Posts in category: Microsoft AI

Editor’s note: All papers referenced here represent collaborations throughout Microsoft and across academia and industry that include authors who contribute to Aether, the Microsoft internal advisory body for AI Ethics and Effects in Engineering and Research.

Artificial intelligence, like all tools we build, is an expression of human creativity. As with all creative expression, AI manifests the perspectives and values of its creators. A stance that encourages reflexivity among AI practitioners is a step toward ensuring that AI systems are human-centered, developed and deployed with the interests and well-being of individuals and society front and center. This is the focus of research scientists and engineers affiliated with Aether, the advisory body for Microsoft leadership on AI ethics and effects. Central to Aether’s work is the question of who we’re creating AI for—and whether we’re creating AI to solve real problems with responsible solutions. With AI capabilities accelerating, our researchers work to understand the sociotechnical implications and find ways to help on-the-ground practitioners envision and realize these capabilities in line with Microsoft AI principles.

The following is a glimpse into the past year’s research for advancing responsible AI with authors from Aether. Throughout this work are repeated calls for reflexivity in AI practitioners’ processes—that is, self-reflection to help us achieve clarity about who we’re developing AI systems for, who benefits, and who may potentially be harmed—and for tools that help practitioners with the hard work of uncovering assumptions that may hinder the potential of human-centered AI. The research discussed here also explores critical components of responsible AI, such as being transparent about technology limitations, honoring the values of the people using the technology, enabling human agency for optimal human-AI teamwork, improving effective interaction with AI, and developing appropriate evaluation and risk-mitigation techniques for multimodal machine learning (ML) models.

Considering who AI systems are for

The need to cultivate broader perspectives and, for society’s benefit, reflect on why and for whom we’re creating AI is not only the responsibility of AI development teams but also of the AI research community. In the paper “REAL ML: Recognizing, Exploring, and Articulating Limitations of Machine Learning Research,” the authors point out that machine learning publishing often exhibits a bias toward emphasizing exciting progress, which tends to propagate misleading expectations about AI. They urge reflexivity on the limitations of ML research to promote transparency about findings’ generalizability and potential impact on society—ultimately, an exercise in reflecting on who we’re creating AI for. The paper offers a set of guided activities designed to help articulate research limitations, encouraging the machine learning research community toward a standard practice of transparency about the scope and impact of their work.

Walk through REAL ML’s instructional guide and worksheet that help researchers with defining the limitations of their research and identifying societal implications these limitations may have in the practical use of their work.

Explore REAL ML

Despite many organizations formulating principles to guide the responsible development and deployment of AI, a recent survey highlights that there’s a gap between the values prioritized by AI practitioners and those of the general public. The survey, which included a representative sample of the US population, found AI practitioners often gave less weight than the general public to values associated with responsible AI. This raises the question of whose values should inform AI systems and shifts attention toward considering the values of the people we’re designing for, aiming for AI systems that are better aligned with people’s needs.

Related papers

• REAL ML: Recognizing, Exploring, and Articulating Limitations of Machine Learning Research
• How Different Groups Prioritize Ethical Values for Responsible AI

Creating AI that empowers human agency

Supporting human agency and emphasizing transparency in AI systems are proven approaches to building appropriate trust with the people systems are designed to help. In human-AI teamwork, interactive visualization tools can enable people to capitalize on their own domain expertise and let them easily edit state-of-the-art models. For example, physicians using GAM Changer can edit risk prediction models for pneumonia and sepsis to incorporate their own clinical knowledge and make better treatment decisions for patients.

A study examining how AI can improve the value of rapidly growing citizen-science contributions found that emphasizing human agency and transparency increased productivity in an online workflow where volunteers provide valuable information to help AI classify galaxies. When choosing to opt in to using the new workflow and receiving messages that stressed human assistance was necessary for difficult classification tasks, participants were more productive without sacrificing the quality of their input and they returned to volunteer more often.

Failures are inevitable in AI because no model that interacts with the ever-changing physical world can be complete. Human input and feedback are essential to reducing risks. Investigating reliability and safety mitigations for systems such as robotic box pushing and autonomous driving, researchers formalize the problem of negative side effects (NSEs), the undesirable behavior of these systems. The researchers experimented with a framework in which the AI system uses immediate human assistance in the form of feedback—either about the user’s tolerance for an NSE occurrence or their decision to modify the environment. Results demonstrate that AI systems can adapt to successfully mitigate NSEs from feedback, but among future considerations, there remains the challenge of developing techniques for collecting accurate feedback from individuals using the system.

The goal of optimizing human-AI complementarity highlights the importance of engaging human agency. In a large-scale study examining how bias in models influences humans’ decisions in a job recruiting task, researchers made a surprising discovery: when working with a black-box deep neural network (DNN) recommender system, people made significantly fewer gender-biased decisions than when working with a bag-of-words (BOW) model, which is perceived as more interpretable. This suggests that people tend to reflect and rely on their own judgment before accepting a recommendation from a system for which they can’t comfortably form a mental model of how its outputs are derived. Researchers call for exploring techniques to better engage human reflexivity when working with advanced algorithms, which can be a means for improving hybrid human-AI decision-making and mitigating bias. 

How we design human-AI interaction is key to complementarity and empowering human agency. We need to carefully plan how people will interact with AI systems that are stochastic in nature and present inherently different challenges than deterministic systems. Designing and testing human interaction with AI systems as early as possible in the development process, even before teams invest in engineering, can help avoid costly failures and redesign. Toward this goal, researchers propose early testing of human-AI interaction through factorial surveys, a method from the social sciences that uses short narratives for deriving insights about people’s perceptions.

But testing for optimal user experience before teams invest in engineering can be challenging for AI-based features that change over time. The ongoing nature of a person adapting to a constantly updating AI feature makes it difficult to observe user behavior patterns that can inform design improvements before deploying a system. However, experiments demonstrate the potential of HINT (Human-AI INtegration Testing), a framework for uncovering over-time patterns in user behavior during pre-deployment testing. Using HINT, practitioners can design test setup, collect data via a crowdsourced workflow, and generate reports of user-centered and offline metrics.

Check out the 2022 anthology of this annual workshop that brings human-computer interaction (HCI) and natural language processing (NLP) research together for improving how people can benefit from NLP apps they use daily.

HCI + NLP 2022 anthology

Related papers

• Assessing Human-AI Interaction Early through Factorial Surveys: A Study on the Guidelines for Human-AI Interaction
• Avoiding Negative Side Effects of Autonomous Systems in the Open World
• Interpretability, Then What? Editing Machine Learning Models to Reflect Human Knowledge and Values
• Investigations of Performance and Bias in Human-AI Teamwork in Hiring
• A new Workflow for Human-AI Collaboration in Citizen Science
• HINT: Integration Testing for AI-based Features with Humans in the Loop

Building responsible AI tools for foundation models

Although we’re still in the early stages of understanding how to responsibly harness the potential of large language and multimodal models that can be used as foundations for building a variety of AI-based systems, researchers are developing promising tools and evaluation techniques to help on-the-ground practitioners deliver responsible AI. The reflexivity and resources required for deploying these new capabilities with a human-centered approach are fundamentally compatible with business goals of robust services and products.

Natural language generation with open-ended vocabulary has sparked a lot of imagination in product teams. Challenges persist, however, including for improving toxic language detection; content moderation tools often over-flag content that mentions minority groups without respect to context while missing implicit toxicity. To help address this, a new large-scale machine-generated dataset, ToxiGen, enables practitioners to fine-tune pretrained hate classifiers for improving detection of implicit toxicity for 13 minority groups in both human- and machine-generated text.

Download the large-scale machine-generated ToxiGen dataset and install source code for fine-tuning toxic language detection systems for adversarial and implicit hate speech for 13 demographic minority groups. Intended for research purposes.

ToxiGen dataset

Multimodal models are proliferating, such as those that combine natural language generation with computer vision for services like image captioning. These complex systems can surface harmful societal biases in their output and are challenging to evaluate for mitigating harms. Using a state-of-the-art image captioning service with two popular image-captioning datasets, researchers isolate where in the system fairness-related harms originate and present multiple measurement techniques for five specific types of representational harm: denying people the opportunity to self-identify, reifying social groups, stereotyping, erasing, and demeaning.

The commercial advent of AI-powered code generators has introduced novice developers alongside professionals to large language model (LLM)-assisted programming. An overview of the LLM-assisted programming experience reveals unique considerations. Programming with LLMs invites comparison to related ways of programming, such as search, compilation, and pair programming. While there are indeed similarities, the empirical reports suggest it is a distinct way of programming with its own unique blend of behaviors. For example, additional effort is required to craft prompts that generate the desired code, and programmers must check the suggested code for correctness, reliability, safety, and security. Still, a user study examining what programmers value in AI code generation shows that programmers do find value in suggested code because it’s easy to edit, increasing productivity. Researchers propose a hybrid metric that combines functional correctness and similarity-based metrics to best capture what programmers value in LLM-assisted programming, because human judgment should determine how a technology can best serve us.

Related papers

• What is it like to program with artificial intelligence?
• ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detection
• Aligning Offline Metrics and Human Judgments of Value of AI-Pair Programmers
• Measuring Representational Harms in Image Captioning

Understanding and supporting AI practitioners

Organizational culture and business goals can often be at odds with what practitioners need for mitigating fairness and other responsible AI issues when their systems are deployed at scale. Responsible, human-centered AI requires a thoughtful approach: just because a technology is technically feasible does not mean it should be created.

