gender bias and doesn’t aim to cover the whole problem. It’s worth mentioning that by releasing this dataset, we don’t aim to be prescriptive in determining what’s the optimal approach to address gender bias. This contribution aims to foster progress on this challenge across the global research community.

Acknowledgements
The datasets were built with help from Anja Austermann, Melvin Johnson, Michelle Linch, Mengmeng Niu, Mahima Pushkarna, Apu Shah, Romina Stella, and Kellie Webster.

Read More

Posted by Andrew Carroll, Product Manager and Cory McLean, Software Engineer, Google Health

Each person’s genome, which collectively encodes the biochemical machinery they are born with, is composed of over 3 billion letters of DNA. However, only a small subset of the genome (~4-5 million positions) varies between two people. Nonetheless, each person’s unique genome interacts with the environment they experience to determine the majority of their health outcomes. A key method of understanding the relationship between genetic variants and traits is a genome-wide association study (GWAS), in which each genetic variant present in a cohort is individually examined for correlation with the trait of interest. GWAS results can be used to identify and prioritize potential therapeutic targets by identifying genes that are strongly associated with a disease of interest, and can also be used to build a polygenic risk score (PRS) to predict disease predisposition based on the combined influence of variants present in an individual. However, while accurate measurement of traits in an individual (called phenotyping) is essential to GWAS, it often requires painstaking expert curation and/or subjective judgment calls.

In “Large-scale machine learning-based phenotyping significantly improves genomic discovery for optic nerve head morphology”, we demonstrate how using machine learning (ML) models to classify medical imaging data can be used to improve GWAS. We describe how models can be trained for phenotypes to generate trait predictions and how these predictions are used to identify novel genetic associations. We then show that the novel associations discovered improve PRS accuracy and, using glaucoma as an example, that the improvements for anatomical eye traits relate to human disease. We have released the model training code and detailed documentation for its use on our Genomics Research GitHub repository.

Identifying genetic variants associated with eye anatomical traits
Previous work has demonstrated that ML models can identify eye diseases, skin diseases, and abnormal mammogram results with accuracy approaching or exceeding state-of-the-art methods by domain experts. Because identifying disease is a subset of phenotyping, we reasoned that ML models could be broadly used to improve the speed and quality of phenotyping for GWAS.

To test this, we chose a model that uses a fundus image of the eye to accurately predict whether a patient should be referred for assessment for glaucoma. This model uses the fundus images to predict the diameters of the optic disc (the region where the optic nerve connects to the retina) and the optic cup (a whitish region in the center of the optic disc). The ratio of the diameters of these two anatomical features (called the vertical cup-to-disc ratio, or VCDR) correlates strongly with glaucoma risk.

A representative retinal fundus image showing the vertical cup-to-disc ratio, which is an important diagnostic measurement for glaucoma.

We applied this model to predict VCDR in all fundus images from individuals in the UK Biobank, which is the world’s largest dataset available to researchers worldwide for health-related research in the public interest, containing extensive phenotyping and genetic data for ~500,000 pseudonymized (the UK Biobank’s standard for de-identification) individuals. We then performed GWAS in this dataset to identify genetic variants that are associated with the model-based predictions of VCDR.

Applying a VCDR prediction model trained on clinical data to generate predicted values for VCDR to enable discovery of genetic associations for the VCDR trait.

The ML-based GWAS identified 156 distinct genomic regions associated with VCDR. We compared these results to a VCDR GWAS conducted by another group on the same UK Biobank data, Craig et al. 2020, where experts had painstakingly labeled all images for VCDR. The ML-based GWAS replicates 62 of the 65 associations found in Craig et al., which indicates that the model accurately predicts VCDR in the UK Biobank images. Additionally, the ML-based GWAS discovered 93 novel associations.

Number of statistically significant GWAS associations discovered by exhaustive expert labeling approach (Craig et al., left), and by our ML-based approach (right), with shared associations in the middle.

The ML-based GWAS improves polygenic model predictions
To validate that the novel associations discovered in the ML-based GWAS are biologically relevant, we developed independent PRSes using the Craig et al. and ML-based GWAS results, and tested their ability to predict human-expert-labeled VCDR in a subset of UK Biobank as well as a fully independent cohort (EPIC-Norfolk). The PRS developed from the ML-based GWAS showed greater predictive ability than the PRS built from the expert labeling approach in both datasets, providing strong evidence that the novel associations discovered by the ML-based method influence VCDR biology, and suggesting that the improved phenotyping accuracy (i.e., more accurate VCDR measurement) of the model translates into a more powerful GWAS.

The correlation between a polygenic risk score (PRS) for VCDR generated from the ML-based approach and the exhaustive expert labeling approach (Craig et al.). In these plots, higher values on the y-axis indicate a greater correlation and therefore greater prediction from only the genetic data. [* — p ≤ 0.05; *** — p ≤ 0.001]

As a second validation, because we know that VCDR is strongly correlated with glaucoma, we also investigated whether the ML-based PRS was correlated with individuals who had either self-reported that they had glaucoma or had medical procedure codes suggestive of glaucoma or glaucoma treatment. We found that the PRS for VCDR determined using our model predictions were also predictive of the probability that an individual had indications of glaucoma. Individuals with a PRS 2.5 or more standard deviations higher than the mean were more than 3 times as likely to have glaucoma in this cohort. We also observed that the VCDR PRS from ML-based phenotypes was more predictive of glaucoma than the VCDR PRS produced from the extensive manual phenotyping.

The odds ratio of glaucoma (self-report or ICD code) stratified by the PRS for VCDR determined using the ML-based phenotypes (in standard deviations from the mean). In this plot, the y-axis shows the probability that the individual has glaucoma relative to the baseline rate (represented by the dashed line). The x-axis shows standard deviations from the mean for the PRS. Data are visualized as a standard box plot, which illustrates values for the mean (the orange line), first and third quartiles, and minimum and maximum.

Conclusion
We have shown that ML models can be used to quickly phenotype large cohorts for GWAS, and that these models can increase statistical power in such studies. Although these examples were shown for eye traits predicted from retinal imaging, we look forward to exploring how this concept could generally apply to other diseases and data types.

Acknowledgments
We would like to especially thank co-author Dr. Anthony Khawaja of Moorfields Eye Hospital for contributing his extensive medical expertise. We also recognize the efforts of Professor Jamie Craig and colleagues for their exhaustive labeling of UK Biobank images, which allowed us to make comparisons with our method. Several authors of that work, as well as Professor Stuart MacGregor and collaborators in Australia and at Max Kelsen have independently replicated these findings, and we value these scientific contributions as well. Last, this work summarizes the work of the following Google contributors, who we would like to thank: Babak Alipanahi, Farhad Hormozdiari, Babak Behsaz, Justin Cosentino, Zachary R. McCaw, Emanuel Schorsch, D. Sculley, Elizabeth H. Dorfman, Sonia Phene, Naama Hammel, Andrew Carroll, and Cory Y. McLean

Read More

Posted by Jarrod McClean, Staff Research Scientist and Hsin-Yuan (Robert) Huang¹, Intern, Google Quantum AI

Quantum computing has rapidly advanced in both theory and practice in recent years, and with it the hope for the potential impact in real applications. One key area of interest is how quantum computers might affect machine learning. We recently demonstrated experimentally that quantum computers are able to naturally solve certain problems with complex correlations between inputs that can be incredibly hard for traditional, or “classical”, computers. This suggests that learning models made on quantum computers may be dramatically more powerful for select applications, potentially boasting faster computation, better generalization on less data, or both. Hence it is of great interest to understand in what situations such a “quantum advantage” might be achieved.

The idea of quantum advantage is typically phrased in terms of computational advantages. That is, given some task with well defined inputs and outputs, can a quantum computer achieve a more accurate result than a classical machine in a comparable runtime? There are a number of algorithms for which quantum computers are suspected to have overwhelming advantages, such as Shor’s factoring algorithm for factoring products of large primes (relevant to RSA encryption) or the quantum simulation of quantum systems. However, the difficulty of solving a problem, and hence the potential advantage for a quantum computer, can be greatly impacted by the availability of data. As such, understanding when a quantum computer can help in a machine learning task depends not only on the task, but also the data available, and a complete understanding of this must include both.

In “Power of data in quantum machine learning”, published in Nature Communications, we dissect the problem of quantum advantage in machine learning to better understand when it will apply. We show how the complexity of a problem formally changes with the availability of data, and how this sometimes has the power to elevate classical learning models to be competitive with quantum algorithms. We then develop a practical method for screening when there may be a quantum advantage for a chosen set of data embeddings in the context of kernel methods. We use the insights from the screening method and learning bounds to introduce a novel method that projects select aspects of feature maps from a quantum computer back into classical space. This enables us to imbue the quantum approach with additional insights from classical machine learning that shows the best empirical separation in quantum learning advantages to date.

Computational Power of Data
The idea of quantum advantage over a classical computer is often framed in terms of computational complexity classes. Examples such as factoring large numbers and simulating quantum systems are classified as bounded quantum polynomial time (BQP) problems, which are those thought to be handled more easily by quantum computers than by classical systems. Problems easily solved on classical computers are called bounded probabilistic polynomial (BPP) problems.

We show that learning algorithms equipped with data from a quantum process, such as a natural process like fusion or chemical reactions, form a new class of problems (which we call BPP/Samp) that can efficiently perform some tasks that traditional algorithms without data cannot, and is a subclass of the problems efficiently solvable with polynomial sized advice (P/poly). This demonstrates that for some machine learning tasks, understanding the quantum advantage requires examination of available data as well.

Geometric Test for Quantum Learning Advantage
Informed by the results that the potential for advantage changes depending on the availability of data, one may ask how a practitioner can quickly evaluate if their problem may be well suited for a quantum computer. To help with this, we developed a workflow for assessing the potential for advantage within a kernel learning framework. We examined a number of tests, the most powerful and informative of which was a novel geometric test we developed.

