We propose a method for sample-efficient optimization of the trade-offs between model accuracy and on-device prediction latency in deep neural networks. Neural architecture search (NAS) aims to provide an automated framework that identifies the optimal architecture for a deep neural network machine learning model given an evaluation criterion such as model accuracy. The continuing trend toward deploying models on end user devices such as mobile phones has led to increased interest in optimizing multiple competing objectives in order to achieve an optimal balance between predictive performance and computational complexity (e.g., total number of flops), memory footprint, and latency of the model. Existing NAS methods that rely on reinforcement learning and/or evolutionary strategies can incur prohibitively high computational costs because they require training and evaluating a large number of architectures. Many other approaches require integrating the optimization framework into the training and evaluation workflows, making it difficult to generalize to different production use-cases.

Each year, PhD students from around the world apply for the Facebook Fellowship, a program designed to encourage and support promising doctoral students who are engaged in innovative and relevant research in areas related to computer science and engineering. Fellowship recipients receive tuition funding for up to two years to conduct their research at their respective universities, independently of Facebook. To learn about award details, eligibility, and more, visit the program page below. Xinyun is a PhD student at UC Berkeley working with Professor Dawn Song and is expected to graduate in 2022. Her research explores the intersection of deep learning, programming languages, and security, focused on neural program synthesis and adversarial machine learning (ML).

This tutorial of Deep Learning on Graphs for Natural Language Processing (DLG4NLP) is timely for the computational linguistics community, and covers relevant and interesting topics, including automatic graph construction for NLP, graph representation learning for NLP, various advanced GNN based models (e.g., graph2seq, graph2tree, and graph2graph) for NLP, and the applications of GNNs in various NLP tasks (e.g., machine translation, natural language generation, information extraction and semantic parsing). The intended audiences for this tutorial mainly include graduate students and researchers in the field of Natural Language Processing and industry professionals who want to know how the state-of-the-art deep learning on graphs techniques can help solve important yet challenging Natural Language Processing problems.

Real-time rendering in virtual reality presents a unique set of challenges -- chief among them being the need to support photorealistic effects, achieve higher resolutions, and reach higher refresh rates than ever before. To address this challenge, researchers at Facebook Reality Labs developed DeepFocus, a rendering system we introduced in December 2018 that uses AI to create ultra-realistic visuals in varifocal headsets. This year at the virtual SIGGRAPH Conference, we're introducing the next chapter of this work, which unlocks a new milestone on our path to create future high-fidelity displays for VR. Our SIGGRAPH technical paper, entitled "Neural Supersampling for Real-time Rendering," introduces a machine learning approach that converts low-resolution input images to high-resolution outputs for real-time rendering. This upsampling process uses neural networks, training on the scene statistics, to restore sharp details while saving the computational overhead of rendering these details directly in real-time applications.

In December 2019, Facebook launched the Towards On-Device AI request for proposals (RFP). The purpose of this RFP was to support the academic community in addressing fundamental challenges in this research area, to accelerate the transition toward a truly "smart" world where AI capabilities permeate all devices and sensors. "We've seen strong progress in moving AI workloads from the cloud to on-device. Running models locally has already helped drive new capabilities like speech assistants, night modes on cameras, and an entirely new class of intelligent devices like smartwatches and smart thermostats," says Vikas Chandra, Director of AI Research. "This is important to push further to preserve privacy, latency, and compute power, and to help create even more experiences that can be useful to us in everyday life." Models must be capable of constantly learning, adapting, and providing proactive assistance.

We introduce kNN-LMs, which extend a pre-trained neural language model (LM) by linearly interpolating it with a k-nearest neighbors (kNN) model. The nearest neighbors are computed according to distance in the pre-trained LM embedding space, and can be drawn from any text collection, including the original LM training data. Applying this augmentation to a strong WIKITEXT-103 LM, with neighbors drawn from the original training set, our kNN-LM achieves a new state-of-the-art perplexity of 15.79 – a 2.9 point improvement with no additional training. We also show that this approach has implications for efficiently scaling up to larger training sets and allows for effective domain adaptation, by simply varying the nearest neighbor datastore, again without further training. Qualitatively, the model is particularly helpful in predicting rare patterns, such as factual knowledge.

Neural Information Processing Systems (NeurIPS) is the largest conference in AI, with machine learning and neuroscience experts traveling from around the world to discuss the latest advances in the field. For 2019, NeurIPS is taking place in Vancouver, British Columbia from Sunday, December 8 to Saturday, December 14, and is projected to attract over 15,000 attendees. Among the attendees are Facebook researchers and engineers in AI, core data science, networking and infrastructure, augmented and virtual reality, and more. Many will be presenting their research in poster sessions, spotlight talks, workshops, and tutorials at the conference. For a day-by-day schedule of the Facebook research being presented at NeurIPS, click here.

Every week, my team at Invector Labs publishes a newsletter that covers the most recent developments in AI research and technology. You can find this week's issue below. You can sign up for it below. Francois Collet is a well-known name in the artificial intelligence(AI) community. A scientist in Google's AI unit, Mr. Collet rose to prominence with the creation of Keras, one of the most popular deep learning frameworks in the market.

In the quest to align deep learning with the sciences to address calls for rigor, safety, and interpretability in machine learning systems, this contribution identifies key missing pieces: the stages of hypothesis formulation and testing, as well as statistical and systematic uncertainty estimation – core tenets of the scientific method. This position paper discusses the ways in which contemporary science is conducted in other domains and identifies potentially useful practices. We present a case study from physics and describe how this field has promoted rigor through specific methodological practices, and provide recommendations on how machine learning researchers can adopt these practices into the research ecosystem. We argue that both domain-driven experiments and application-agnostic questions of the inner workings of fundamental building blocks of machine learning models ought to be examined with the tools of the scientific method, to ensure we not only understand effect, but also begin to understand cause, which is the raison d'être of science.

Facebook Research and MIT researchers are using the popular video game'Minecraft' to build a new AI assistant that can juggle multiple tasks at once. Facebook is using the popular video game Minecraft to help train a new artificially intelligent assistant that, in the future, could help humans perform a wide range of tasks with a broad range of spoken commands. Facebook Research and MIT researchers quietly published a paper in July outlining how they intend to use Minecraft to train an AI assistant that can multitask rather than perform one task at superhuman levels. "In this work, we have argued for building a virtual assistant situated in the game of Minecraft, in order to study learning from interaction, and especially learning from language interaction," the researchers explain in the published paper. According to the researcher team, Minecraft is the perfect environment to train artificial intelligence because it's what is known as a "sandbox" game, which allows players to roam freely, fight, craft, explore and build objects in an online world.