Similarly, just because a dataset is available doesn’t mean it’s appropriate to use. Knowing why and how a dataset was created is crucial for helping AI practitioners decide on whether it should be used for their purposes and what its implications are for fairness, reliability, safety, and privacy. A study focusing on how AI practitioners approach datasets and documentation reveals current practices are informal and inconsistent. It points to the need for data documentation frameworks designed to fit within practitioners’ existing workflows and that make clear the responsible AI implications of using a dataset. Based on these findings, researchers iterated on Datasheets for Datasets and proposed the revised Aether Data Documentation Template.

Use this flexible template to reflect and help document underlying assumptions, potential risks, and implications of using your dataset.

Document your data

AI practitioners find themselves balancing the pressures of delivering to meet business goals and the time requirements necessary for the responsible development and evaluation of AI systems. Examining these tensions across three technology companies, researchers conducted interviews and workshops to learn what practitioners need for measuring and mitigating AI fairness issues amid time pressure to release AI-infused products to wider geographic markets and for more diverse groups of people. Participants disclosed challenges in collecting appropriate datasets and finding the right metrics for evaluating how fairly their system will perform when they can’t identify direct stakeholders and demographic groups who will be affected by the AI system in rapidly broadening markets. For example, hate speech detection may not be adequate across cultures or languages. A look at what goes into AI practitioners’ decisions around what, when, and how to evaluate AI systems that use natural language generation (NLG) further emphasizes that when practitioners don’t have clarity about deployment settings, they’re limited in projecting failures that could cause individual or societal harm. Beyond concerns for detecting toxic speech, other issues of fairness and inclusiveness—for example, erasure of minority groups’ distinctive linguistic expression—are rarely a consideration in practitioners’ evaluations.

Coping with time constraints and competing business objectives is a reality for teams deploying AI systems. There are many opportunities for developing integrated tools that can prompt AI practitioners to think through potential risks and mitigations for sociotechnical systems.

Related papers

• Deconstructing NLG Evaluation: Evaluation Practices, Assumptions, and Their Implications
• Understanding Machine Learning Practitioners’ Data Documentation Perceptions, Needs, Challenges, and Desiderata
• Assessing the Fairness of AI Systems: AI Practitioners’ Processes, Challenges, and Needs for Support

Thinking about it: Reflexivity as an essential for society and industry goals

As we continue to envision what all is possible with AI’s potential, one thing is clear: developing AI designed with the needs of people in mind requires reflexivity. We have been thinking about human-centered AI as being focused on users and stakeholders. Understanding who we are designing for, empowering human agency, improving human-AI interaction, and developing harm mitigation tools and techniques are as important as ever. But we also need to turn a mirror toward ourselves as AI creators. What values and assumptions do we bring to the table? Whose values get to be included and whose are left out? How do these values and assumptions influence what we build, how we build, and for whom? How can we navigate complex and demanding organizational pressures as we endeavor to create responsible AI? With technologies as powerful as AI, we can’t afford to be focused solely on progress for its own sake. While we work to evolve AI technologies at a fast pace, we need to pause and reflect on what it is that we are advancing—and for whom.

The post Advancing human-centered AI: Updates on responsible AI research appeared first on Microsoft Research.

Read More

High-throughput ab initio reaction mechanism exploration in the cloud with automated multi-reference validation

Jan P. Unsleber, Hongbin Liu, Leopold Talirz, Thomas Weymuth, Maximilian Mörchen, Adam Grofe, Dave Wecker, Christopher J. Stein, Ajay Panyala, Bo Peng, Karol Kowalski, Matthias Troyer, Markus Reiher

Quantum chemical calculations on atomistic systems have evolved into a standard approach to studying molecular matter. These calculations often involve a significant amount of manual input and specific process considerations, which could be automated and allow for further efficiencies. In our recent paper: High-throughput ab initio reaction mechanism exploration in the cloud with automated multi-reference validation, we present the AutoRXN workflow, an automated workflow for exploratory high-throughput electronic structure calculations of molecular systems. In this workflow, (i) density functional theory methods are exploited to deliver minimum and transition-state structures and corresponding energies and properties, (ii) coupled cluster calculations are then launched for optimized structures to provide more accurate energy and property estimates, and (iii) multi-reference diagnostics are evaluated to back check the coupled cluster results and subject them to automated multi-configurational calculations for potential multi-configurational cases. All calculations are carried out in a cloud environment and support massive computational campaigns. Key features of all components of the AutoRXN workflow are autonomy, stability, and minimum operator interference. We highlight the AutoRXN workflow at the example of an autonomous reaction mechanism exploration of the mode of action of a homogeneous catalyst for the asymmetric reduction of ketones.

Spotlight: On-Demand EVENT

Microsoft Research Summit 2022

On-Demand
Watch now to learn about some of the most pressing questions facing our research community and listen in on conversations with 120+ researchers around how to ensure new technologies have the broadest possible benefit for humanity.

Explore sessions

Disparate Impacts on Online Information Access during the COVID-19 Pandemic

Jina Suh, Eric Horvitz, Ryen W. White, Tim Althoff

Despite efforts to close the long-term and emergent health equity gap, studies during the COVID-19 pandemic show that socioeconomically and environmentally disadvantaged subpopulations have been disproportionately harmed by the disease[1]. Digital access to health services and information has also emerged as an important factor modulating health outcomes. During the pandemic, digital engagement in resources across health, educational, economic, and social needs became a necessity due to lockdown mandates and increased use of internet-based communication by public institutions. Unfortunately, disparities in digital access also reflect socioeconomic and environmental dimensions, which can lead to negative offline consequences, creating a “digital vicious cycle”[2]. Therefore, it is a public health priority to identify vulnerable populations and to understand potential barriers to critical digital resources.

In a new paper: Disparate Impacts on Online Information Access during the COVID-19 Pandemic, published in Nature Communications, researchers from Microsoft Research and the University of Washington have collaborated to harness the centrality of web search engines for online information access to observe digital disparities during the pandemic. They analyzed over 55 billion web search interactions on Bing during the pandemic across 25,150 U.S. ZIP codes to reveal that socioeconomic and environmental factors are associated with the differential use of digital resources across different communities – even if they were digitally connected.

DeepSpeed Data Efficiency library: Towards less data, faster training, and higher model quality

DeepSpeed Team, Andrey Proskurin

DeepSpeed has released a new Data Efficiency library to optimize deep learning training efficiency and cost. The library offers new algorithms on efficient data sampling/scheduling via curriculum learning and efficient data routing via random layerwise token dropping, together with composable and customizable library support. The library greatly reduces training cost while maintaining model quality (1.5-2x less data and time for GPT-3/BERT pretraining), or further improves model quality under the same training cost (>1 point gain for GPT-3-1.3B zero/few-shot evaluation). The code is open-sourced at https://github.com/microsoft/DeepSpeed.

You can learn more in our blog post and in the papers below.

Research Fellows Program at Microsoft Research India – Apply now

The Research Fellows Program at Microsoft Research India is now accepting applications for Fall 2023. This is an opportunity to work with world-class researchers on state-of-the-art technology. The program prepares students for careers in research, engineering, and entrepreneurship, while pushing the frontiers of computer science and technology. Previous Research Fellows have contributed to all aspects of the research lifecycle, spanning ideation, implementation, evaluation, and deployment.

Selected candidates spend one to two years with Microsoft Research India. Candidates should have completed BS/BE/BTech or MS/ME/MTech in Computer Science or related areas, graduating by summer 2023. Apply before February 3, 2023.

The post Research Focus: Week of January 9, 2023 appeared first on Microsoft Research.

Read More

2022 has seen remarkable progress in foundational technologies that have helped to advance human knowledge and create new possibilities to address some of society’s most challenging problems. Significant advances in AI have also enabled Microsoft to bring new capabilities to customers through our products and services, including GitHub Copilot, an AI pair programmer capable of turning natural language prompts into code, and a preview of Microsoft Designer, a graphic design app that supports the creation of social media posts, invitations, posters, and one-of-a-kind images.

These offerings provide an early glimpse of how new AI capabilities, such as large language models, can enable people to interact with machines in increasingly powerful ways. They build on a significant, long-term commitment to fundamental research in computing and across the sciences, and the research community at Microsoft plays an integral role in advancing the state of the art in AI, while working closely with engineering teams and other partners to transform that progress into tangible benefits.

In 2022, Microsoft Research established AI4Science, a global organization applying the latest advances in AI and machine learning toward fundamentally transforming science; added to and expanded the capabilities of the company’s family of foundation models; worked to make these models and technologies more adaptable, collaborative, and efficient; further developed approaches to ensure that AI is used responsibly and in alignment with human needs; and pursued different approaches to AI, such as causal machine learning and reinforcement learning.

We shared our advances across AI and many other disciplines during our second annual Microsoft Research Summit, where members of our research community gathered virtually with their counterparts across industry and academia to discuss how emerging technologies are being explored and deployed to bring the greatest possible benefits to humanity.  

Plenary sessions at the event focused on the transformational impact of deep learning on the way we practice science, research that empowers medical practitioners and reduces inequities in healthcare, and emerging foundations for planet-scale computing. Further tracks and sessions over three days provided deeper dives into the future of the cloud; efficient large-scale AI; amplifying human productivity and creativity; delivering precision healthcare; building user trust through privacy, identity, and responsible AI; and enabling a resilient and sustainable world.