In quantum machine learning methods, such as quantum neural networks or quantum kernel methods, a quantum program is often divided into two parts, a quantum embedding of the data (an embedding map for the feature space using a quantum computer), and the evaluation of a function applied to the data embedding. In the context of quantum computing, quantum kernel methods make use of traditional kernel methods, but use the quantum computer to evaluate part or all of the kernel on the quantum embedding, which has a different geometry than a classical embedding. It was conjectured that a quantum advantage might arise from the quantum embedding, which might be much better suited to a particular problem than any accessible classical geometry.

We developed a quick and rigorous test that can be used to quickly compare a particular quantum embedding, kernel, and data set to a range of classical kernels and assess if there is any opportunity for quantum advantage across, e.g., possible label functions such as those used for image recognition tasks. We define a geometric constant g, which quantifies the amount of data that could theoretically close that gap, based on the geometric test. This is an extremely useful technique for deciding, based on data constraints, if a quantum solution is right for the given problem.

Projected Quantum Kernel Approach
One insight revealed by the geometric test, was that existing quantum kernels often suffered from a geometry that was easy to best classically because they encouraged memorization, instead of understanding. This inspired us to develop a projected quantum kernel, in which the quantum embedding is projected back to a classical representation. While this representation is still hard to compute with a classical computer directly, it comes with a number of practical advantages in comparison to staying in the quantum space entirely.

Geometric quantity g, which quantifies the potential for quantum advantage, depicted for several embeddings, including the projected quantum kernel introduced here.

By selectly projecting back to classical space, we can retain aspects of the quantum geometry that are still hard to simulate classically, but now it is much easier to develop distance functions, and hence kernels, that are better behaved with respect to modest changes in the input than was the original quantum kernel. In addition the projected quantum kernel facilitates better integration with powerful non-linear kernels (like a squared exponential) that have been developed classically, which is much more challenging to do in the native quantum space.

This projected quantum kernel has a number of benefits over previous approaches, including an improved ability to describe non-linear functions of the existing embedding, a reduction in the resources needed to process the kernel from quadratic to linear with the number of data points, and the ability to generalize better at larger sizes. The kernel also helps to expand the geometric g, which helps to ensure the greatest potential for quantum advantage.

Data Sets Exhibit Learning Advantages
The geometric test quantifies potential advantage for all possible label functions, however in practice we are most often interested in specific label functions. Using learning theoretic approaches, we also bound the generalization error for specific tasks, including those which are definitively quantum in origin. As the advantage of a quantum computer relies on its ability to use many qubits simultaneously but previous approaches scale poorly in number of qubits, it is important to verify the tasks at reasonably large qubit sizes ( > 20 ) to ensure a method has the potential to scale to real problems. For our studies we verified up to 30 qubits, which was enabled by the open source tool, TensorFlow-Quantum, enabling scaling to petaflops of compute.

Interestingly, we showed that many naturally quantum problems, even up to 30 qubits, were readily handled by classical learning methods when sufficient data were provided. Hence one conclusion is that even for some problems that look quantum, classical machine learning methods empowered by data can match the power of quantum computers. However, using the geometric construction in combination with the projected quantum kernel, we were able to construct a data set that exhibited an empirical learning advantage for a quantum model over a classical one. Thus, while it remains an open question to find such data sets in natural problems, we were able to show the existence of label functions where this can be the case. Although this problem was engineered and a quantum computational advantage would require the embeddings to be larger and more challenging, this work represents an important step in understanding the role data plays in quantum machine learning.

Prediction accuracy as a function of the number of qubits (n) for a problem engineered to maximize the potential for learning advantage in a quantum model. The data is shown for two different sizes of training data (N).

For this problem, we scaled up the number of qubits (n) and compared the prediction accuracy of the projected quantum kernel to existing kernel approaches and the best classical machine learning model in our dataset. Moreover, a key takeaway from these results is that although we showed the existence of datasets where a quantum computer has an advantage, for many quantum problems, classical learning methods were still the best approach. Understanding how data can affect a given problem is a key factor to consider when discussing quantum advantage in learning problems, unlike traditional computation problems for which that is not a consideration.

Conclusions
When considering the ability of quantum computers to aid in machine learning, we have shown that the availability of data fundamentally changes the question. In our work, we develop a practical set of tools for examining these questions, and use them to develop a new projected quantum kernel method that has a number of advantages over existing approaches. We build towards the largest numerical demonstration to date, 30 qubits, of potential learning advantages for quantum embeddings. While a complete computational advantage on a real world application remains to be seen, this work helps set the foundation for the path forward. We encourage any interested readers to check out both the paper and related TensorFlow-Quantum tutorials that make it easy to build on this work.

Acknowledgements
We would like to acknowledge our co-authors on this paper — Michael Broughton, Masoud Mohseni, Ryan Babbush, Sergio Boixo, and Hartmut Neven, as well as the entirety of the Google Quantum AI team. In addition, we acknowledge valuable help and feedback from Richard Kueng, John Platt, John Preskill, Thomas Vidick, Nathan Wiebe, Chun-Ju Wu, and Balint Pato.

¹Current affiliation — Institute for Quantum Information and Matter and Department of Computing and Mathematical Sciences, Caltech, Pasadena, CA, USA^↩

Read More

Posted by Emily Knapp and Tim Herrmann, Program Managers

This week marks the start of the 2021 Conference on Computer Vision and Pattern Recognition (CVPR 2021), the premier annual computer vision event consisting of the main conference, workshops and tutorials. As a leader in computer vision research and a Champion Level Sponsor, Google will have a strong presence at CVPR 2021, with over 70 publications accepted, along with the organization of and participation in multiple workshops and tutorials.

If you are participating in CVPR this year, please visit our virtual booth to learn about Google research into the next generation of intelligent systems that utilize the latest machine learning techniques applied to various areas of machine perception.

You can also learn more about our research being presented at CVPR 2021 in the list below (Google affiliations in bold).

Organizing Committee Members

General Chair: Rahul Sukthankar
Finance Chair: Ramin Zabih
Workshop Chair: Caroline Pantofaru
Area Chairs: Chen Sun, Golnaz Ghiasi, Jonathan Barron, Kostas Rematas, Negar Rostamzadeh, Noah Snavely, Sanmi Koyejo, Tsung-Yi Lin

Publications

Cross-Modal Contrastive Learning for Text-to-Image Generation (see the blog post)
Han Zhang, Jing Yu Koh, Jason Baldridge, Honglak Lee*, Yinfei Yang

Learning Graph Embeddings for Compositional Zero-Shot Learning
Muhammad Ferjad Naeem, Yongqin Xian, Federico Tombari, Zeynep Akata

SPSG: Self-Supervised Photometric Scene Generation From RGB-D Scans
Angela Dai, Yawar Siddiqui, Justus Thies, Julien Valentin, Matthias Nießner

3D-MAN: 3D Multi-Frame Attention Network for Object Detection
Zetong Yang*, Yin Zhou, Zhifeng Chen, Jiquan Ngiam

MIST: Multiple Instance Spatial Transformer
Baptiste Angles, Yuhe Jin, Simon Kornblith, Andrea Tagliasacchi, Kwang Moo Yi

OCONet: Image Extrapolation by Object Completion
Richard Strong Bowen*, Huiwen Chang, Charles Herrmann*, Piotr Teterwak*, Ce Liu, Ramin Zabih

Ranking Neural Checkpoints
Yandong Li, Xuhui Jia, Ruoxin Sang, Yukun Zhu, Bradley Green, Liqiang Wang, Boqing Gong

LipSync3D: Data-Efficient Learning of Personalized 3D Talking Faces From Video Using Pose and Lighting Normalization
Avisek Lahiri, Vivek Kwatra, Christian Frueh, John Lewis, Chris Bregler

Differentiable Patch Selection for Image Recognition
Jean-Baptiste Cordonnier*, Aravindh Mahendran, Alexey Dosovitskiy, Dirk Weissenborn, Jakob Uszkoreit, Thomas Unterthiner

HumanGPS: Geodesic PreServing Feature for Dense Human Correspondences
Feitong Tan, Danhang Tang, Mingsong Dou, Kaiwen Guo, Rohit Pandey, Cem Keskin, Ruofei Du, Deqing Sun, Sofien Bouaziz, Sean Fanello, Ping Tan, Yinda Zhang

VIP-DeepLab: Learning Visual Perception With Depth-Aware Video Panoptic Segmentation (see the blog post)
Siyuan Qiao*, Yukun Zhu, Hartwig Adam, Alan Yuille, Liang-Chieh Chen

DeFMO: Deblurring and Shape Recovery of Fast Moving Objects
Denys Rozumnyi, Martin R. Oswald, Vittorio Ferrari, Jiri Matas, Marc Pollefeys

HDMapGen: A Hierarchical Graph Generative Model of High Definition Maps
Lu Mi, Hang Zhao, Charlie Nash, Xiaohan Jin, Jiyang Gao, Chen Sun, Cordelia Schmid, Nir Shavit, Yuning Chai, Dragomir Anguelov

Wide-Baseline Relative Camera Pose Estimation With Directional Learning
Kefan Chen, Noah Snavely, Ameesh Makadia

MobileDets: Searching for Object Detection Architectures for Mobile Accelerators
Yunyang Xiong, Hanxiao Liu, Suyog Gupta, Berkin Akin, Gabriel Bender, Yongzhe Wang, Pieter-Jan Kindermans, Mingxing Tan, Vikas Singh, Bo Chen

SMURF: Self-Teaching Multi-Frame Unsupervised RAFT With Full-Image Warping
Austin Stone, Daniel Maurer, Alper Ayvaci, Anelia Angelova, Rico Jonschkowski

