Posts in category: Facebook AI

Facebook Connectivity’s mission is to enable better, broader global connectivity to bring more people online to a faster internet. We collaborate with the industry, including telecom operators, community leaders, technology developers, and researchers in order to find solutions that are scalable and sustainable.

COVID-19 has shed light on the ever-increasing need for high-speed connectivity in the home. Since March, we have seen incredible increases in group calls, both in terms of volume and group sizes. People rely on a stable connection for remote learning and education, video conferencing, remote work, collaboration, and to simply connect with friends and family. Beyond the pandemic, having a reliable and fast connection plays an important role in making sure everyone in the world has access to online resources, tools, and experiences.

To learn more about Facebook Connectivity, we chatted with Julius Kusuma, Research Scientist at Facebook Connectivity. Julius came to Facebook about two years ago after spending some time working on subsurface and subsea connectivity systems. He’s worked in a number of areas in connectivity, including acoustic telemetry, electro-magnetics, and wireless — from megameter-wave and kilometer-wave to millimeter-wave systems.

In this Q&A, we ask Julius about his background, the role of millimeter-wave technology in connectivity, the importance of academic collaborations, and more.

Q: What brought you to Facebook Connectivity?

Julius Kusuma: When I first read about Facebook Connectivity, I was surprised to hear that half of the world’s population is still not connected to the internet. And for those who are, connectivity is often poor. Access to the internet has never been more important than it is today for education, work, and building global community. This speaks to me personally, as I rely on the internet to connect to my friends and family members around the world, especially now in light of the recent coronavirus pandemic.

Growing up, I was fortunate to have been able to learn about education opportunities thanks to the voiceband modem that my parents bought for me in middle school. Having access to basic email, a web browser, and newsgroups opened up the world to me. I went down countless rabbit holes learning about so many topics, and I even had to ask my parents for more data! Fortunately, having become friends with the local ISP that provided the service, they were kind enough to give me extra megabytes of data every once in a while to keep me going.

Today’s internet is much richer and more powerful than what I had access to at that time. We’d like to enable that for people everywhere.

At Facebook, I work in a small team that provides research-driven innovation to support Connectivity’s mission. My personal research spans several areas, including rural connectivity and wireless technologies. Connectivity challenges are complex and multi-dimensional, so I work with academic and industry partners with expertise in wireless communication, radio propagation modeling, electronics, algorithms, econometrics, and many others.

Q: What is millimeter-wave (mmWave) wireless technology and why is there so much interest in it within the field of connectivity?

JK: MmWave refers to radio waves in the 30–300 GHz frequency range, which corresponds to wavelengths in the 0.01–1 range mm, hence the name. This frequency band is higher than conventional bands used for wireless communication. MmWave allows for higher bandwidth, but its range is limited due to the physics of signal propagation. Therefore, we have to develop technologies to maintain good links and manage the network.

Due to its high frequency, it is possible to build very smart devices that can self-organize and self-optimize, using new technologies such as smart antennas, smart data management, and smart routing algorithms. It’s amazing to think that the mobile phone and Wi-Fi can work the way they do today. They have benefited from significant technology development over decades. With mmWave, we can provide another order of magnitude in performance gains, and to do so, we need to continue developing the right technologies to unleash the potential of mmWave, including tools for network planning and management.

We think mmWave plays an important role in connectivity because it allows for us to deploy a high-speed wireless link quickly and inexpensively, by using mmWave to build a self-organizing, self-optimizing network. Therefore, it can solve the “last mile problem” — the last connection to your home or premise.

Q: What is Facebook doing in mmWave technology development?

JK: One major project we are working on is Terragraph. It is a gigabit wireless technology that operates on 60 GHz unlicensed frequency band and delivers fiber-like speeds over the air in areas where trenching new fiber cables may be difficult or cost-prohibitive.

In many parts of the world, fiber access to consumers is cost-prohibitive and slow to deploy due to many factors, including permitting, trenching, and lack of access. Terragraph can be a better alternative to provide fiber-like speeds at a significantly lower cost. It’s also much faster to deploy and can be brought to market in a matter of weeks. To help evaluate Terragraph as a solution for high-speed fixed broadband connectivity and public Wi-Fi, YTL Communications conducted a large-scale trial in George Town, Malaysia, using the technology to connect businesses and offices that had only copper/DSL connections formerly.