In this blog post, we look back at some of the key achievements and notable work in AI and highlight other advances across our diverse, multidisciplinary, and global organization.

Advancing AI foundations and accelerating progress

Over the past year, the research community at Microsoft made significant contributions to the rapidly evolving landscape of powerful large-scale AI models. Microsoft Research and the Microsoft Turing team unveiled a new Turing Universal Language Representation model capable of performing both English and multilingual understanding tasks. In computer vision, advancements for the Project Florence-VL (Florence-Vision and Language) team spanned still imagery and video: its GIT model was the first to surpass human performance on the image captioning benchmark TextCaps; LAVENDER showed strong performance in video question answering, text-to-video retrieval, and video captioning; and GLIP and GLIPv2 combined localization and vision-language understanding. The group also introduced NUWA-Infinity, a model capable of converting text, images, and video into high-resolution images or long-duration video. Meanwhile, the Visual Computing Group scaled up its Transformer-based general-purpose computer vision architecture, Swin Transformer, achieving applicability across more vision tasks than ever before.

Researchers from Microsoft Research Asia and the Microsoft Turing team also introduced BEiT-3, a general-purpose multimodal foundation model that achieves state-of-the-art transfer performance on both vision and vision-language tasks. In BEiT-3, researchers introduce Multiway Transformers for general-purpose modeling, where the modular architecture enables both deep fusion and modality-specific encoding. Based on the shared backbone, BEiT-3 performs masked “language” modeling on images (Imglish), texts (English), and image-text pairs (“parallel sentences”) in a unified manner. The code and pretrained models will be available at GitHub.

One of the most crucial accelerators of progress in AI is the ability to optimize training and inference for large-scale models. In 2022, the DeepSpeed team made a number of breakthroughs to improve mixture of experts (MoE) models, making them more efficient, faster, and less costly. Specifically, they were able to reduce training cost by 5x, reduce MoE parameter size by up to 3.7x, and reduce MoE inference latency by 7.3x while offering up to 4.5x faster and 9x cheaper inference for MoE models compared to quality-equivalent dense models.

Transforming scientific discovery and adding societal value

Our ability to comprehend and reason about the natural world has advanced over time, and the new AI4Science organization, announced in July, represents another turn in the evolution of scientific discovery. Machine learning is already being used in the natural sciences to model physical systems using observational data. AI4Science aims to dramatically accelerate our ability to model and predict natural phenomena by creating deep learning emulators that learn by using computational solutions to fundamental equations as training data.

This new paradigm can help scientists gain greater insight into natural phenomena, right down to their smallest components. Such molecular understanding and powerful computational tools can help accelerate the discovery of new materials to combat climate change, and new drugs to help support the prevention and treatment of disease.  

For instance, AI4Science’s Project Carbonix is working on globally accessible, at-scale solutions for decarbonizing the world economy, including reverse engineering materials that can pull carbon out of the environment and recycling carbon into materials. Collaborating on these efforts through the Microsoft Climate Research Initiative (MCRI) are domain experts from academia, industry, and government. Announced in June, MCRI is focused on areas such as carbon accounting, climate risk assessments, and decarbonization.

As part of the Generative Chemistry project, Microsoft researchers have been working with the global medicines company Novartis to develop and execute machine learning tools and human-in-the-loop approaches to enhance the entire drug discovery process. In April, they introduced MoLeR, a graph-based generative model for designing compounds that is more reflective of how chemists think about the process and is more efficient and practical than an earlier generative model the team developed. 

While AI4Science is focused on computational simulation, we have seen with projects like InnerEye that AI can have societal value in many other ways. In March, Microsoft acquired Nuance Communications Inc., further cementing the companies’ shared commitment to outcome-based AI across industries, particularly in healthcare. Tools like the integration of Microsoft Teams and Dragon Ambient eXperience (Nuance DAX) to help ease the administrative burden of physicians and support meaningful doctor-patient interactions are already making a difference.

Making AI more adaptable, collaborative, and efficient 

To help accelerate the capabilities of large-scale AI while building a landscape in which everyone can benefit from it, the research community at Microsoft aimed to drive progress in three areas: adaptability, collaboration, and efficiency.

To provide consistent value, AI systems must respond to changes in task and environment. Research in this area includes multi-task learning with task-aware routing of inputs, knowledge-infused decoding, model repurposing with data-centric ML, pruning and cognitive science or brain-inspired AI. A good example of our work toward adaptability is GODEL, or Grounded Open Dialogue Language Model, which ushers in a new class of pretrained language models that enable chatbots to help with tasks and then engage in more general conversations.  

Microsoft’s research into more collaborative AI includes AdaTest, which leverages human expertise alongside the generative power of large language models to help people more efficiently find and correct bugs in natural language processing models. Researchers have also explored expanding the use of AI in creative processes, including a project in which science fiction writer Gabrielle Loisel used OpenAI’s GPT-3 to co-author a novella and other stories. 

To enable more people to make use of AI in an efficient and sustainable way, Microsoft researchers are pursuing several new architectures and training paradigms. This includes new modular architectures and novel techniques, such as DeepSpeed Compression, a composable library for extreme compression and zero-cost quantization, and Z-Code Mixture of Experts models, which boost translation efficiency and were deployed in Microsoft Translator in 2022.  

In December, researchers unveiled AutoDistil, a new technique that leverages knowledge distillation and neural architecture search to improve the balance between cost and performance when generating compressed models. They also introduced AdaMix, which improves the fine-tuning of large pretrained models for downstream tasks using mixture of adaptations modules for parameter-efficient model tuning. And vision-language model compression research on the lottery ticket hypothesis showed that pretrained language models can be significantly compressed without hurting their performance.

Building and deploying AI responsibly

Building AI that maximizes its benefit to humanity, and does so equitably, requires considering both the opportunities and risks that come with each new advancement in line with our guiding principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability.

Helping to put these principles into practice is Microsoft’s Responsible AI Standard, which the company made publicly available in June. The standard comprises tools and steps that AI practitioners can execute in their workflows today to help ensure that building AI responsibly is baked into every stage of development. These standards will evolve as the tools and resources to responsibly build AI evolve in response to the rapid pace of AI advancement, particularly pertaining to the growing size of AI models and the new challenges they bring. 

With FedKD and InclusiveFL, researchers tackled some of the obstacles in applying federated learning, an ML method for protecting privacy, to model training. Two separate teams explored solutions for the harmful language that large generative models can reproduce—one presenting a unified framework for both detoxifying and debiasing models and another introducing methods for making content moderation tools more robust. Meanwhile, researchers sought to strengthen human-AI collaboration by giving users more insight into how models arrive at their outputs via explanations provided by the models themselves.

The responsible development of AI also means deploying technologies that operate the way they were designed to—and the way people expect them to. In a pair of blog posts, researchers draw on their respective experiences developing a technology to support social agency in children who are born blind and another to support mental health practitioners in guiding patient treatment to stress the need for multiple measures of performance in determining the readiness of increasingly complex AI systems and the incorporation of domain experts and user research throughout the development process.

Advancing AI for decision making

Building the next generation of AI requires continuous research into fundamental new AI innovations. Two significant areas of study in 2022 were causal ML and reinforcement learning.

Causal ML

Identifying causal effects is an integral part of scientific inquiry. It helps us understand everything from educational outcomes to the effects of social policies to risk factors for diseases. Questions of cause and effect are also critical for the design and data-driven evaluation of many technological systems we build today.  

This year, Microsoft Research continued its work on causal ML, which combines traditional machine learning with causal inference methods. To help data scientists better understand and deploy causal inference, Microsoft researchers built the DoWhy library, an end-to-end causal inference tool, in 2018. To broaden access to this critical knowledge base, DoWhy has now migrated to an independent open-source governance model in a new PyWhy GitHub organization. As part of this new collaborative model, Amazon Web Services is contributing new technology based on structural causal models.

At this year’s Conference on Neural Information Processing Systems (NeurIPS), researchers presented a suite of open-source causal tools and libraries that aims to simultaneously provide core causal AI functionality to practitioners and create a platform for research advances to be rapidly deployed. This includes ShowWhy, a no-code user interface suite that empowers domain experts to become decision scientists. We hope that our work accelerates use-inspired basic research for improvement of causal AI.

Reinforcement learning (RL)

Reinforcement learning is a powerful tool for learning which behaviors are likely to produce the best outcomes in a given scenario, typically through trial and error. But this powerful tool faces some challenges. Trial and error can consume enormous resources when applied to large datasets. And for many real-time applications, there’s no room to learn from mistakes.   

To address RL’s computational bottleneck, Microsoft researchers developed Path Predictive Elimination, a reinforcement learning method that is robust enough to remove noise from continuously changing environments. Also in 2022, a Microsoft team released MoCapAct, a library of pretrained simulated models to enable advanced research on artificial humanoid control at a fraction of the compute resources currently required.  

Researchers also developed a new method for using offline RL to augment human-designed strategies for making critical decisions. This team deployed game theory to design algorithms that can use existing data to learn policies that improve on current strategies.