Conceptual 12M: Pushing Web-Scale Image-Text Pre-Training To Recognize Long-Tail Visual Concepts
Soravit Changpinyo, Piyush Sharma, Nan Ding, Radu Soricut

Uncalibrated Neural Inverse Rendering for Photometric Stereo of General Surfaces
Berk Kaya, Suryansh Kumar, Carlos Oliveira, Vittorio Ferrari, Luc Van Gool

MeanShift++: Extremely Fast Mode-Seeking With Applications to Segmentation and Object Tracking
Jennifer Jang, Heinrich Jiang

Repopulating Street Scenes
Yifan Wang*, Andrew Liu, Richard Tucker, Jiajun Wu, Brian L. Curless, Steven M. Seitz, Noah Snavely

MaX-DeepLab: End-to-End Panoptic Segmentation With Mask Transformers (see the blog post)
Huiyu Wang*, Yukun Zhu, Hartwig Adam, Alan Yuille, Liang-Chieh Chen

IBRNet: Learning Multi-View Image-Based Rendering
Qianqian Wang, Zhicheng Wang, Kyle Genova, Pratul Srinivasan, Howard Zhou, Jonathan T. Barron, Ricardo Martin-Brualla, Noah Snavely, Thomas Funkhouser

From Points to Multi-Object 3D Reconstruction
Francis Engelmann*, Konstantinos Rematas, Bastian Leibe, Vittorio Ferrari

Learning Compositional Representation for 4D Captures With Neural ODE
Boyan Jiang, Yinda Zhang, Xingkui Wei, Xiangyang Xue, Yanwei Fu

Guided Integrated Gradients: An Adaptive Path Method for Removing Noise
Andrei Kapishnikov, Subhashini Venugopalan, Besim Avci, Ben Wedin, Michael Terry, Tolga Bolukbasi

De-Rendering the World’s Revolutionary Artefacts
Shangzhe Wu*, Ameesh Makadia, Jiajun Wu, Noah Snavely, Richard Tucker, Angjoo Kanazawa

Spatiotemporal Contrastive Video Representation Learning
Rui Qian, Tianjian Meng, Boqing Gong, Ming-Hsuan Yang, Huisheng Wang, Serge Belongie, Yin Cui

Decoupled Dynamic Filter Networks
Jingkai Zhou, Varun Jampani, Zhixiong Pi, Qiong Liu, Ming-Hsuan Yang

NeuralHumanFVV: Real-Time Neural Volumetric Human Performance Rendering Using RGB Cameras
Xin Suo, Yuheng Jiang, Pei Lin, Yingliang Zhang, Kaiwen Guo, Minye Wu, Lan Xu

Regularizing Generative Adversarial Networks Under Limited Data
Hung-Yu Tseng*, Lu Jiang, Ce Liu, Ming-Hsuan Yang, Weilong Yang

SceneGraphFusion: Incremental 3D Scene Graph Prediction From RGB-D Sequences
Shun-Cheng Wu, Johanna Wald, Keisuke Tateno, Nassir Navab, Federico Tombari

NeRV: Neural Reflectance and Visibility Fields for Relighting and View Synthesis
Pratul P. Srinivasan, Boyang Deng, Xiuming Zhang, Matthew Tancik, Ben Mildenhall, Jonathan T. Barron

Adversarially Adaptive Normalization for Single Domain Generalization
Xinjie Fan*, Qifei Wang, Junjie Ke, Feng Yang, Boqing Gong, Mingyuan Zhou

Adaptive Prototype Learning and Allocation for Few-Shot Segmentation
Gen Li, Varun Jampani, Laura Sevilla-Lara, Deqing Sun, Jonghyun Kim, Joongkyu Kim

Adversarial Robustness Across Representation Spaces
Pranjal Awasthi, George Yu, Chun-Sung Ferng, Andrew Tomkins, Da-Cheng Juan

Background Splitting: Finding Rare Classes in a Sea of Background
Ravi Teja Mullapudi, Fait Poms, William R. Mark, Deva Ramanan, Kayvon Fatahalian

Searching for Fast Model Families on Datacenter Accelerators
Sheng Li, Mingxing Tan, Ruoming Pang, Andrew Li, Liqun Cheng, Quoc Le, Norman P. Jouppi

Objectron: A Large Scale Dataset of Object-Centric Videos in the Wild With Pose Annotations (see the blog post)
Adel Ahmadyan, Liangkai Zhang, Jianing Wei, Artsiom Ablavatski, Matthias Grundmann

CutPaste: Self-Supervised Learning for Anomaly Detection and Localization
Chun-Liang Li, Kihyuk Sohn, Jinsung Yoon, Tomas Pfister

Nutrition5k: Towards Automatic Nutritional Understanding of Generic Food
Quin Thames, Arjun Karpur, Wade Norris, Fangting Xia, Liviu Panait, Tobias Weyand, Jack Sim

CReST: A Class-Rebalancing Self-Training Framework for Imbalanced Semi-Supervised Learning
Chen Wei*, Kihyuk Sohn, Clayton Mellina, Alan Yuille, Fan Yang

DetectoRS: Detecting Objects With Recursive Feature Pyramid and Switchable Atrous Convolution
Siyuan Qiao, Liang-Chieh Chen, Alan Yuille

DeRF: Decomposed Radiance Fields
Daniel Rebain, Wei Jiang, Soroosh Yazdani, Ke Li, Kwang Moo Yi, Andrea Tagliasacchi

Variational Transformer Networks for Layout Generation (see the blog post)
Diego Martin Arroyo, Janis Postels, Federico Tombari

Rich Features for Perceptual Quality Assessment of UGC Videos
Yilin Wang, Junjie Ke, Hossein Talebi, Joong Gon Yim, Neil Birkbeck, Balu Adsumilli, Peyman Milanfar, Feng Yang

Complete & Label: A Domain Adaptation Approach to Semantic Segmentation of LiDAR Point Clouds
Li Yi, Boqing Gong, Thomas Funkhouser

Neural Descent for Visual 3D Human Pose and Shape
Andrei Zanfir, Eduard Gabriel Bazavan, Mihai Zanfir, William T. Freeman, Rahul Sukthankar, Cristian Sminchisescu

GDR-Net: Geometry-Guided Direct Regression Network for Monocular 6D Object Pose Estimation
Gu Wang, Fabian Manhardt, Federico Tombari, Xiangyang Ji

Look Before You Speak: Visually Contextualized Utterances
Paul Hongsuck Seo, Arsha Nagrani, Cordelia Schmid

LASR: Learning Articulated Shape Reconstruction From a Monocular Video
Gengshan Yang*, Deqing Sun, Varun Jampani, Daniel Vlasic, Forrester Cole, Huiwen Chang, Deva Ramanan, William T. Freeman, Ce Liu

MoViNets: Mobile Video Networks for Efficient Video Recognition
Dan Kondratyuk, Liangzhe Yuan, Yandong Li, Li Zhang, Mingxing Tan, Matthew Brown, Boqing Gong

No Shadow Left Behind: Removing Objects and Their Shadows Using Approximate Lighting and Geometry
Edward Zhang, Ricardo Martin-Brualla, Janne Kontkanen, Brian Curless

On Robustness and Transferability of Convolutional Neural Networks
Josip Djolonga, Jessica Yung, Michael Tschannen, Rob Romijnders, Lucas Beyer, Alexander Kolesnikov, Joan Puigcerver, Matthias Minderer, Alexander D’Amour, Dan Moldovan, Sylvain Gelly, Neil Houlsby, Xiaohua Zhai, Mario Lucic

Robust and Accurate Object Detection via Adversarial Learning
Xiangning Chen, Cihang Xie, Mingxing Tan, Li Zhang, Cho-Jui Hsieh, Boqing Gong

To the Point: Efficient 3D Object Detection in the Range Image With Graph Convolution Kernels
Yuning Chai, Pei Sun, Jiquan Ngiam, Weiyue Wang, Benjamin Caine, Vijay Vasudevan, Xiao Zhang, Dragomir Anguelov

Bottleneck Transformers for Visual Recognition
Aravind Srinivas, Tsung-Yi Lin, Niki Parmar, Jonathon Shlens, Pieter Abbeel, Ashish Vaswani

Faster Meta Update Strategy for Noise-Robust Deep Learning
Youjiang Xu, Linchao Zhu, Lu Jiang, Yi Yang

Correlated Input-Dependent Label Noise in Large-Scale Image Classification
Mark Collier, Basil Mustafa, Efi Kokiopoulou, Rodolphe Jenatton, Jesse Berent

Learned Initializations for Optimizing Coordinate-Based Neural Representations
Matthew Tancik, Ben Mildenhall, Terrance Wang, Divi Schmidt, Pratul P. Srinivasan, Jonathan T. Barron, Ren Ng

Simple Copy-Paste Is a Strong Data Augmentation Method for Instance Segmentation
Golnaz Ghiasi, Yin Cui, Aravind Srinivas*, Rui Qian, Tsung-Yi Lin, Ekin D. Cubuk, Quoc V. Le, Barret Zoph

Function4D: Real-Time Human Volumetric Capture From Very Sparse Consumer RGBD Sensors
Tao Yu, Zerong Zheng, Kaiwen Guo, Pengpeng Liu, Qionghai Dai, Yebin Liu

RSN: Range Sparse Net for Efficient, Accurate LiDAR 3D Object Detection
Pei Sun, Weiyue Wang, Yuning Chai, Gamaleldin Elsayed, Alex Bewley, Xiao Zhang, Cristian Sminchisescu, Dragomir Anguelov