As an emerging technology, there are many opportunities for exploration and research and development in mmWave. In our work, we engage closely with the research community through several channels.

First, along with Deutsche Telekom, we created and led the V-band mmWave Channel Sounder program in the Telecom Infra Project (TIP). Second, we are supporting research efforts such as the National Science Foundation’s Platforms for Advanced Wireless Research (PAWR). Third, we have direct research collaborations with a number of universities.

Q: What is the Telecom Infra Project?

JK: Founded by Deutsche Telekom, Intel, Facebook, Nokia, and SK Telecom in 2016, TIP is a global community of companies and organizations that are driving infrastructure solutions to advance global connectivity. Lack of flexibility in the current telecom infrastructure solutions makes it challenging for operators to efficiently build and upgrade networks in a scalable and sustainable way. Therefore, to succeed in Facebook’s mission, we need to help develop a strong ecosystem of researchers, engineering companies, product developers, telecom operators, and service providers. We see TIP as the vehicle to accelerate the telecom industry. Naturally, technology is at the heart of this endeavor.

As a research scientist, I am excited by the open and collaborative nature of TIP. This is exemplified in the recently completed mmWave Channel Sounder program. We wanted to empower the research community to do experimentation and exploration, so we invited applicants to submit proposals describing the experiments that could be done to characterize the mmWave radio channel. Awardees were then given mmWave channel sounder kits. Academic groups also received financial grants to assist in offsetting costs associated with performing the channel measurements.

This program supports the overall goals of the mmWave Networks Project Group to help facilitate the deployment of high-speed applications, such as Fixed Wireless Access, Smart City Connectivity, and Small Cell Backhaul with a focus on 60 GHz spectrum and 802.11ad/802.11ay technologies for outdoor transmission at street-level.

We recently published the findings as a report. Contributors to the report demonstrated mmWave performance in a wide range of use-cases and scenarios including indoor office spaces, urban canyons, suburban areas, and agricultural settings. The report also contains links to the data sets where available, hosted by the National Institute of Standards and Technology.

Q: Why are research collaborations important for Facebook Connectivity’s mission?

JK: Connectivity’s challenges are complex, vast, and diverse. We need the brightest minds to help us not only develop solutions, but also to better understand the problems, identify the opportunities, and build specialized solutions. This means that we need continuing investment in research and development to push the envelope further and further.

Therefore, we seek to complement and extend offerings from existing providers, community leaders, technology developers, and researchers. Open source and open innovations are key elements that empower collaborations. Speaking more personally, I work with a number of university and industry research groups, and a good example is the aforementioned Platform for Advanced Wireless Research, funded by the National Science Foundation (NSF).

PAWR is a research consortium that allows industry and academic collaborators to use research platforms to conduct experiments, build prototypes, and show demonstrations in an open, accessible way. This will accelerate the pathways from research to impact, so that fundamental research can quickly get to the field, benefit from testing, and be validated. This allows us to do experiments in realistic settings, which is very valuable and important. Access to such platforms allow us not only to prove the technology, but also to develop end-to-end workflows from planning to deployment to maintenance — to make sure it performs well, can provide reliable connectivity, and is sustainable.

Facebook Connectivity is a member of this consortium and we provide technologies, equipment, and expertise.

Q: How can the research community get involved?

JK: Development of solutions within TIP takes place in project groups, focusing on three strategic networks areas that collectively make up an end-to-end network: Access, Transport, and Core and Services. By dividing a network into these areas, TIP members can best identify areas in need of innovation and work together to build the right products. These project groups run research efforts and research projects, including grants programs.

Researchers are encouraged to join project groups in their areas of expertise, and get in touch with project group leads for more information. We will also be updating the Facebook Research and TIP websites for collaboration opportunities.

We are excited about the work done in the mmWave Channel Sounder program. Those interested in learning more can have a look at the report. We also have a number of publications showcasing our work in connectivity made available on our website. To learn about Facebook’s broader efforts in Connectivity, visit connectivity.fb.com.

To learn more about PAWR, check out their website.