Readers’ choice: Notable blog posts for 2022 

• Microsoft has demonstrated the underlying physics required to create a new kind of qubit
• µTransfer: A technique for hyperparameter tuning of enormous neural networks
• Powering the next generation of trustworthy AI in a confidential cloud using NVIDIA GPUs
• DeepSpeed Compression: A composable library for extreme compression and zero-cost quantization
• DoWhy evolves to independent PyWhy model to help causal inference grow
• PeopleLens: Using AI to support social interaction between children who are blind and their peers
• Microsoft Research Summit 2022: What’s Next for Technology and Humanity?
• Using reinforcement learning to identify high-risk states and treatments in healthcare
• GODEL: Combining goal-oriented dialog with real-world conversations
• Swin Transformer supports 3-billion-parameter vision models that can train with higher-resolution images for greater task applicability
• FLUTE: A scalable federated learning simulation platform
• (De)ToxiGen: Leveraging large language models to build more robust hate speech detection tools

Thank you for reading

2022 was an exciting year for research, and we look forward to the future breakthroughs our global research community will deliver. In the coming year, you can expect to hear more from us about our vision, and the impact we hope to achieve. We appreciate the opportunity to share our work with you, and we hope you will subscribe to the Microsoft Research Newsletter for the latest developments.

The post Research @ Microsoft 2022: A look back at a year of accelerating progress in AI appeared first on Microsoft Research.

Read More

Microsoft is home to a diverse team of researchers focused on supporting a healthy global society, including finding ways technology can address human rights problems affecting the most vulnerable populations around the world. With a multi-disciplinary background in human-computer interaction, data science, and the social sciences, the research team partners with community, governmental, and nongovernmental organizations to create open technologies that enable scalable responses to such challenges.  

The United Nations’ International Organization for Migration (IOM) provides direct assistance and support to migrants around the world, as well as victims and survivors of human trafficking. IOM is dedicated to promoting humane and orderly migration by providing services to governments and migrants in its 175 member countries. It recently reported 50 million victims of forced labor globally, including 3.3 million children, 6.3 million in commercial sexual exploitation, and 22 million trapped in forced marriages. Understanding and addressing problems at this scale requires technology to help anti-trafficking actors and domain experts gather and translate real-world data into evidence that can inform policies and build support systems. 

According to IOM, migrants and displaced people represent some of the most vulnerable populations in society. The organization explains that, “while human mobility can be a source of prosperity, innovation, and sustainable development, every migration journey can include risks to safety, which are exacerbated during times of crisis, or when people face extreme vulnerability as they are forced to migrate amid a lack of safe and regular migration pathways.

Spotlight: On-Demand EVENT

Microsoft Research Summit 2022

On-Demand
Watch now to learn about some of the most pressing questions facing our research community and listen in on conversations with 120+ researchers around how to ensure new technologies have the broadest possible benefit for humanity.

Explore sessions

Today, using software developed by Microsoft researchers, IOM released its second synthetic dataset from trafficking victim case records, the first ever public dataset to describe victim-perpetrator relations. The synthetic dataset is also the first of its kind to be generated with differential privacy, providing an additional security guarantee for multiple data releases, which enables the sharing of more data and allows more rigorous research to be conducted while protecting privacy and civil liberties. 

The new data release builds on several years of collaboration between Microsoft and IOM to support safe data sharing of victim case records in ways that can inform collective action across the anti-trafficking community. This collaboration began in July 2019 when IOM joined the accelerator program of the Tech Against Trafficking (TAT) coalition, with the goal of advancing the privacy and utility of data made available through the Counter Trafficking Data Collaborative (CTDC) data hub – the first global portal on human trafficking case data. Since then, IOM and Microsoft have collaborated to improve the ways data on identified victims and survivors—as well as their accounts of perpetrators—can be used to combat the proliferation of human trafficking.  

“We are grateful to Microsoft Research for our partnership over almost four years to share data while protecting the safety and privacy of victims and survivors of trafficking.”

– Monica Goracci, IOM’s Director of Programme Support and Migration Management

The critical importance of data privacy when working with vulnerable populations 

When publishing data on victims of trafficking, all efforts must be taken to ensure that traffickers are wholly prevented from identifying known victims in published datasets. It is also important to protect individuals’ privacy to avoid stigma or other potential forms of harm or (re)traumatization. Data statistics accuracy is another concern: the statistics must simultaneously enable researchers and analysts to guarantee victims’ privacy and extract useful insights from the dataset containing personal information. This is critically important: if a privacy method were to over- or under-report a given pattern in victim cases, it could mislead decision makers to misdirect scarce resources and therefore fail to tackle the originating problem.

The collaboration between IOM and Microsoft was founded on the idea that rather than redacting sensitive data to create privacy, synthetic datasets can be generated in ways that accurately capture the structure and statistics of underlying sensitive datasets, while remaining private by design. But not all synthetic data comes with formal guarantees of data privacy or accuracy. Therefore, building trust in synthetic data requires communicating how well the synthetic data represents the actual sensitive data, while ensuring that these comparisons do not create privacy risks themselves.

From this founding principle, along with the need to accurately report case counts broken down by different combinations of attributes (e.g., age range, gender, nationality), a solution emerged: to release synthetic data alongside privacy-preserving counts of cases, matching all short combinations of case attributes. The aggregate data thereby supports both evaluation of synthetic data quality and retrieval of accurate counts for official reporting. Through this collaboration and the complementary nature of synthetic data and aggregate data—together with interactive interfaces with which to view and explore both datasets—the open-source Synthetic Data Showcase software was developed.

In September 2021, IOM used Synthetic Data Showcase to release its first downloadable Global Synthetic Dataset, representing data from over 156,000 victims and survivors of trafficking across 189 countries and territories (where victims were first identified and supported by CTDC partners). The new Global Victim-Perpetrator Synthetic Dataset, released today, is CTDC’s second synthetic dataset produced using an updated version of Synthetic Data Showcase with added support for differential privacy. This new dataset includes IOM data from over 17,000 trafficking victim case records and their accounts of over 37,000 perpetrators who facilitated the trafficking process from 2005 to 2022.  Together, these datasets provide vital first-hand information on the socio-demographic profiles of victims, their accounts of perpetrators, types of exploitation, and the overall trafficking process—all of which are critical to better assist survivors and prosecute perpetrators. 

“Data privacy is crucial to the pursuit of efficient, targeted counter-trafficking policies and good migration governance.”

– Irina Todorova, Head of the Assistance to Vulnerable Migrants Unit at IOM’s Protection Division

A differentially private dataset 

In 2006, Microsoft researchers led the initial development of differential privacy, and today it represents the gold standard in privacy protection. It helps ensure that answers to data queries are similar, whether or not any individual data subject is in the dataset, and therefore cannot be used to infer the presence of specific individuals, either directly or indirectly.  

Existing algorithms for differentially private data synthesis typically create privacy by “hiding” actual combinations of attributes in a sea of fabricated or spurious attribute combinations that don’t specifically reflect what was in the original sensitive dataset.

This can be problematic if the presence of these fabricated attribute combinations misrepresents the real-world situation and misleads downstream decision making, policy making, or resource allocation to the detriment of the underlying population (e.g., encouraging policing of trafficking routes that have not actually been observed). 

When the research team encountered these challenges with existing differentially private synthesizers, they engaged fellow researchers at Microsoft to explore possible solutions. They explained the critical importance of reporting accurate counts of actual attribute combinations in support of statistical reporting and evidence-based intervention, and how the “feature” of fabricating unobserved combinations as a way of preserving privacy could be harmful when attempting to understand real-world patterns of exploitation.

Those colleagues had recently solved a similar problem in a different context: how to extract accurate counts of n-gram word combinations from a corpus of private text data. Their solution, recently published at the 2021 Conference on Neural Information Processing Systems, significantly outperformed the state of the art. In collaboration with the research team working with IOM, they adapted this solution into a new approach to generating differentially private marginals—counts of all short combinations of attributes that represented a differentially-private aggregate dataset.

Because differentially private data has the property that subsequent processing cannot increase privacy loss, any datasets generated from such aggregates retain the same level of privacy. This enabled the team to modify their existing approach to data synthesis—creating synthetic records by sampling attribute combinations until all attributes are accounted for—to extrapolate these noisily reported attribute combinations into full, differentially-private synthetic records. The result is precisely what IOM and similar organizations need to create a thriving data ecosystem in the fight against human trafficking and other human rights violations: accurate aggregate data for official reporting, synthetic data for interactive exploration and machine learning, and differential privacy guarantees that provide protection even over multiple overlapping data releases. 

This new synthesizer is now available to the community via Microsoft’s SmartNoise library within the OpenDP initiative. Unlike existing synthesizers, it provides strong control over the extent to which fabrication of spurious attribute combinations is allowed and augments synthetic datasets with “actual” aggregate data protected by differential privacy.

Access to private-yet-accurate patterns of attributes characterizing victim-perpetrator relationships allows stakeholders to advance the understanding of risk factors for vulnerability and carry out effective counter-trafficking interventions, all while keeping the victims’ identities private.

“The new dataset represents the first global collection of case data linking the profiles of trafficking victims and perpetrators ever made available to the public, while enabling strong privacy guarantees. It provides critical information to better assist survivors and prosecute offenders.” – Claire Galez-Davis, Data Scientist at IOM’s Protection Division. 