NeRF in the Wild: Neural Radiance Fields for Unconstrained Photo Collections
Ricardo Martin-Brualla, Noha Radwan, Mehdi S. M. Sajjadi, Jonathan T. Barron, Alexey Dosovitskiy, Daniel Duckworth

Robust Neural Routing Through Space Partitions for Camera Relocalization in Dynamic Indoor Environments
Siyan Dong, Qingnan Fan, He Wang, Ji Shi, Li Yi, Thomas Funkhouser, Baoquan Chen, Leonidas Guibas

Taskology: Utilizing Task Relations at Scale
Yao Lu, Sören Pirk, Jan Dlabal, Anthony Brohan, Ankita Pasad*, Zhao Chen, Vincent Casser, Anelia Angelova, Ariel Gordon

Omnimatte: Associating Objects and Their Effects in Video
Erika Lu, Forrester Cole, Tali Dekel, Andrew Zisserman, William T. Freeman, Michael Rubinstein

AutoFlow: Learning a Better Training Set for Optical Flow
Deqing Sun, Daniel Vlasic, Charles Herrmann, Varun Jampani, Michael Krainin, Huiwen Chang, Ramin Zabih, William T. Freeman, and Ce Liu

Unsupervised Multi-Source Domain Adaptation Without Access to Source Data
Sk Miraj Ahmed, Dripta S. Raychaudhuri, Sujoy Paul, Samet Oymak, Amit K. Roy-Chowdhury

Meta Pseudo Labels
Hieu Pham, Zihang Dai, Qizhe Xie, Minh-Thang Luong, Quoc V. Le

Spatially-Varying Outdoor Lighting Estimation From Intrinsics
Yongjie Zhu, Yinda Zhang, Si Li, Boxin Shi

Learning View-Disentangled Human Pose Representation by Contrastive Cross-View Mutual Information Maximization
Long Zhao*, Yuxiao Wang, Jiaping Zhao, Liangzhe Yuan, Jennifer J. Sun, Florian Schroff, Hartwig Adam, Xi Peng, Dimitris Metaxas, Ting Liu

Benchmarking Representation Learning for Natural World Image Collections
Grant Van Horn, Elijah Cole, Sara Beery, Kimberly Wilber, Serge Belongie, Oisin Mac Aodha

Scaling Local Self-Attention for Parameter Efficient Visual Backbones
Ashish Vaswani, Prajit Ramachandran, Aravind Srinivas, Niki Parmar, Blake Hechtman, Jonathon Shlens

KeypointDeformer: Unsupervised 3D Keypoint Discovery for Shape Control
Tomas Jakab*, Richard Tucker, Ameesh Makadia, Jiajun Wu, Noah Snavely, Angjoo Kanazawa

HITNet: Hierarchical Iterative Tile Refinement Network for Real-time Stereo Matching
Vladimir Tankovich, Christian Häne, Yinda Zhang, Adarsh Kowdle, Sean Fanello, Sofien Bouaziz

POSEFusion: Pose-Guided Selective Fusion for Single-View Human Volumetric Capture
Zhe Li, Tao Yu, Zerong Zheng, Kaiwen Guo, Yebin Liu

Workshops (only Google affiliations are noted)

Media Forensics
Organizers: Christoph Bregler

Safe Artificial Intelligence for Automated Driving
Invited Speakers: Been Kim

VizWiz Grand Challenge
Organizers: Meredith Morris

3D Vision and Robotics
Invited Speaker: Andy Zeng

New Trends in Image Restoration and Enhancement Workshop and Challenges on Image and Video Processing
Organizers: Ming-Hsuan Yang Program Committee: George Toderici, Ming-Hsuan Yang

2nd Workshop on Extreme Vision Modeling
Invited Speakers: Quoc Le, Chen Sun

First International Workshop on Affective Understanding in Video
Organizers: Gautam Prasad, Ting Liu

Adversarial Machine Learning in Real-World Computer Vision Systems and Online Challenges
Program Committee: Nicholas Carlini, Nicolas Papernot

Ethical Considerations in Creative Applications of Computer Vision
Invited Speaker: Alex Hanna Organizers: Negar Rostamzadeh, Emily Denton, Linda Petrini

Visual Question Answering Workshop
Invited Speaker: Vittorio Ferrari

Sixth International Skin Imaging Collaboration (ISIC) Workshop on Skin Image Analysis
Invited Speakers: Sandra Avila Organizers: Yuan Liu Steering Committee: Yuan Liu, Dale Webster

The 4th Workshop and Prize Challenge: Bridging the Gap between Computational Photography and Visual Recognition (UG2+) in Conjunction with IEEE CVPR 2021
Invited Speakers: Peyman Milanfar, Chelsea Finn

The 3rd CVPR Workshop on 3D Scene Understanding for Vision, Graphics, and Robotics
Invited Speaker: Andrea Tagliasacchi

Robust Video Scene Understanding: Tracking and Video Segmentation
Organizers: Jordi Pont-Tuset, Sergi Caelles, Jack Valmadre, Alex Bewley

4th Workshop and Challenge on Learned Image Compression
Invited Speaker: Rianne van den Berg Organizers: George Toderici, Lucas Theis, Johannes Ballé, Eirikur Agustsson, Nick Johnston, Fabian Mentzer

The Third Workshop on Precognition: Seeing Through the Future
Invited Speaker: Anelia Angelova
Organizers: Utsav Prabhu Program Committee: Chen Sun, David Ross

Computational Cameras and Displays
Organizers: Tali Dekel Keynote Talks: Paul Debevec Program Committee: Ayan Chakrabarti, Tali Dekel

2nd Embodied AI Workshop
Organizing Committee: Anthony Francis Challenge Organizers: Peter Anderson, Anthony Francis, Alex Ku, Alexander Toshev Scientific Advisory Board: Alexander Toshev

Responsible Computer Vision
Program Committee: Caroline Pantofaru, Utsav Prabhu, Susanna Ricco, Negar Rostamzadeh, Candice Schumann

Dynamic Neural Networks Meets Computer Vision
Invited Speaker: Azalia Mirhoseini

Interactive Workshop on Bridging the Gap between Subjective and Computational Measurements of Machine Creativity
Invited Speaker: David Bau

GAZE 2021: The 3rd International Workshop on Gaze Estimation and Prediction in the Wild
Organizer: Thabo Beeler Program Committee: Thabo Beeler

Sight and Sound
Organizers: William Freeman

Future of Computer Vision Datasets
Invited Speaker: Emily Denton, Caroline Pantofaru

Open World Vision
Invited Speakers: Rahul Sukthankar

The 3rd Workshop on Learning from Unlabeled Videos
Organizers: Anelia Angelova, Honglak Lee Program Committee: AJ Piergiovanni

4th International Workshop on Visual Odometry and Computer Vision Applications Based on Location Clues — With a Focus on Mobile Platform Applications
Organizers: Anelia Angelova

4th Workshop on Efficient Deep Learning for Computer Vision
Invited Speaker: Andrew Howard
Organizers: Pete Warden, Andrew Howard

Second International Workshop on Large Scale Holistic Video Understanding
Invited Speaker: Cordelia Schmid Program Committee: AJ Piergiovanni Organizers: David Ross

Neural Architecture Search 1st Lightweight NAS Challenge and Moving Beyond
Invited Speakers: Sara Sabour

The Second Workshop on Fair, Data-Efficient, and Trusted Computer Vision
Invited Speakers: Gaurav Aggarwal

The 17th Embedded Vision Workshop
General Chair: Anelia Angelova

8th Workshop on Fine-Grained Visual Categorization
Organizers: Christine Kaeser-Chen, Kimberly Wilber

AI for Content Creation
Invited Speaker: Tali Dekel, Jon Barron, Emily Denton Organizers: Deqing Sun

Frontiers of Monocular 3D Perception
Invited Speakers: Anelia Angelova, Cordelia Schmid, Noah Snavely

Beyond Fairness: Towards a Just, Equitable, and Accountable Computer Vision
Organizers: Emily Denton

The 1st Workshop on Future Video Conferencing
Invited Speakers: Chuo-Ling Chang, Sergi Caelles

Tutorials (only Google affiliations are noted)

Tutorial on Fairness Accountability Transparency and Ethics in Computer Vision
Organizer: Emily Denton

Data-Efficient Learning in An Imperfect World
Organizers: Boqing Gong, Ting Chen

Semantic Segmentation of Point Clouds: a Deep Learning Framework for Cultural Heritage
Invited Speaker: Manzil Zaheer

From VQA to VLN: Recent Advances in Vision-and-Language Research
Organizer: Peter Anderson

* Indicates work done while at Google

Read More

Yifeng Jiang, Research Intern and Jie Tan, Research Scientist, Robotics at Google

Simulation empowers various engineering disciplines to quickly prototype with minimal human effort. In robotics, physics simulations provide a safe and inexpensive virtual playground for robots to acquire physical skills with techniques such as deep reinforcement learning (DRL). However, as the hand-derived physics in simulations does not match the real world exactly, control policies trained entirely within simulation can fail when tested on real hardware — a challenge known as the sim-to-real gap or the domain adaptation problem. The sim-to-real gap for perception-based tasks (such as grasping) has been tackled using RL-CycleGAN and RetinaGAN, but there is still a gap caused by the dynamics of robotic systems. This prompts us to ask, can we learn a more accurate physics simulator from a handful of real robot trajectories? If so, such an improved simulator could be used to refine the robot controller using standard DRL training, so that it succeeds in the real world.

In our ICRA 2021 publication “SimGAN: Hybrid Simulator Identification for Domain Adaptation via Adversarial Reinforcement Learning”, we propose to treat the physics simulator as a learnable component that is trained by DRL with a special reward function that penalizes discrepancies between the trajectories (i.e., the movement of the robots over time) generated in simulation and a small number of trajectories that are collected on real robots. We use generative adversarial networks (GANs) to provide such a reward, and formulate a hybrid simulator that combines learnable neural networks and analytical physics equations, to balance model expressiveness and physical correctness. On robotic locomotion tasks, our method outperforms multiple strong baselines, including domain randomization.