Collaborative research is important to us, and from time to time we host events. For example, last year we hosted a Rural Connectivity Research Workshop on Facebook’s campus. Another avenue for academic researchers to engage with industry is through the TIP Summit. Finally, startups and early-stage technology developers should look at our new Facebook Accelerator program.

The post Accelerating mmWave wireless research: How collaborations help us connect the unconnected appeared first on Facebook Research.

Read More

While preventive measures are currently the only tools people can use to protect themselves from contracting COVID-19, knowledge about the effectiveness of these measures varies around the world. The COVID-19 Preventive Health Survey is designed to help researchers understand people’s knowledge, attitudes, and practices about COVID-19 prevention. By reporting people’s knowledge about measures such as physical distancing and use of masks, the survey can support policymakers and researchers to develop improved messaging and outreach to communities. This data can also help health systems prepare to address communities where future vaccine acceptance appears low.

This survey was launched in July 2020 in 67 countries and territories through a partnership between Facebook, the Massachusetts Institute of Technology (MIT), and Johns Hopkins University Center for Communication Programs (JHU CCP), and advised on by the World Health Organization Global Outbreak Alert and Response Network. The survey asks people to self-report their adherence to preventive measures, such as washing hands and wearing masks, as well as what they know about COVID-19, including symptoms of the disease, risk factors, and how their community is handling the pandemic. The survey also asks respondents about perceived behaviors of others in their communities and assesses whether communities would want a future COVID-19 vaccine.

“This new survey gives us a different view on COVID-19 by measuring beliefs, knowledge, and behaviors that are critical to effective responses,” says Dean Eckles, Associate Professor at MIT.

Privacy-preserving approach

Facebook and partners designed the survey with privacy in mind from the start. This survey leverages the same model as the COVID-19 Symptom Surveys, where Facebook invites a sample of Facebook users to take a survey off of the Facebook platform by our academic partner. In addition to language preference, Facebook shares a single de-identified statistic known as a survey weight, which doesn’t identify a person but helps researchers correct for nonresponse bias. This helps ensure that the sample more accurately reflects the characteristics of the population represented in the data. Facebook does not share who took the survey with partners, and they do not share individual survey responses with Facebook. Facebook has access only to the aggregate survey data that is publicly available.

Data available for research

• Researchers and others can access aggregate, de-identified survey data via MIT’s public API.
• Academic and nonprofit researchers can also request microdata access here and sign a data use agreement.

Analysis

We used publicly available aggregate data from the Preventive Health Survey in the following analysis. These charts include data on adults ages 20 years or older from countries and territories that received a one-time snapshot survey, as well as countries and territories that received wave surveys, fielded three times between July and August 2020.

Mask knowledge and mask usage

We first analyze the correlation between knowledge of mask effectiveness and mask usage in countries and territories with at least an 1,000 effective sample size. Mask usage includes respondents who self-reported that they wore a mask at least one time in the last week to prevent contracting COVID-19. Mask knowledge includes those who responded that wearing a mask is very or extremely effective in preventing COVID-19. Though we do not control for mask mandates, we see that the percent of people wearing a mask is moderately correlated with the percent of people who believe wearing a mask is effective (correlation coefficient = 0.64).

Preventive behaviors in the United States

We then examine preventive behaviors in the United States by age group, including whether respondents said they washed their hands, wore a mask, or kept at least 1 meter of distance from others in the last week to avoid contracting COVID-19. We see that the 41-to-50 age group is the least likely to practice preventive behaviors overall: This group consistently has the lowest or second-lowest percentage of people across all three preventive behaviors, even though older populations are at an increased risk of COVID-19 complications. Of the three behaviors, keeping a 1-meter distance is the least adopted behavior, especially among the younger age groups. About 10 percent fewer respondents between 20 and 50 years old report keeping a 1-meter distance from others, compared with washing hands and wearing a mask.

COVID-19 vaccine acceptance

Given the number of efforts underway to develop a COVID-19 vaccine, we analyzed the percentage of respondents who said they would take a COVID-19 vaccine, if one were to become available. The following figure shows the countries and territories with the five highest and five lowest percentages of people who said they would take a COVID-19 vaccine. Future research may elucidate how these sentiments change over time and how they are affected by COVID-19 and flu vaccine developments in the near future.