An intuitive new interface and public utility web application 

Solving problems at a global scale requires tools that make safe data sharing accessible wherever there is a need and in a way that is understandable by all stakeholders. The team wanted to construct an intuitive interface to help develop a shared evidence base and motivate collective action by the anti-trafficking community. They also wanted to ensure that the solution was available to anyone with a need to share sensitive data safely and responsibly. The new user interface developed through this work is now available as a public utility web application in which private data aggregation and synthesis are performed locally in the web browser, with no data ever leaving the user’s machine.

“I find the locally run web application incredibly interactive and intuitive. It is a lot easier for me to explain the data generation process and teach others to use the new web interface. As the data is processed locally in our computers, I don’t need to worry about data leaks.” – Lorraine Wong, Research Officer at IOM’s Protection Division.  

What’s next for the IOM and Microsoft collaboration 

Microsoft and IOM have made the solution publicly accessible for other organizations, including central government agencies. It can be used by any stakeholder who wants to collect and publish sensitive data while protecting individual privacy.

Through workshops and guidance on how to produce high-quality administrative data, the organizations plan to share evidence on exploitation and abuse to support Member States, other UN agencies, and counter-trafficking organizations around the world. This kind of administrative data is a key source of information providing baseline statistics that can be used to understand patterns, risk factors, trends, and modus operandi that are critical for policy response formulation.

For example, IOM has been collaborating with the UN Office on Drugs and Crime (UNODC) to establish international standards and guidance to support governments in producing high-quality administrative data. It has also been collaborating with the UN International Labour Organization (ILO) to index policy-oriented research on trafficking in a bibliography. Finally, IOM is producing an online course, including a module that includes guidance on synthetic data, to encourage safe data sharing from governments and frontline counter-trafficking agencies.

“Being able to publish more data than we have done in the past, and in an even safer way, is a great achievement,” explained Phineas Jasi, Data Management and Research Specialist at IOM’s Protection Division. He added that “The aim is for these data to inform the evidence base on human trafficking, which in turns helps devise efficient and targeted counter-trafficking policies and achieve good migration governance.” 

Translating data into evidence is the goal of the related ShowWhy application from the same Microsoft research team, which guides domain experts through the end-to-end process of developing causal evidence from observational data. Just like Synthetic Data Showcase, it makes advanced data science capabilities accessible to domain experts through a suite of interactive, no-code user interfaces. 

“Driving a coordinated global response against human trafficking requires removing traditional barriers to both data access and data analysis,” said Darren Edge, Director at Microsoft Research. “With our Synthetic Data Showcase and ShowWhy applications, we are aiming to empower domain experts to develop causal evidence for themselves, from sensitive data that couldn’t otherwise be shared, and use this to inform collective action with a precision and scale that couldn’t otherwise be imagined.” 

The post IOM and Microsoft release first-ever differentially private synthetic dataset to counter human trafficking appeared first on Microsoft Research.

Read More

In this issue, we continue to feature some of our 100+ papers accepted at NeurIPS 2022.

Outstanding paper: Gradient Estimation with Discrete Stein Operators

Jiaxin Shi, Yuhao Zhou, Jessica Hwang, Michalis Titsias, Lester Mackey

Gradient estimation — approximating the gradient of an expectation with respect to the parameters of a distribution — is central to the solution of many machine learning problems. However, when the distribution is discrete, most common gradient estimators suffer from excessive variance. To improve the quality of gradient estimation, we introduce a variance reduction technique based on Stein operators for discrete distributions in our paper: Gradient Estimation with Discrete Stein Operators. We then use this technique to build flexible control variates for the REINFORCE leave-one-out estimator. Our control variates can be adapted online to minimize variance and do not require extra evaluations of the target function. In benchmark generative modeling tasks such as training binary variational autoencoders, our gradient estimator achieves substantially lower variance than state-of-the-art estimators with the same number of function evaluations.

Learning Modular Simulations for Homogeneous Systems

Jayesh K. Gupta, Sai Vemprala, Ashish Kapoor

Data-driven simulations have emerged as an effective alternative to building simulators from scratch, which involves complex handcrafting and expert knowledge. However, complex real-life systems are often decomposable into modular subsystems and are usually designed in such a manner for engineering tractability. Unfortunately, current data-driven methods tend to learn monolithic blocks in an end-to-end fashion, which often results in simulators that only work for a particular configuration and are not generalizable. In our paper: Learning Modular Simulations for Homogeneous Systems, we present a modular simulation framework for modeling homogeneous multibody systems, which combines ideas from graph neural networks and neural differential equations. We propose the Message-Passing Neural Ordinary Differential Equation (MP-NODE), a paradigm for modeling individual subsystems as neural ODEs along with spatio-temporal message-passing capability, which is then used to orchestrate full simulations. We evaluate our framework on a variety of systems and show that message passing allows coordination between multiple modules over time for accurate predictions and can also enable zero-shot generalization to new system configurations. Furthermore, we show that our models can be transferred to new system configurations with lower data requirement and training effort, compared to those trained from scratch.

Self-explaining deep models with logic rule reasoning 

Seungeon Lee, Xiting Wang, Sungwon Han, Xiaoyuan Yi, Xing Xie, Meeyoung Cha

Deep learning has shown high predictive accuracy in a wide range of tasks, but its inner working mechanisms are obscured by complex model designs. This raises important questions about whether a deep model is ethical, trustworthy, or capable of performing as intended under various conditions.

In our paper, Self-explaining deep models with logic rule reasoning, we present SELOR, a framework for integrating self-explaining capabilities into a given deep model to achieve both high prediction performance and human precision. By “human precision”, we refer to the degree to which humans agree with the reasons models provide for their predictions. Human precision affects user trust and allows users to collaborate closely with the model. We demonstrate that logic rule explanations naturally satisfy human precision with the expressive power required for good predictive performance. We then show how to enable a deep model to predict and explain with logic rules. Our method does not require predefined logic rule sets or human annotations and can be learned efficiently and easily with widely-used deep learning modules in a differentiable way. Extensive experiments show that our method gives explanations closer to human decision logic than other methods while maintaining the performance of deep learning models.

3DB: A Framework for Debugging Computer Vision Models 

Guillaume Leclerc, Hadi Salman, Andrew Ilyas, Sai Vemprala, Logan Engstrom, Vibhav Vineet, Kai Xiao, Pengchuan Zhang, Shibani Santurkar, Greg Yang, Ashish Kapoor, Aleksander Madry 

Computer vision models such as classification and object detection are known to fail in many different ways. While the importance of recognizing and addressing such shortcomings is well understood, we lack a scalable and efficient means of identifying such failure cases. To address this issue, we introduce 3DB: an extendable, unified framework for testing and debugging vision models using photorealistic simulation. In our paper, 3DB: A Framework for Debugging Computer Vision Models, we show how 3DB enables users to test vision models on synthetic images, where a diverse set of factors can be controlled. We demonstrate, through a wide range of use cases, that 3DB allows users to discover shortcomings in computer vision systems and gain insights into how models make decisions. 3DB captures and generalizes many robustness analyses from prior work and enables one to study their interplay. Finally, we find that the insights generated by the system transfer to the physical world. We are releasing 3DB as a library alongside a set of example analyses, guides, and documentation: https://3db.github.io/3db/.

The post Research Focus: Week of December 5, 2022 appeared first on Microsoft Research.

Read More

Microsoft is proud to be a platinum sponsor of the 36th annual conference on Neural Information Processing Systems (NeurIPS), which is widely regarded as the world’s most prestigious research conference on artificial intelligence and machine learning.

Microsoft has a strong presence at NeurIPS again this year, with more than 150 of our researchers participating in the conference and 122 of our research papers accepted. Our researchers are also taking part in 10 workshops, four competitions and a tutorial.

In one of the workshops, AI for Science: Progress and Promises, a panel of leading researchers will discuss how artificial intelligence and machine learning have the potential to advance scientific discovery. The panel will include two Microsoft researchers: Max Welling, Vice President and Distinguished Scientist, Microsoft Research AI4Science, who will serve as moderator, and Peter Lee, Corporate Vice President, Microsoft Research and Incubations.

Of the 122 Microsoft research papers accepted for the conference, seven have been selected for oral presentations during the virtual NeurIPS experience the week of December 4^th. The oral presentations provide a deeper dive into each of the featured research topics.

In addition, two other Microsoft research papers received Outstanding Paper Awards for NeurIPS 2022. One of those papers, Gradient Estimation with Discrete Stein Operators, explains how researchers developed a gradient estimator that achieves substantially lower variance than state-of-the-art estimators with the same number of function evaluations, which has the potential to improve problem solving in machine learning. In the other paper, A Neural Corpus Indexer for Document Retrieval, researchers demonstrate that an end-to-end deep neural network that unifies training and indexing stages can significantly improve the recall performance of traditional document retrieval methods.

Spotlight: On-Demand EVENT

Microsoft Research Summit 2022

On-Demand
Watch now to learn about some of the most pressing questions facing our research community and listen in on conversations with 120+ researchers around how to ensure new technologies have the broadest possible benefit for humanity.