A Learnable Hybrid Simulator
A traditional physics simulator is a program that solves differential equations to simulate the movement or interactions of objects in a virtual world. For this work, it is necessary to build different physical models to represent different environments – if a robot walks on a mattress, the deformation of the mattress needs to be taken into account (e.g., with the finite element method). However, due to the diversity of the scenarios that robots could encounter in the real world, it would be tedious (or even impossible) for such environment-specific modeling techniques, which is why it is useful to instead take an approach based on machine learning. Although simulators can be learned entirely from data, if the training data does not include a wide enough variety of situations, the learned simulator might violate the laws of physics (i.e., deviate from the real-world dynamics) if it needs to simulate situations for which it was not trained. As a result, the robot that is trained in such a limited simulator is more likely to fail in the real world.

To overcome this complication, we construct a hybrid simulator that combines both learnable neural networks and physics equations. Specifically, we replace what are often manually-defined simulator parameters — contact parameters (e.g., friction and restitution coefficients) and motor parameters (e.g., motor gains) — with a learnable simulation parameter function because the unmodeled details of contact and motor dynamics are major causes of the sim-to-real gap. Unlike conventional simulators in which these parameters are treated as constants, in the hybrid simulator they are state-dependent — they can change according to the state of the robot. For example, motors can become weaker at higher speed. These typically unmodeled physical phenomena can be captured using the state-dependent simulation parameter functions. Moreover, while contact and motor parameters are usually difficult to identify and subject to change due to wear-and-tear, our hybrid simulator can learn them automatically from data. For example, rather than having to manually specify the parameters of a robot’s foot against every possible surface it might contact, the simulation learns these parameters from training data.

Comparison between a conventional simulator and our hybrid simulator.

The other part of the hybrid simulator is made up of physics equations that ensure the simulation obeys fundamental laws of physics, such as conservation of energy, making it a closer approximation to the real world and thus reducing the sim-to-real gap.

In our earlier mattress example, the learnable hybrid simulator is able to mimic the contact forces from the mattress. Because the learned contact parameters are state-dependent, the simulator can modulate contact forces based on the distance and velocity of the robot’s feet relative to the mattress, mimicking the effect of the stiffness and damping of a deformable surface. As a result, we do not need to analytically devise a model specifically for deformable surfaces.

Using GANs for Simulator Learning
Successfully learning the simulation parameter functions discussed above would result in a hybrid simulator that can generate similar trajectories to the ones collected on the real robot. The key that enables this learning is defining a metric for the similarity between trajectories. GANs, initially designed to generate synthetic images that share the same distribution, or “style,” with a small number of real images, can be used to generate synthetic trajectories that are indistinguishable from real ones. GANs have two main parts, a generator that learns to generate new instances, and a discriminator that evaluates how similar the new instances are to the training data. In this case, the learnable hybrid simulator serves as the GAN generator, while the GAN discriminator provides the similarity scores.

The GAN discriminator provides the similarity metric that compares the movements of the simulated and the real robot.

Fitting parameters of simulation models to data collected in the real world, a process called system identification (SysID), has been a common practice in many engineering fields. For example, the stiffness parameter of a deformable surface can be identified by measuring the displacements of the surface under different pressures. This process is typically manual and tedious, but using GANs can be much more efficient. For example, SysID often requires a hand-crafted metric for the discrepancy between simulated and real trajectories. With GANs, such a metric is automatically learned by the discriminator. Furthermore, to calculate the discrepancy metric, conventional SysID requires pairing each simulated trajectory to a corresponding real-world one that is generated using the same control policy. Since the GAN discriminator takes only one trajectory as the input and calculates the likelihood that it is collected in the real world, this one-to-one pairing is not needed.

Using Reinforcement Learning (RL) to Learn the Simulator and Refine the Policy
Putting everything together, we formulate simulation learning as an RL problem. A neural network learns the state-dependent contact and motor parameters from a small number of real-world trajectories. The neural network is optimized to minimize the error between the simulated and the real trajectories. Note that it is important to minimize this error over an extended period of time — a simulation that accurately predicts a more distant future will lead to a better control policy. RL is well suited to this because it optimizes the accumulated reward over time, rather than just optimizing a single-step reward.

After the hybrid simulator is learned and becomes more accurate, we use RL again to refine the robot’s control policy within the simulation (e.g., walking across a surface, shown below).

Following the arrows clockwise: (upper left) recording a small number of robot’s failed attempts in the target domain (e.g., a real-world proxy in which the leg in red is modified to be much heavier than the source domain); (upper right) learning the hybrid simulator to match trajectories collected in the target domain; (lower right) refining control policies in this learned simulator; (lower left) testing the refined controller directly in the target domain.

Evaluation
Due to limited access to real robots during 2020, we created a second and different simulation (target domain) as a proxy of the real-world. The change of dynamics between the source and the target domains are large enough to approximate different sim-to-real gaps (e.g., making one leg heavier, walking on deformable surfaces instead of hard floor). We assessed whether our hybrid simulator, with no knowledge of these changes, could learn to match the dynamics in the target domain, and if the refined policy in this learned simulator could be successfully deployed in the target domain.

Qualitative results below show that simulation learning with less than 10 minutes of data collected in the target domain (where the floor is deformable) is able to generate a refined policy that performs much better for two robots with different morphologies and dynamics.

Comparison of performance between the initial and refined policy in the target domain (deformable floor) for the hopper and the quadruped robot.

Quantitative results below show that SimGAN outperforms multiple state-of-the-art baselines, including domain randomization (DR) and direct finetuning in target domains (FT).

Comparison of policy performance using different sim-to-real transfer methods in three different target domains for the Quadruped robot: locomotion on deformable surface, with weakened motors, and with heavier bodies.

Conclusion
The sim-to-real gap is one of the key bottlenecks that prevents robots from tapping into the power of reinforcement learning. We tackle this challenge by learning a simulator that can more faithfully model real-world dynamics, while using only a small amount of real-world data. The control policy that is refined in this simulator can be successfully deployed. To achieve this, we augment a classical physics simulator with learnable components, and train this hybrid simulator using adversarial reinforcement learning. To date we have tested its application to locomotion tasks, we hope to build on this general framework by applying it to other robot learning tasks, such as navigation and manipulation.

Acknowledgements
We’d like to thank our paper co-authors: Tingnan Zhang, Daniel Ho, Yunfei Bai, C. Karen Liu, and Sergey Levine. We would also like to thank the team members of Robotics at Google for discussions and feedback.

Read More

Posted by Candice Schumann and Susanna Ricco, Software Engineers, Google Research

In 2016, we introduced Open Images, a collaborative release of ~9 million images annotated with image labels spanning thousands of object categories and bounding box annotations for 600 classes. Since then, we have made several updates, including the release of crowdsourced data to the Open Images Extended collection to improve diversity of object annotations. While the labels provided with these datasets were expansive, they did not focus on sensitive attributes for people, which are critically important for many machine learning (ML) fairness tasks, such as fairness evaluations and bias mitigation. In fact, finding datasets that include thorough labeling of such sensitive attributes is difficult, particularly in the domain of computer vision.

Today, we introduce the More Inclusive Annotations for People (MIAP) dataset in the Open Images Extended collection. The collection contains more complete bounding box annotations for the person class hierarchy in 100k images containing people. Each annotation is also labeled with fairness-related attributes, including perceived gender presentation and perceived age range. With the increasing focus on reducing unfair bias as part of responsible AI research, we hope these annotations will encourage researchers already leveraging Open Images to incorporate fairness analysis in their research.

Examples of new boxes in MIAP. In each subfigure the magenta boxes are from the original Open Images dataset, while the yellow boxes are additional boxes added by the MIAP Dataset. Original photo credits — left: Boston Public Library; middle: jen robinson; right: Garin Fons; all used with permission under the CC- BY 2.0 license.

Annotations in Open Images
Each image in the original Open Images dataset contains image-level annotations that broadly describe the image and bounding boxes drawn around specific objects. To avoid drawing multiple boxes around the same object, less specific classes were temporarily pruned from the label candidate set, a process that we refer to as hierarchical de-duplication. For example, an image with labels animal, cat, and washing machine has bounding boxes annotated for cat and washing machine, but not for the redundant class animal.

The MIAP dataset addresses the five classes that are part of the person hierarchy in the original Open Images dataset: person, man, woman, boy, girl. The existence of these labels make the Open Images dataset uniquely valuable for research advancing responsible AI, allowing one to train a general person detector with access to gender- and age-range-specific labels for fairness analysis and bias mitigation.

However, we found that the combination of hierarchical de-duplication and societally imposed distinctions between woman/girl and man/boy introduced limitations in the original annotations. For example, if annotators were asked to draw boxes for the class girl, they would not draw a box around a boy in the image. They may or may not draw a box around a woman depending on their assessment of the age of the individual and their cultural understanding of the concept of girl. These decisions could be applied inconsistently between images, depending on the cultural background of the individual annotator, the appearance of an individual, and the context of the scene. Consequently, the bounding box annotations in some images were incomplete, with some people who appeared prominently not being annotated.

Annotations in MIAP
The new MIAP annotations are designed to address these limitations and fulfill the promise of Open Images as a dataset that will enable new advances in machine learning fairness research. Rather than asking annotators to draw boxes for the most specific class from the hierarchy (e.g., girl), we invert the procedure, always requesting bounding boxes for the gender- and age-agnostic person class. All person boxes are then separately associated with labels for perceived gender presentation (predominantly feminine, predominantly masculine, or unknown) and age presentation (young, middle, older, or unknown). We recognize that gender is not binary and that an individual’s gender identity may not match their perceived or intended gender presentation and, in an effort to mitigate the effects of unconscious bias on the annotations, we reminded annotators that norms around gender expression vary across cultures and have changed over time.