Partnership opportunities and next steps

The Preventive Health Survey adds to a number of data sets and resources that Facebook’s Data for Good program has made available to support partners with public health responses during COVID-19. The survey can be used to tailor risk communication to specific populations, evaluate the efficacy of policies such as stay-at-home orders and masks, and inform decision-making from both policy and program perspectives.

“Covering dozens of countries with this kind of data collection and analysis will be unparalleled,” says J. Douglas Storey, PhD, JHU CCP’s Director of Communication Science and Research. “There has never been any global source of information like this, and it will capture the more nuanced drivers of behavior and a range of psychosocial factors not currently being collected in a systematic way.”

To generate actionable insights from the survey that can be used quickly in national responses, our partners are developing public dashboards. MIT’s dashboard is available here and JHU CCP’s dashboard is coming soon.

We welcome feedback on the usefulness of the data and how similar survey efforts in the future may be most helpful for researchers aiding public health responses. For any questions related to the Preventive Health Survey, please reach out to PHSurvey@fb.com.

The post New available survey data shares insights on COVID-19 prevention and a potential vaccine appeared first on Facebook Research.

Read More

Nima Noorshams, Saurabh Verma, and Aude Hofleitner are Research Scientists at Facebook working within Core Data Science, a research and development team focused on improving Facebook’s processes, infrastructure, and products.

What we did: Sequence modeling for integrity

Billions of people rely on Facebook products and services to connect with family and friends, build new communities, share experiences, and run their businesses. However, the rise of inauthentic accounts and activities as well as disparaging and threatening content on social media has introduced several integrity challenges. Needless to say, maintaining the integrity of such a large and growing network in a fast and scalable manner is of utmost importance for the safety and security of the online community.

Entities on the platform, such as accounts, posts, pages, and groups, are not static. They interact with one another over time, which can reveal a lot about their nature. For instance, fake accounts and misinformation posts elicit different types of reactions from other accounts than do normal/benign accounts and posts (see Figure 1). In the paper “TIES: Temporal Interaction Embeddings for enhancing social media integrity at Facebook,” we focus on the problem of leveraging these interactions in order to enhance the integrity of the platform.

In short, TIES is a deep learning, application-agnostic, scalable framework for embedding sequences of entity interactions. It encodes not only the sequence of actions but also various features of sources and targets of the interactions. The embedding vectors can then be used for various integrity applications, such as detecting fake accounts, identifying misinformation or hate speech, detecting high-risk ad accounts, and many others.

How we did it: Combining graph representations and sequence learning

Past studies have mainly focused on either static or dynamic behaviors of the networks, but not both at the same time. In contrast, the core of TIES consist of two embeddings:

1. Graph-based embedding, which captures the static (or slow-changing) information encoded in the large social graph.
2. Sequence-based embedding, which captures the more dynamic actions.

Prior knowledge, such as friending and group or page memberships, are captured in the social graph. Large-scale embedding algorithms, such as PyTorch-BigGraph, can be used to encode graph information. These graph-based embeddings are then used to initialize the sequence encoder piece of the framework. Figure 2 illustrates the model architecture.

We first convert the sequence of triplets (source, target, action) into feature vectors. These vectors consist of trainable action embeddings, pretrained source and target embeddings (which are produced by PyTorch-BigGraph), as well as miscellaneous features such as time-gap between the actions. The features are then fed into a sequence encoder, which consists of a seq2seq encoding layer, self-attention, and pooling layer. The model parameters are trained by minimizing a loss function over a labeled data set, thus creating supervised embeddings.

TIES applications at Facebook

We have tested this framework on several applications, including detecting misinformation, detecting fake accounts and engagements, and identifying high-risk ad accounts. Different types of actions and features were used for each application. For instance, in detecting misinformation, we used sequences of user actions on posts, such as likes, comments, shares, and so on.

In all the aforementioned applications, we used a portion of the training samples (up to millions of samples) for training the TIES model and then passed that as an additional feature to baseline models (generally complicated models consisting of several hundred carefully engineered features and/or deep learning frameworks that were already deployed into production). In all instances, we observed uniform and statistically significant gains over existing baselines that can contribute to enhancing the integrity of our platform, and TIES features have been deployed into production since.

The post Leveraging online social interactions for enhancing integrity at Facebook appeared first on Facebook Research.

Read More