Explore sessions

Below we have provided the titles, authors and abstracts for all seven of the Microsoft research papers chosen for oral presentations at NeurIPS, with links to additional information for those who want to explore the topics more fully:

Uni[MASK]: Unified Inference in Sequential Decision Problems

Micah Carroll, Orr Paradise, Jessy Lin, Raluca Georgescu, Mingfei Sun, David Bignell, Stephanie Milani, Katja Hofmann, Matthew Hausknecht, Anca Dragan, Sam Devlin

Abstract: Randomly masking and predicting word tokens has been a successful approach in pre-training language models for a variety of downstream tasks. In this work, we observe that the same idea also applies naturally to sequential decision making, where many well-studied tasks like behavior cloning, offline RL, inverse dynamics, and waypoint conditioning correspond to different sequence maskings over a sequence of states, actions, and returns. We introduce the UniMASK framework, which provides a unified way to specify models which can be trained on many different sequential decision-making tasks. We show that a single UniMASK model is often capable of carrying out many tasks with performance similar to or better than single-task models. Additionally, after fine tuning, our UniMASK models consistently outperform comparable single-task models.

K-LITE: Learning Transferable Visual Models with External Knowledge

Sheng Shen, Chunyuan Li, Xiaowei Hu, Yujia Xie, Jianwei Yang, Pengchuan Zhang, Zhe Gan, Lijuan Wang, Lu Yuan, Ce Liu, Kurt Keutzer, Trevor Darrell, Anna Rohrbach, Jianfeng Gao

Abstract: The new generation of state-of-the-art computer vision systems are trained from natural language supervision, ranging from simple object category names to descriptive captions. This form of supervision ensures high generality and usability of the learned visual models, based on the broad concept coverage achieved through large-scale data collection process. Alternatively, we argue that learning with external knowledge about images is a promising way which leverages a much more structured source of supervision and offers sample efficiency.

In this paper, we propose K-LITE (Knowledge-augmented Language-Image Training and Evaluation), a simple strategy to leverage external knowledge for building transferable visual systems: In training, it enriches entities in natural language with WordNet and Wiktionary knowledge, leading to an efficient and scalable approach to learning image representations that uses knowledge about the visual concepts; In evaluation, the natural language is also augmented with external knowledge and then used to reference learned visual concepts (or describe new ones) to enable zero-shot and few-shot transfer of the pre-trained models. We study the performance of K-LITE on two important computer vision problems, image classification and object detection, benchmarking on 20 and 13 different existing datasets, respectively. The proposed knowledge-augmented models show significant improvement in transfer learning performance over existing methods. Our code is released at https://github.com/microsoft/klite.

Extreme Compression for Pre-trained Transformers Made Simple and Efficient

Xiaoxia Wu, Zhewei Yao, Minjia Zhang, Conglong Li, Yuxiong He

Abstract: Extreme compression, particularly ultra-low bit precision (binary/ternary) quantization, has been proposed to fit large NLP models on resource-constraint devices. However, to preserve the accuracy for such aggressive compression schemes, cutting-edge methods usually introduce complicated compression pipelines, e.g., multi-stage expensive knowledge distillation with extensive hyperparameter tuning. Also, they oftentimes focus less on smaller transformer models that have already been heavily compressed via knowledge distillation and lack a systematic study to show the effectiveness of their methods.

In this paper, we perform a very comprehensive systematic study to measure the impact of many key hyperparameters and training strategies from previous. As a result, we find out that previous baselines for ultra-low bit precision quantization are significantly under-trained. Based on our study, we propose a simple yet effective compression pipeline for extreme compression.

Our simplified pipeline demonstrates that:

(1) we can skip the pre-training knowledge distillation to obtain a 5-layer bert while achieving better performance than previous state-of-the-art methods, like TinyBERT;

(2) extreme quantization plus layer reduction is able to reduce the model size by 50x, resulting in new state-of-the-art results on GLUE tasks.

On the Complexity of Adversarial Decision Making

Dylan J Foster, Alexander Rakhlin, Ayush Sekhari, Karthik Sridharan

Abstract: A central problem in online learning and decision making—from bandits to reinforcement learning—is to understand what modeling assumptions lead to sample-efficient learning guarantees. We consider a general adversarial decision-making framework that encompasses (structured) bandit problems with adversarial rewards and reinforcement learning problems with adversarial dynamics. Our main result is to show—via new upper and lower bounds—that the Decision-Estimation Coefficient, a complexity measure introduced by Foster et al. in the stochastic counterpart to our setting, is necessary and sufficient to obtain low regret for adversarial decision making. However, compared to the stochastic setting, one must apply the Decision-Estimation Coefficient to the convex hull of the class of models (or, hypotheses) under consideration. This establishes that the price of accommodating adversarial rewards or dynamics is governed by the behavior of the model class under convexification, and recovers a number of existing results –both positive and negative. En route to obtaining these guarantees, we provide new structural results that connect the Decision-Estimation Coefficient to variants of other well-known complexity measures, including the Information Ratio of Russo and Van Roy and the Exploration-by-Optimization objective of Lattimore and György.

Maximum Class Separation as Inductive Bias in One Matrix

Tejaswi Kasarla, Gertjan J. Burghouts, Max van Spengler, Elise van der Pol, Rita Cucchiara, Pascal Mettes

Abstract: Maximizing the separation between classes constitutes a well-known inductive bias in machine learning and a pillar of many traditional algorithms. By default, deep networks are not equipped with this inductive bias and therefore many alternative solutions have been proposed through differential optimization. Current approaches tend to optimize classification and separation jointly: aligning inputs with class vectors and separating class vectors angularly.

This paper proposes a simple alternative: encoding maximum separation as an inductive bias in the network by adding one fixed matrix multiplication before computing the softmax activations. The main observation behind our approach is that separation does not require optimization but can be solved in closed-form prior to training and plugged into a network. We outline a recursive approach to obtain the matrix consisting of maximally separable vectors for any number of classes, which can be added with negligible engineering effort and computational overhead. Despite its simple nature, this one matrix multiplication provides real impact. We show that our proposal directly boosts classification, long-tailed recognition, out-of-distribution detection, and open-set recognition, from CIFAR to ImageNet. We find empirically that maximum separation works best as a fixed bias; making the matrix learnable adds nothing to the performance. The closed-form implementation and code to reproduce the experiments are available on GitHub.

Censored Quantile Regression Neural Networks for Distribution-Free Survival Analysis

Tim Pearce, Jong-Hyeon Jeong, Yichen Jia, Jun Zhu

Abstract: This paper considers doing quantile regression on censored data using neural networks (NNs). This adds to the survival analysis toolkit by allowing direct prediction of the target variable, along with a distribution-free characterization of uncertainty, using a flexible function approximator. We begin by showing how an algorithm popular in linear models can be applied to NNs. However, the resulting procedure is inefficient, requiring sequential optimization of an individual NN at each desired quantile. Our major contribution is a novel algorithm that simultaneously optimizes a grid of quantiles output by a single NN. To offer theoretical insight into our algorithm, we show firstly that it can be interpreted as a form of expectation-maximization, and secondly that it exhibits a desirable `self-correcting’ property. Experimentally, the algorithm produces quantiles that are better calibrated than existing methods on 10 out of 12 real datasets.

Learning (Very) Simple Generative Models Is Hard

Sitan Chen, Jerry Li, Yuanzhi Li

Abstract: Motivated by the recent empirical successes of deep generative models, we study the computational complexity of the following unsupervised learning problem. For an unknown neural network (F:mathbb{R}^dtomathbb{R}^{d’}), let (D) be the distribution over (mathbb{R}^{d’}) given by pushing the standard Gaussian (mathcal{N}(0,textrm{Id}_d)) through (F). Given i.i.d. samples from (D), the goal is to output ({any}) distribution close to (D) in statistical distance.

We show under the statistical query (SQ) model that no polynomial-time algorithm can solve this problem even when the output coordinates of (F) are one-hidden-layer ReLU networks with (log(d)) neurons. Previously, the best lower bounds for this problem simply followed from lower bounds for (supervised) (learning) and required at least two hidden layers and (poly(d)) neurons [Daniely-Vardi ’21, Chen-Gollakota-Klivans-Meka ’22].

The key ingredient in our proof is an ODE-based construction of a compactly supported, piecewise-linear function (f) with polynomially-bounded slopes such that the pushforward of (mathcal{N}(0,1)) under (f) matches all low-degree moments of (mathcal{N}(0,1)).

The post NeurIPS 2022: Seven Microsoft Research Papers Selected for Oral Presentations appeared first on Microsoft Research.

Read More

Few-shot Task-agnostic Neural Architecture Search for Distilling Large Language Models

Dongkuan Xu, Subhabrata Mukherjee, Xiaodong Liu, Debadeepta Dey, Wenhui Wang, Xiang Zhang, Ahmed Hassan Awadallah, Jianfeng Gao

Knowledge distillation (KD) is effective in compressing large pre-trained language models, where we train a small student model to mimic the output distribution of a large teacher model (e.g., BERT, GPT-X). KD relies on hand-designed student model architectures that require several trials and pre-specified compression rates. In our paper, Few-shot Task-agnostic Neural Architecture Search for Distilling Large Language Models, we discuss AutoDistil, a new technique pioneered by Microsoft Research that leverages advances in and neural architecture search (NAS) to automatically generate a suite of compressed models with variable computational cost (e.g., varying sizes, FLOPs and latency). NAS for distillation addresses customization challenges of hand-engineering compressed model architectures for diverse deployment environments having variable resource constraints with an automated framework. AutoDistil-generated compressed models obtain up to 41x reduction in FLOPs with limited regression in task performance and 6x FLOPs reduction with parity in performance with large teacher model. Given any state-of-the-art compressed model, AutoDistil finds a better compressed variant with better trade-off in task performance vs. computational cost during inference.