This procedure adds a significant number of boxes that were previously missing.

Over the 100k images that include people, the number of person bounding boxes have increased from ~358k to ~454k. The number of bounding boxes per perceived gender presentation and perceived age presentation increased consistently. These new annotations provide more complete ground truth for training a person detector as well as more accurate subgroup labels for incorporating fairness into computer vision research.

Comparison of number of person bounding boxes between the original Open Images and the new MIAP dataset.

Intended Use
We include annotations for perceived age range and gender presentation for person bounding boxes because we believe these annotations are necessary to advance the ability to better understand and work to mitigate and eliminate unfair bias or disparate performance across protected subgroups within the field of image understanding. We note that the labels capture the gender and age range presentation as assessed by a third party based on visual cues alone, rather than an individual’s self-identified gender or actual age. We do not support or condone building or deploying gender and/or age presentation classifiers trained from these annotations as we believe the risks associated with the use of these technologies outside fairness research outweigh any potential benefits.

Acknowledgements
The core team behind this work included Utsav Prabhu, Vittorio Ferrari, and Caroline Pantofaru. We would also like to thank Alex Hanna, Reena Jana, Alina Kuznetsova, Matteo Malloci, Stefano Pellegrini, Jordi Pont-Tuset, and Mahima Pushkarna, for their contributions to the project.

Read More

Arnab Bose and Yuheng Kuang, Staff Software Engineers, Robotics at Google

Disciplines in the natural sciences, social sciences, and medicine all have to grapple with how to evaluate and compare results within the context of the continually changing real world. In contrast, a significant body of machine learning (ML) research uses a different method that relies on the assumption of a fixed world: measure the performance of a baseline model on fixed data sets, then build a new model aimed at improving on the baseline, and evaluate its performance (on the same fixed data) by comparing its performance to the baseline.

Research into robotics systems and their applications to the real world requires a rethinking of this experiment design. Even in controlled robotic lab environments, it is possible that real-world changes cause the baseline model to perform inconsistently over time, making it unclear whether new models’ performance is an improvement compared to the baseline, or just the result of unintentional, random changes in the experiment setup. As robotics research advances into more complex and challenging real-world scenarios, there is a growing need for both understanding the impact of the ever-changing world on baselines and developing systematic methods to generate informative and clear results.

In this post, we demonstrate how robotics research, even in the relatively controlled environment of a lab, is meaningfully affected by changes in the environment, and discuss how to address this fundamental challenge using random assignment and A/B testing. Although these are classical research methods, they are not generally employed by default in robotics research — yet, they are critical to producing meaningful and measurable scientific results for robotics in real-world scenarios. Additionally, we cover the costs, benefits, and other considerations of using these methods.

The Ever-Changing Real World in Robotics
Even in a robotics lab environment, which is designed to minimize all changes that are not experimental conditions, it is notoriously difficult to set up a perfectly reproducible experiment. Robots get bumped and are subject to wear and tear, lighting changes affect perception, battery charge influences the torque applied to motors — all things that can affect results in ways large and small.

To illustrate this on real robot data, we collected success rate data on one of our simplest setups — moving identical foam dice from one bin to another. For this task, we ran about 33k task trials on two robots over more than five months with the same software and ML model, and took the overall success rate of the last two weeks as baseline. We then measured the historic performance over time in this “very well controlled” environment.

Video of a real robot completing the task: moving identical foam dice from one bin to another.

Given that we did not purposefully change anything during data collection, one would expect the success rate to be statistically similar over time. And yet, this is not what was observed.

The y-axis represents the 95% confidence interval of % change in success rate relative to baseline. If the confidence intervals contain zero, that indicates the success rate is statistically similar to the success rate of baseline. Confidence intervals were computed using Jackknife, with Cochran-Mantel-Haenszel correction to remove operator bias.

Using the sequential data from the plot above, one might conclude that the model ran during weeks 13-14 performed best and that ran during weeks 9-10 performed the worst. One might also expect most, if not all, of the confidence intervals above to contain 0, but only one did. Because no changes were made at any time during these trials, this example effectively demonstrates the impact of unintentional, random real-world changes on even very simple setups. It’s also worth noting that having more trials per experiment wouldn’t remove these differences, instead they will more likely produce a narrower confidence interval making the impact more obvious.

However, what happens when one uses random assignment to compare results, grouping the data randomly rather than sequentially? To answer this, we randomly assigned the above data to the same number of groups for comparison with the baseline. This is equivalent to performing A/B testing where all groups receive the same treatment.

Looking at the chart, we observe that the confidence intervals include zero, indicating success similar to the baseline, as expected.

We performed similar studies with a few other robotics tasks, comparing between sequential and random assignments. They all yielded similar results.

We see that even with no intentional changes, there are statistically significant differences observed for sequential assignment, while random assignment shows the expected result of no statistically significant differences.

Considerations for A/B testing in robotics
While it’s clear based on the above that A/B testing with random assignment is an effective way to control for the unexplainable variance of the real world in robotics, there are some considerations when adopting this approach. Here are several, along with their accompanying pros, cons, and solutions:

• Absolute vs relative performance: Each experiment needs to be measured against a baseline that is run concurrently. The relative performance metric between baseline and experiment is published with a confidence interval. The absolute performance metric (in baseline or experiment) is less informative, because it depends to an unknown degree on the state of the world when the measurement was taken. However, the statistical differences we’ve measured between the experiment and baseline are sound and robust to reproduction.
• Data efficiency: With this approach, the baseline always needs to run in parallel with the experimental conditions so they can be compared against each other. Although this may seem wasteful, it is worth the cost when compared against the drawbacks of making an invalid inference against a stale baseline. Furthermore, as the number of random assignment experiments scale up, we can use a single baseline arm with multiple simultaneous experiment arms across independent factors leveraging Google’s overlapping experiment infrastructure. Data efficiency improves with scale.
• Environmental biases: If there’s any external factor affecting performance overall (lighting, slicker surfaces, etc.), both the baseline and all experiment arms will encounter this factor with similar probability, so its effect will cancel if there’s no relative impact. If there is a correlation between environmental factors and experiment arms, this will show up as differences over time (each environmental factor accumulates in the episodes collected). This can substantially reduce or eliminate the need for effortful environmental resets, and lets us run lifelong experiments and still measure improvements across experimental arms.
• Human biases: One advantage of random assignment is a reduction in biases introduced by humans. Since human operators cannot know which data sample gets routed to which arm of the experiment, it is harder to have biased experimenters influence any particular outcome.

The Path Forward
The A/B testing experiment framework has been successfully used for a long time in many scientific disciplines to measure performance against changing, unpredictable real-world environments. In this blog post, we show that robotics research can benefit from using this same methodology: it improves the quality and confidence of research results, and avoids the impossible task of perfectly controlling all elements of a fundamentally changing environment. Doing this well requires infrastructure to continuously operate robots, collect data, and tools to make the statistical framework easily accessible to researchers.

Acknowledgements
Arnab Bose, Tuna Toksoz, Yuheng Kuang, Anthony Brohan, Razvan Sudulescu developed the experiment infrastructure and conducted the research. Matthieu Devin suggested the A/A analysis to showcase the differences using existing data. Special thanks to Bill Heavlin, Chris Harris, Vincent Vanhoucke who provided invaluable feedback and support to the work.

Read More

Posted by Joel Shor, Software Engineer, Google Research, Tokyo and Sachin Joglekar, Software Engineer, TensorFlow

Representation learning is a machine learning (ML) method that trains a model to identify salient features that can be applied to a variety of downstream tasks, ranging from natural language processing (e.g., BERT and ALBERT) to image analysis and classification (e.g., Inception layers and SimCLR). Last year, we introduced a benchmark for comparing speech representations and a new, generally-useful speech representation model (TRILL). TRILL is based on temporal proximity, and tries to map speech that occurs close together in time to a lower-dimensional embedding that captures temporal proximity in the embedding space. Since its release, the research community has used TRILL on a diverse set of tasks, such as age classification, video thumbnail selection, and language identification. However, despite achieving state-of-the-art performance, TRILL and other neural network-based approaches require more memory and take longer to compute than signal processing operations that deal with simple features, like loudness, average energy, pitch, etc.

In our recent paper “FRILL: A Non-Semantic Speech Embedding for Mobile Devices“, to appear at Interspeech 2021, we create a new model that is 40% the size of TRILL and and a feature set that can be computed over 32x faster on mobile phone, with an average decrease in accuracy of less than 2%. This marks an important step towards fully on-device applications of speech ML models, which will lead to better personalization, improved user experiences and greater privacy, an important aspect of developing AI responsibly. We release the code to create FRILL on github, and a pre-trained FRILL model on TensorFlow Hub.

FRILL: Smaller, Faster TRILL
The TRILL architecture is based on a modified version of ResNet50, an architecture that is computationally taxing for constrained hardware, like mobile phones or smart home devices. On the other hand, architectures like MobileNetV3 have been designed with hardware-aware AutoML to perform well on mobile devices. To take advantage of this, we leverage knowledge distillation to combine the benefits of MobileNetV3’s performance with TRILL’s representations.

In the distillation process, the smaller model (i.e., the “student”) tries to match the output of the larger model (“teacher”) on the AudioSet dataset. Whereas the original TRILL model learned its weights by optimizing a self-supervised loss that clustered audio segments close in time, the student model learns its weights through a fully-supervised loss that ignores temporal matching and instead tries to match TRILL outputs on the training data. The fully-supervised learning signal is often stronger than self-supervision, and allows us to train more quickly.

Knowledge distillation for non-semantic speech embeddings. The dashed line shows the student model output. The “teacher network” is the TRILL network, where “Layer 19” was the best-performing internal representation. The “Student Hyperparameters” on the left are the options explored in this study, the result of which are 144 distinct models. These models were trained with mean-squared error (MSE) to try to match TRILL’s Layer 19.

Choosing the Best Student Model
We perform distillation with a variety of student models, each trained with a specific combination of architecture choices (explained below). To measure each student model’s latency, we leverage TensorFlow Lite (TFLite), a framework that enables execution of TensorFlow models on edge devices. Each candidate model is first converted into TFLite’s flatbuffer format for 32-bit floating point inference and then sent to the target device (in this case, a Pixel 1) for benchmarking. These measurements help us to accurately assess the latency versus quality tradeoffs across all student models and to minimize the loss of quality in the conversion process.

Architecture Choices and Optimizations
We explored different neural network architectures and features that balance latency and accuracy — models with fewer parameters are usually smaller and faster, but have less representational power and therefore generate less generally-useful representations. We trained 144 different models across a number of hyperparameters, all based on the MobileNetV3 architecture:

1. MobileNetV3 size and width: MobileNetV3 was released in different sizes for use in different environments. The size refers to which MobileNetV3 architecture we used. The width, sometimes known as alpha, proportionally decreases or increases the number of filters in each layer. A width of 1.0 corresponds to the number of filters in the original paper.
2. Global average pooling: MobileNetV3 normally produces a set of two-dimensional feature maps. These are flattened, concatenated, and passed to the bottleneck layer. However, this bottleneck is often still too large to be computed quickly. We reduce the size of the bottleneck layer kernel by taking the global average of all ”pixels” in each output feature map. Our intuition is that the discarded temporal information is less important for learning a non-semantic speech representation due to the fact that relevant aspects of the signal are stable across time.
3. Bottleneck compression: A significant portion of the student model’s weights are located in the bottleneck layer. To reduce the size of this layer, we apply a compression operator based on singular value decomposition (SVD) that learns a low-rank approximation of the bottleneck weight matrix.
4. Quantization-aware training: Since the bottleneck layer has most of the model weights, we use quantization-aware training (QAT) to gradually reduce the numerical precision of the bottleneck weights during training. QAT allows the model to adjust to the lower numerical precision during training, instead of potentially causing performance degradation by introducing quantization after training finishes.

Results
We evaluated each of these models on the Non-Semantic Speech Benchmark (NOSS) and two new tasks — a challenging task to detect whether a speaker is wearing a mask and the human-noise subset of the Environment Sound Classification dataset, which includes labels like “coughing” and “sneezing”. After eliminating models that have strictly better alternatives, we are left with eight ”frontier” models on the quality vs. latency curve, which are the models that had no faster and better performance alternatives at a corresponding quality threshold or latency in our batch of 144 models. We plot the latency vs. quality curve of only these “frontier” models below, and we ignore models that are strictly worse.

Embedding quality and latency tradeoff. The x-axis represents the inference latency and the y-axis shows the difference in accuracy from TRILL’s performance, averaged across benchmark datasets.

FRILL is the best performing sub-10ms inference model, with an inference time of 8.5 ms on a Pixel 1 (about 32x faster than TRILL), and is also roughly 40% the size of TRILL. The frontier curve plateaus at about 10ms latency, which means that at low latency, one can achieve much better performance with minimal latency costs, while achieving improved performance at latencies beyond 10ms is more difficult. This supports our choice of experiment hyperparameters. FRILL’s per-task performance is shown in the table below.

Finally, we evaluate the relative contribution of each of our hyperparameters. We find that for our experiments, quantization-aware training, bottleneck compression and global average pooling most reduced the latency of the resulting models. At the same time bottleneck compression most reduced the quality of the resulting model, while pooling reduced the model performance the least. The architecture width parameter was an important factor in reducing the model size, with minimal performance degradation.

Linear regression weight magnitudes for predicting model quality, latency, and size. The weights indicate the expected impact of changing the input hyperparameter. A higher weight magnitude indicates a greater expected impact.

Our work is an important step in bringing the full benefits of speech machine learning research to mobile devices. We also provide our public model, corresponding model card, and evaluation code to help the research community responsibly develop even more applications for on-device speech representation research.

Acknowledgements
We’d like to thank our paper co-authors: Jacob Peplinski, Jake Garrison and Shwetak Patel. We’d like to thank Aren Jansen for his technical support on this project, Françoise Beaufays, and Tulsee Doshi for help open sourcing the model, and Google Research, Tokyo for logistical support.

Read More

Posted by Diego Martin Arroyo, Software Engineer and Federico Tombari, Research Scientist, Google Research

Information in a written document is not only conveyed by the meaning of the words contained in it, but also by the overall document layout. Layouts are commonly used to direct the order in which the reader parses a document to enable a better understanding (e.g., with columns or paragraphs), to provide helpful summaries (e.g., with titles) or for aesthetic purposes (e.g., when displaying advertisements).

While these design rules are easy to follow, it is difficult to explicitly define them without quickly needing to include exceptions or encountering ambiguous cases. This makes the automation of document design difficult, as any system with a hardcoded set of production rules will either be overly simplistic and thus incapable of producing original layouts (causing a lack of diversity in the layout of synthesized data), or too complex, with a large set of rules and their accompanying exceptions. In an attempt to solve this challenge, some have proposed machine learning (ML) techniques to synthesize document layouts. However, most ML-based solutions for automatic document design do not scale to a large number of layout components, or they rely on additional information for training, such as the relationships between the different components of a document.

In “Variational Transformer Networks for Layout Generation”, to be presented at CVPR 2021, we create a document layout generation system that scales to an arbitrarily large number of elements and does not require any additional information to capture the relationships between design elements. We use self-attention layers as building blocks of a variational autoencoder (VAE), which is able to model document layout design rules as a distribution, rather than using a set of predetermined heuristics, increasing the diversity of the generated layouts. The resulting Variational Transformer Network (VTN) model is able to extract meaningful relationships between the layout elements (paragraphs, tables, images, etc.), resulting in realistic synthetic documents (e.g., better alignment and margins). We show the effectiveness of this combination across different domains, such as scientific papers, UI layouts, and even furniture arrangements.

VAEs for Layout Generation
The ultimate goal of this system is to infer the design rules for a given type of layout from a collection of examples. If one considers these design rules as the distribution underlying the data, it is possible to use probabilistic models to discover it. We propose doing this with a VAE (widely used for tasks like image generation or anomaly detection), an autoencoder architecture that consists of two distinct subparts, the encoder and decoder. The encoder learns to compress the input to fewer dimensions, retaining only the necessary information to reconstruct the input, while the decoder learns to undo this operation. The compressed representation (also called the bottleneck) can be forced to behave like a known distribution (e.g., a uniform Gaussian). Feeding samples from this a priori distribution to the decoder segment of the network results in outputs similar to the training data.

An additional advantage of the VAE formulation is that it is agnostic to the type of operations used to implement the encoder and decoder segments. As such, we use self-attention layers (typically seen in Transformer architectures) to automatically capture the influence that each layout element has over the rest.

Transformers use self-attention layers to model long, sequenced relationships, often applied to an array of natural language understanding tasks, such as translation and summarization, as well as beyond the language domain in object detection or document layout understanding tasks. The self-attention operation relates every element in a sequence to every other and determines how they influence each other. This property is ideal to model relationships across different elements in a layout without the need for explicit annotations.

In order to synthesize new samples from these relationships, some approaches for layout generation [e.g., 1] and even for other domains [e.g., 2, 3] rely on greedy search algorithms, such as beam search, nucleus sampling or top-k sampling. Since these strategies are often based on exploration rules that tend to favor the most likely outcome at every step, the diversity of the generated samples is not guaranteed. However, by combining self-attention with the VAE’s probabilistic techniques, the model is able to directly learn a distribution from which it can extract new elements.

Modeling the Variational Bottleneck
The bottleneck of a VAE is commonly modeled as a vector representing the input. Since self-attention layers are a sequence-to-sequence architecture, i.e., a sequence of n input elements is mapped onto n output elements, the standard VAE formulation is difficult to apply. Inspired by BERT, we append an auxiliary token to the beginning of the sequence and treat it as the autoencoder bottleneck vector z. During training, the vector associated with this token is the only piece of information passed to the decoder, so the encoder needs to learn how to compress the entire document information in this vector. The decoder then learns to infer the number of elements in the document as well as the locations of each element in the input sequence from this vector alone. This strategy allows us to use standard techniques to regularize the bottleneck, such as the KL divergence.

Decoding
In order to synthesize documents with varying numbers of elements, the network needs to model sequences of arbitrary length, which is not trivial. While self-attention enables the encoder to adapt automatically to any number of elements, the decoder segment does not know the number of elements in advance. We overcome this issue by decoding sequences in an autoregressive way — at every step, the decoder produces an element, which is concatenated to the previously decoded elements (starting with the bottleneck vector z as input), until a special stop element is produced.

A visualization of our proposed architecture

Turning Layouts into Input Data
A document is often composed of several design elements, such as paragraphs, tables, images, titles, footnotes, etc. In terms of design, layout elements are often represented by the coordinates of their enclosing bounding boxes. To make this information easily digestible for a neural network, we define each element with four variables (x, y, width, height), representing the element’s location on the page (x, y) and size (width, height).

Results
We evaluate the performance of the VTN following two criteria: layout quality and layout diversity. We train the model on publicly available document datasets, such as PubLayNet, a collection of scientific papers with layout annotations, and evaluate the quality of generated layouts by quantifying the amount of overlap and alignment between elements. We measure how well the synthetic layouts resemble the training distribution using the Wasserstein distance over the distributions of element classes (e.g., paragraphs, images, etc.) and bounding boxes. In order to capture the layout diversity, we find the most similar real sample for each generated document using the DocSim metric, where a higher number of unique matches to the real data indicates a more diverse outcome.