Neuron with steady response leads to better generalization

Qiang Fu, Lun Du, Haitao Mao, Xu Chen, Wei Fang, Shi Han, Dongmei Zhang

Improving models’ ability to generalize is one of the most important research problems in machine learning. Deep neural networks with diverse architectures have been invented and widely applied to various domains and tasks. Our goal was to study and identify the fundamental properties commonly shared by different kinds of deep neural networks, and then design a generic technique applicable for all of them to improve their generalization.

In this paper, from the neural level granularity, we study the characteristics of individual neurons’ response during the training dynamics. We find that keeping the response of activated neurons stable for the same class helps improve models’ ability to generalize. This is a new regularization perspective based on the neuron-level class-dependent response distribution. Meanwhile, we observed that the traditional vanilla model usually lacks good steadiness of intra-class response. Based on these observations, we designed a generic regularization method, Neuron Steadiness Regularization (NSR), to reduce large intra-class neuron response variance. NSR is computationally efficient and applicable to various architectures and tasks. Significant improvements are obtained on extensive experiments with multiple types of datasets and various network architectures. We will continue the research for improving the model generalization ability.

Long-form video-language pre-training with multimodal temporal contrastive learning

Yuchong Sun, Hongwei Xue, Ruihua Song, Bei Liu, Huan Yang, Jianlong Fu

Huge numbers of videos on diverse topics and of various lengths are shared on social media. Analyzing and understanding these videos is an important but challenging problem. Previous work on action and scene recognition has been limited to certain labels, while neglecting the rich semantic and dynamic information in other videos. Inspired by the cross-modal pre-training paradigm in image-language domain (e.g., CLIP, Florence), researchers have explored video-language joint pre-training, which mainly use short-form videos (e.g.,

In this research, we propose a Long-Form VIdeo-LAnguage pre-training model (LF-VILA) to explore long-form video representation learning, and train it on a long-form video-language dataset (LF-VILA-8M) on the basis of our new collected video-language dataset (HD-VILA-100M). We then design a Multimodal Temporal Contrastive (MTC) loss to capture the temporal relation between video clips and single sentences. We also propose the Hierarchical Temporal Window Attention (HTWA) mechanism on video encoder to reduce the training time by one-third. Our model achieves significant improvements on nine benchmarks, including paragraph-to-video retrieval, long-form video question-answering, and action recognition tasks. In the future, we will explore using it for broader scenarios, such as ego-centric video understanding.

Microsoft Research Causality and ML team features multiple papers and workshops at NeurIPS 2022

Parikshit Bansal, Ranveer Chandra, Eleanor Dillon, Saloni Dash, Rui Ding, Darren Edge, Adam Foster, Wenbo Gong, Shi Han, Agrin Hilmkil, Joel Jennings, Jian Jiao, Emre Kıcıman, Hua Li, Chao Ma, Sara Malvar, Robert Ness, Nick Pawlowski, Yashoteja Prabhu, Eduardo Rodrigues, Amit Sharma, Swati Sharma, Cheng Zhang, Dongmei Zhang

Identifying causal effects is an integral part of scientific inquiry, helping us to understand everything from educational outcomes to the effects of social policies to risk factors for diseases. Questions of cause-and-effect are also critical for the design and data-driven improvement and evaluation of business and technological systems we build today. The intersection of causal analysis and machine learning is driving rapid advances. Microsoft researchers are excited to be presenting three papers at NeurIPS, along with workshops on new methods and their applications. This includes work improving deep methods for causal discovery, applying causal insights to improve responsible language models, and improving soil carbon modeling with causal approaches. To accelerate research and broaden adoption of the latest causal methods, Microsoft researchers are co-organizing the Workshop on Causality for Real-world Impact and releasing new no-code interactive ShowWhy tools for causal discovery and analysis. We encourage NeurIPS attendees to learn more via the links below or stop by the Microsoft booth for demos and talks.

Main conference papers

Workshop papers

Workshop on Causality for Real-world Impact

• Making A Causal AI Suite for Decision-Making
• The Counterfactual-Shapley Value: Attributing Change in System Metrics
• Counterfactual Generation Under Confounding
• Deep End-to-end Causal Inference
• Rhino: Deep Causal Temporal Relationship Learning with history-dependent noise
• Causal Reasoning in the Presence of Latent Confounders via Neural ADMG Learning

Workshop on Tackling Climate Change with Machine Learning

• Causal Modeling of Soil Processes for Improved Generalization

Workshop on Distribution Shifts

• Using Interventions to Improve Out-of-Distribution Generalization of Text-Matching Recommendation Systems

Workshop on Understanding Deep Learning Through Empirical Falsification (“I can’t believe it’s not better”)
We’ll be participating in the panel.

New research on generative models

Two papers covering new research on generative models will be presented at NeurIPS 2022.

Vikas Raunak, Matt Post, Arul Menezes

The first paper, Operationalizing Specifications, In Addition to Test Sets for Evaluating Constrained Generative Models, presents recommendations on the evaluation of state-of-the-art generative models for constrained generation tasks. The progress on generative models has been rapid in recent years. These large-scale models have had three impacts: 1) The fluency of generation in both language and vision modalities has rendered common average-case evaluation metrics much less useful in diagnosing system errors; 2) The same substrate models now form the basis of a number of applications, driven both by the utility of their representations as well as phenomena such as in-context learning, which raise the abstraction level of interacting with such models; 3) The user expectations around these models have made the technical challenge of out-of-domain generalization much less excusable in practice. Subsequently, our evaluation methodologies haven’t adapted to these changes. More concretely, while the associated utility and methods of interacting with generative models have expanded, a similar expansion has not been observed in their evaluation practices. In this paper, we argue that the scale of generative models could be exploited to raise the abstraction level at which evaluation itself is conducted and provide recommendations for the same. Our recommendations are based on leveraging specifications as a powerful instrument to evaluate generation quality and are readily applicable to a variety of tasks. 

• Venue: Workshop on Human Evaluation of Generative Models, 36th Conference on Neural Information Processing Systems

Vikas Raunak, Arul Menezes

The second paper is Rank-One Editing of Encoder-Decoder Models. Here, we look at large sequence-to-sequence models for tasks such as neural machine translation (NMT), which are usually trained over hundreds of millions of samples. However, training is just the origin of a model’s life-cycle. Real-world deployments of models require further behavioral adaptations as new requirements emerge or shortcomings become known. Typically, in the space of model behaviors, behavior deletion requests are addressed through model retrainings, whereas model finetuning is done to address behavior addition requests. Both procedures are instances of data-based model intervention. In this work, we present a preliminary study investigating rank-one editing as a direct intervention method for behavior deletion requests in encoder-decoder transformer models. We propose four editing tasks for NMT and show that the proposed editing algorithm achieves high efficacy, while requiring only a single instance of positive example to fix an erroneous (negative) model behavior. This research therefore explores a path towards fixing the deleterious behaviors of encoder-decoder models for tasks such as translation, making them safer and more reliable without investing in a huge computational budget. 

• Venue: The Second Workshop On Interactive Learning For Natural Language Processing (InterNLP 2022)

Award Winner: A Neural Corpus Indexer for Document Retrieval

Yujing Wang, Yingyan Hou, Haonan Wang, Ziming Miao, Shibin Wu, Hao Sun, Qi Chen, Yuqing Xia, Chengmin Chi, Guoshuai Zhao, Zheng Liu, Xing Xie, Hao Allen Sun, Weiwei Deng, Qi Zhang, Mao Yang

Note: this paper was named an Outstanding Paper at NeurIPS 2022

Current state-of-the-art document retrieval solutions typically follow an index-retrieve paradigm, where the index is not directly optimized towards the final target. The proposed Neural Corpus Indexer (NCI) model, instead, leverages a sequence-to-sequence architecture, which serves as a model-based index that takes a query as input and outputs the most relevant document identifiers. For the first time, we demonstrate that an end-to-end differentiable document retrieval model can significantly outperform both sparse inverted index and dense retrieval methods. Specifically, NCI achieves +17.6% and +16.8% relative enhancement for Recall@1 on NQ320k dataset and R-Precision on TriviaQA dataset respectively, and a competitive MRR without using an explicit re-ranking model. This work has received a NeurIPS 2022 Outstanding Paper award.

The pipeline is composed of three stages. In the first stage, documents are encoded into semantic identifiers by the hierarchical k-means algorithm. In the second stage, a query generation model is employed to prepare training pairs. At the third stage, the NCI is trained with cross-entropy and consistency-based regularization losses. To further align with the hierarchical nature of the semantic identifiers, a weight adaptation mechanism is introduced to make the decoder aware of semantic prefixes. During inference, top N relevant documents can be easily obtained via beam search. The proposed approach introduces architectural and training choices that demonstrate the promising future of neural indexers as a viable alternative. And the discussed open questions can serve as an inspiration for future research.

Microsoft Research career opportunities – come join us!

We’re hiring for multiple roles including internships and researchers at all levels in multiple Microsoft Research labs. Join us and work on causal ML, precision health, genomics, deep learning, robotics, or computational chemistry. If you’re attending the conference, stop by the Microsoft booth (Expo Hall G, Booth #202) to speak with researchers and recruiters about working at Microsoft and open job opportunities. Or you can browse our current openings at NeurIPS 2022 – Microsoft Research career opportunities.