We compare the VTN approach to previous works like LayoutVAE and Gupta et al. The former is a VAE-based formulation with an LSTM backbone, whereas Gupta et al. use a self-attention mechanism similar to ours, combined with standard search strategies (beam search). The results below show that LayoutVAE struggles to comply with design rules, like strict alignments, as in the case of PubLayNet. Thanks to the self-attention operation, Gupta et al. can model these constraints much more effectively, but the usage of beam search affects the diversity of the results.

We also explore the ability of our approach to learn design rules in other domains, such as Android UIs (RICO), natural scenes (COCO) and indoor scenes (SUN RGB-D). Our method effectively learns the design rules of these datasets and produces synthetic layouts of similar quality as the current state of the art and a higher degree of diversity.

Below are some examples of layouts produced by our method compared to existing methods. The design rules learned by the network (location, margins, alignment) resemble those of the original data and show a high degree of variability.

LayoutVAE
Gupta et al.
VTN

Conclusion
In this work we show the feasibility of using self-attention as part of the VAE formulation. We validate the effectiveness of this approach for layout generation, achieving state-of-the-art performance on various datasets and across different tasks. Our research paper also explores alternative architectures for the integration of self-attention and VAEs, exploring non-autoregressive decoding strategies and different types of priors, and analyzes advantages and disadvantages. The layouts produced by our method can help to create synthetic training data for downstream tasks, such as document parsing or automating graphic design tasks. We hope that this work provides a foundation for continued research in this area, as many subproblems are still not completely solved, such as how to suggest styles for the elements in the layout (text font, which image to choose, etc.) or how to reduce the amount of training data necessary for the model to generalize.

AcknowledgementsWe thank our co-author Janis Postels, as well as Alessio Tonioni and Luca Prasso for helping with the design of several of our experiments. We also thank Tom Small for his help creating the animations for this post.

Read More

Posted by AJ Maschinot, Senior Software Engineer and Jenny Huang, Product Manager, Google Research

In recent years, self-supervised representation learning, which is used in a variety of image and video tasks, has significantly advanced due to the application of contrastive learning. These contrastive learning approaches typically teach a model to pull together the representations of a target image (a.k.a., the “anchor”) and a matching (“positive”) image in embedding space, while also pushing apart the anchor from many non-matching (“negative”) images. Because labels are assumed to be unavailable in self-supervised learning, the positive is often an augmentation of the anchor, and the negatives are chosen to be the other samples from the training minibatch. However, because of this random sampling, false negatives, i.e., negatives generated from samples of the same class as the anchor, can cause a degradation in the representation quality. Furthermore, determining the optimal method to generate positives is still an area of active research.

In contrast to the self-supervised approach, a fully-supervised approach could use labeled data to generate positives from existing same-class examples, providing more variability in pretraining than could typically be achieved by simply augmenting the anchor. However, very little work has been done to successfully apply contrastive learning in the fully-supervised domain.

In “Supervised Contrastive Learning”, presented at NeurIPS 2020, we propose a novel loss function, called SupCon, that bridges the gap between self-supervised learning and fully supervised learning and enables contrastive learning to be applied in the supervised setting. Leveraging labeled data, SupCon encourages normalized embeddings from the same class to be pulled closer together, while embeddings from different classes are pushed apart. This simplifies the process of positive selection, while avoiding potential false negatives. Because it accommodates multiple positives per anchor, this approach results in an improved selection of positive examples that are more varied, while still containing semantically relevant information. SupCon also allows label information to play an active role in representation learning rather than restricting it to be used only in downstream training, as is the case for conventional contrastive learning. To the best of our knowledge, this is the first contrastive loss to consistently perform better on large-scale image classification problems than the common approach of using cross-entropy loss to train the model directly. Importantly, SupCon is straightforward to implement and stable to train, provides consistent improvement to top-1 accuracy for a number of datasets and architectures (including Transformer architectures), and is robust to image corruptions and hyperparameter variations.

Self-supervised (left) vs supervised (right) contrastive losses: The self-supervised contrastive loss contrasts a single positive for each anchor (i.e., an augmented version of the same image) against a set of negatives consisting of the entire remainder of the minibatch. The supervised contrastive loss considered in this paper, however, contrasts the set of all samples from the same class as positives against the negatives from the remainder of the batch.

The Supervised Contrastive Learning Framework
SupCon can be seen as a generalization of both the SimCLR and N-pair losses — the former uses positives generated from the same sample as that of the anchor, and the latter uses positives generated from different samples by exploiting known class labels. The use of many positives and many negatives for each anchor allows SupCon to achieve state-of-the-art performance without the need for hard negative mining (i.e., searching for negatives similar to the anchor), which can be difficult to tune properly.

SupCon subsumes multiple losses from the literature and is a generalization of the SimCLR and N-Pair losses.

This method is structurally similar to those used in self-supervised contrastive learning, with modifications for supervised classification. Given an input batch of data, we first apply data augmentation twice to obtain two copies, or “views,” of each sample in the batch (though one could create and use any number of augmented views). Both copies are forward propagated through an encoder network, and the resulting embedding is then L2-normalized. Following standard practice, the representation is further propagated through an optional projection network to help identify meaningful features. The supervised contrastive loss is computed on the normalized outputs of the projection network. Positives for an anchor consist of the representations originating from the same batch instance as the anchor or from other instances with the same label as the anchor; the negatives are then all remaining instances. To measure performance on downstream tasks, we train a linear classifier on top of the frozen representations.

Cross-entropy, self-supervised contrastive loss and supervised contrastive loss Left: The cross-entropy loss uses labels and a softmax loss to train a classifier. Middle: The self-supervised contrastive loss uses a contrastive loss and data augmentations to learn representations. Right: The supervised contrastive loss also learns representations using a contrastive loss, but uses label information to sample positives in addition to augmentations of the same image.

Key Findings
SupCon consistently boosts top-1 accuracy compared to cross-entropy, margin classifiers (with use of labels), and self-supervised contrastive learning techniques on CIFAR-10 and CIFAR-100 and ImageNet datasets. With SupCon, we achieve excellent top-1 accuracy on the ImageNet dataset with the ResNet-50 and ResNet-200 architectures. On ResNet-200, we achieve a top-1 accuracy of 81.4%, which is a 0.8% improvement over the state-of-the-art cross-entropy loss using the same architecture (which represents a significant advance for ImageNet). We also compared cross-entropy and SupCon on a Transformer-based ViT-B/16 model and found a consistent improvement over cross-entropy (77.8% versus 76% for ImageNet; 92.6% versus 91.6% for CIFAR-10) under the same data augmentation regime (without any higher-resolution fine-tuning).

The SupCon loss consistently outperforms cross-entropy with standard data augmentation strategies (AutoAugment, RandAugment and CutMix). We show top-1 accuracy for ImageNet, on ResNet-50, ResNet-101 and ResNet200.

We also demonstrate analytically that the gradient of our loss function encourages learning from hard positives and hard negatives. The gradient contributions from hard positives/negatives are large while those for easy positives/negatives are small. This implicit property allows the contrastive loss to sidestep the need for explicit hard mining, which is a delicate but critical part of many losses, such as triplet loss. See the supplementary material of our paper for a full derivation.

SupCon is also more robust to natural corruptions, such as noise, blur and JPEG compression. The mean Corruption Error (mCE) measures the average degradation in performance compared to the benchmark ImageNet-C dataset. The SupCon models have lower mCE values across different corruptions compared to cross-entropy models, showing increased robustness.

We show empirically that the SupCon loss is less sensitive than cross-entropy to a range of hyperparameters. Across changes in augmentations, optimizers, and learning rates, we observe significantly lower variance in the output of the contrastive loss. Moreover, applying different batch sizes while holding all other hyperparameters constant results in consistently better top-1 accuracy of SupCon to that of cross-entropy at each batch size.

Accuracy of cross-entropy and supervised contrastive loss as a function of hyperparameters and training data size, measured on ImageNet with a ResNet-50 encoder. Left: Boxplot showing Top-1 accuracy vs changes in augmentation, optimizer and learning rates. SupCon yields more consistent results across variations in each, which is useful when the best strategies are unknown a priori. Right: Top-1 accuracy as a function of batch size shows both losses benefit from larger batch sizes while SupCon has higher Top-1 accuracy, even when trained with small batch sizes.

Accuracy of supervised contrastive loss as a function of training duration and the temperature hyperparameter, measured on ImageNet with a ResNet-50 encoder. Left: Top-1 accuracy as a function of SupCon pre-training epochs. Right: Top-1 accuracy as a function of temperature during the pre-training stage for SupCon. Temperature is an important hyperparameter in contrastive learning and reducing sensitivity to temperature is desirable.

Broader Impact and Next Steps
This work provides a technical advancement in the field of supervised classification. Supervised contrastive learning can improve both the accuracy and robustness of classifiers with minimal complexity. The classic cross-entropy loss can be seen as a special case of SupCon where the views correspond to the images and the learned embeddings in the final linear layer corresponding to the labels. We note that SupCon benefits from large batch sizes, and being able to train the models on smaller batches is an important topic for future research.

Our Github repository includes Tensorflow code to train the models in the paper. Our pre-trained models are also released on TF-Hub.

Acknowledgements
The NeurIPS paper was jointly co-authored with Prannay Khosla, Piotr Teterwak, Chen Wang, Aaron Sarna, Yonglong Tian, Phillip Isola, Aaron Maschinot, Ce Liu, and Dilip Krishnan. Special thanks to Jenny Huang for leading the writing process for this blogpost.

Read More