The post Research Focus: Week of November 28, 2022 appeared first on Microsoft Research.

Read More

Trust is essential for people and organizations to use technology with confidence. At Microsoft, we strive to earn the trust of our customers, employees, communities, and partners by committing to privacy, security, the responsible use of AI, and transparency.

At Microsoft Research, we take on this challenge by creating and using state-of-the-art tools and technologies that support a proactive, integrated approach to security across all layers of the digital estate.

Threats to cybersecurity are constant and they continue to grow, impacting organizations and individuals everywhere. Attack tools are readily available and well-funded adversaries now have the capability to cause unprecedented harm. These threats help explain why U.S. President Joe Biden issued an executive order in 2021 calling for cybersecurity improvements. Similarly, the European Union recently called for stronger protection of its information and communication technology (ICT) supply chains.

Against that backdrop, Microsoft Research is focused on what comes next in security and privacy. New and emerging computing frontiers, like the metaverse and web3, will require consistent advances in identity, transparency and other security principles, in order to learn from the past and unlock these technologies’ potential. Developments in quantum computing and advances in machine learning and artificial intelligence offer great potential to advance science and the human condition. Our research aims to ensure that future breakthroughs come with robust safety and privacy protections, even as they accelerate profound changes and new business opportunities.

At Microsoft Research, we pursue ambitious projects to improve the privacy and security of everyone on the planet. This is the first blog post in a series exploring the work we do in privacy, security and cryptography. In future installments, we will dive deeper into the research challenges we are addressing, and the opportunities we see.

Spotlight: On-Demand EVENT

Microsoft Research Summit 2022

On-Demand
Watch now to learn about some of the most pressing questions facing our research community and listen in on conversations with 120+ researchers around how to ensure new technologies have the broadest possible benefit for humanity.

Explore sessions

Digital identities

While the internet was not originally built with an identity layer, digital identities have grown to become foundational elements of today’s web and impacti people’s lives even beyond the digital world. Our research is aimed at modernizing digital identities and building more robust, usable, private and secure user-centric identity systems, putting each of us in control of our own digital identities.

This work includes researching cryptographic algorithms that enable privacy-preserving open-source user-centric identity systems. Such systems would let people present cryptographically signed electronic claims and selectively choose which information they wish to disclose, while preventing tracking of people between presentations of the claim. Our approach would preserve an individual’s privacy and work with existing web protocols to provide easy and safe access to a wide range of resources and activities.

Our research also includes investigating innovative ways for people to manage their identity secrets reliably and safely without having to provide any centralized party with full access to them. Success in this area will also require scalable and verifiable methods to distribute identity public keys, so people can know who exactly they are interacting with.

Media provenance and authenticity 

Advances in graphics and machine learning algorithms have enabled the creation of easy-to-use tools for editing While useful in many ways, this technology has also enabled fraud and manipulation of digital images and media – or deepfakes. Early fakes were easy to spot, but current versions are becoming nearly impossible for machines or people to detect. The potential proliferation of fakes that are indistinguishable from reality undermines society’s trust in everything we see and hear.

Rather than trying to detect fakes, Microsoft Research has developed technology to determine the source of any digital media and whether it has been altered. We do this by adding digitally signed manifests to video, audio or images. The source of these media objects might be well-known news organizations, governments or even individuals using apps on mobile devices. 

Since media creation, distribution, and consumption are complex and involve many industries, Microsoft has helped forma standards organization to stipulate how these signatures are added to media objects. We are also working with news organizations such as the BBC, New York Times, and CBC to promote media provenance as a mitigation for misinformation on social media networks. 

Hardware security foundations 

To promote cyber-resilience, we are developing systems which can detect a cyberattack and safely shut down protecting data and blocking the attacker. The systems are designed to be repaired quickly and securely, if compromised. These systems are built with simple hardware features that provide very high levels of protection for repair and recovery modules. To enable reliable detection of compromised systems, we are also developing storage features that can be used to protect security event logs. This makes it harder for attackers to cover their tracks.

Security analytics 

Modern-day computers and networks are under constant attack by hackers of all kinds. In this seemingly never-ending cat-and-mouse contest, securing and defending today’s global systems is a multi-billion-dollar enterprise. Managing the massive quantities of security data collected is increasingly challenging, which creates an urgent need for disruptive innovation in security analytics. 

We are investigating a transformer-based approach to modeling and analyzing large-scale security data. Applying and tuning such models is a novel field of study that could change the game for security analytics.

Privacy-preserving machine learning

A privacy-preserving AI system should generalize so well that its behavior reveals no personal or sensitive details that may have been contained in the original data on which it was trained.

How close can we get to this ideal? Differential privacy can enable analysts to extract useful insights from datasets containing personal information even while strengthening privacy protections. This method introduces “statistical noise.” The noise is significant enough that AI models are prevented from compromising the privacy of any individual, but still provide accurate, useful research findings. Our recent results show that large language models can be particularly effective differentially private learners.

Another approach, federated learning, enables large models to be trained and fine-tuned on customers’ own devices to protect the privacy of their data, and to respect data boundaries and data-handling policies. At Microsoft Research, we are creating an orchestration infrastructure for developers to deploy cross-platform, cross-device federated learning solutions.

Protecting data in training or fine-tuning is just one piece of the puzzle. Whenever AI is used in a personalized context, it may unintentionally leak information about the target of the personalization. Therefore, we must be able to describe the threat model for a complete deployment of a system with AI components, rather than just a single part of it.

Read more about our work on these and other related topics in an earlier blog post.

Confidential computing

Confidential computing has emerged as a practical solution to securing compute workloads in cloud environments, even from malicious cloud administrators. Azure already offers confidential computing environments in multiple regions, leveraging Trusted Execution Environments (TEEs) available in multiple hardware platforms.

Imagine if all computation were taking place in TEEs, where services would be able to access sensitive data only after they had been attested to perform specific tasks. This is not practical today and much research remains to be done. For example, there are no formal standards to even describe what a TEE is, what kind of programming interface a TEE cloud should have, or how different TEEs should interact.

Additionally, it is important to continuously improve the security guarantees of TEEs. For instance, understanding which side-channel attacks are truly realistic and developing countermeasures remains a major topic for research. Furthermore, we need to continue researching designs for confidential databases, confidential ledgers and confidential storage. Finally, even if we build both confidential computing and storage environments, how can we establish trust in the code that we want to run? As a cloud provider, our customers expect us to work continuously on improving the security of our infrastructure and the services that run on it.

Secure-by-design cloud

In the future, we can imagine Azure customers compiling their software for special hardware with memory tagging capabilities, eliminating problems like buffer overflows for good. To detect compromise, VM memory snapshots could be inspected and studied with AI-powered tools. In the worst case, system security could always be bootstrapped from a minimal hardware root of trust. At Microsoft Research, we are taking a step further and asking how we can build the cloud from the ground up, with security in mind.

New cryptography

The advance of quantum computing presents many exciting potential opportunities. As a leader in both quantum computing development and cryptographic research, Microsoft has a responsibility to ensure that the groundbreaking innovations on the horizon don’t compromise classical (non-quantum) computing systems and information. Working across Microsoft, we are learning more about the weaknesses of classical cryptography and how to build new cryptographic systems strong enough to resist future attacks.

Our active participation in the National Institute of Standards and Technology (NIST) Post-Quantum Cryptography projects has allowed Microsoft Research to examine deeply how the change to quantum-resistant algorithms will impact Microsoft services and Microsoft customers. With over seven years of work in this area, Microsoft Research’s leadership in quantum cryptography will help customers prepare for the upcoming change of cryptographic algorithms.

We’ve joined with the University of Waterloo and others to build a platform for experimenting with the newly proposed cryptographic systems and applying them to real-world protocols and scenarios. We’ve implemented real-world tests of post-quantum cryptography, to learn how these new systems will work at scale and how we can deploy them quickly to protect network tunnels Our specialized hardware implementations and cryptanalysis provide feedback to the new cryptosystems, which improves their performance, making post-quantum cryptosystems smaller and stronger.

ElectionGuard

Advances in cryptography are enabling end-to-end verifiable elections and risk-limiting audits for elections. Our open-source ElectionGuard project uses cryptography to confirm all votes have been correctly counted. Individual voters can see that their vote has been accurately recorded and anyone can check that all votes have been correctly tallied—yet individual ballots are kept secret. Risk-limiting audits use advanced statistical methods that can determine when an election audit has hit a pre-determined level of confidence with greater efficiency than traditional audits.

The cryptography tools that enable verifiable voting are Shamir Secret Sharing, Threshold Encryption, and additive Homomorphic Encryption. The math is interesting, and we will explore that in future blog posts, but there’s much more than math to ElectionGuard.

Securing the future

Through our work, we aim to continue to earn customer trust, striving to ensure that Microsoft’s products and services and our customer’s information will remain safe and secure for years to come.

Forthcoming entries in this blog series will include more details on the areas covered in this post and more. Much of our work is open-source and published, so we will be highlighting our GitHub projects and other ways you can interact directly with our work.

Have a question or topic that you would like to see us address in a future post? Please contact us!

The post Research trends in privacy, security and cryptography appeared first on Microsoft Research.

Read More