AI research

AI research April 28, 2022

The graph connection
Graph neural networks (GNNs) can process graph-based information to make predictions. They learn to encode information about the local surroundings of each node in a graph. GNNs can perform prediction or classification tasks at the level of the whole graph, or for each node or edge. They have proven to be useful in scientific applications such as genomics, molecular design, drug development and physics simulations.

Anticipating others’ behavior on the road
MIT researchers break down complex problem into smaller pieces to solve it in real-time. They use relationships between two road users to predict future trajectories for multiple agents. The MIT technique even outperformed Waymo’s recently published model. The simplicity is definitely a plus, says co-lead author Xin “Cyrus” Huang, a graduate student.

Generating new molecules with graph grammar
Researchers at MIT and IBM use a graph-graph model to build new synthesizable molecules. They treat the formation of atoms and chemical bonds as a graph and develop a graph grammar. Using the grammar and production rules that were inferred from the training set, the model can not only reverse engineer its examples, but can create new compounds in a systematic and data-efficient way.

Understanding Deep Learning Algorithms that Leverage Unlabeled Data, Part 2: Contrastive Learning
Labels for large-scale datasets are expensive to curate, so leveraging abundant unlabeled data before fine-tuning them on the smaller, labeled, data sets is an important and promising direction for pre-training machine learning models. Contrastive learning is a powerful class of self-supervised visual representation learning methods that learn feature extractors by (1) minimizing the distance between the representations of positive pairs, or samples that are similar in some sense, and (2) maximizing the distance.

Reusability report: Capturing properties of biological objects and their relationships using graph neural networks
Graph embedding on biomedical networks: methods, applications and evaluations. Researchers use graph neural networks to identify new cancer genes and their associated molecular mechanisms. They also improve the prediction of cancer driver genes using graph convolutional networks. Their findings will be published at the 27th International ACM SIGKDD Conference on Knowledge Discovery and Data Mining (2021). The authors also discuss the role of deep learning on graph networks in data mining.

Half a decade of graph convolutional networks
Graphs provide a powerful way to model data in many real-world applications, such as the World Wide Web, social networks and communication networks. Predictions and classification tasks related to such systems can be tackled with graph neural networks. Graph neural networks learn a function that maps each node of a graph to a vector in a low-dimensional vector space. These embeddings can be fed into different machine learning algorithms to perform various tasks.

The transformational role of GPU computing and deep learning in drug discovery
Advances in DL, particularly in computer vision and language processing, revived the recent interest of CADD researchers in neural networks. Merck is credited with popularizing DL for CADD through the Kaggle competition on Molecular Activity Challenge in 2012 (ref. 40) The emergence of GPU-enabled DL architectures, along with the proliferation of chemical genomics data, has led to meaningful CADD-enabled discoveries of clinical drug candidates. AI-driven companies are reporting successes in augmented drug discovery.

Solving the challenges of robotic pizza-making
Researchers at MIT, Carnegie Mellon University, and the University of California at San Diego, have created a framework for a robotic manipulation system that uses a two-stage learning process. A “teacher’s algorithm solves each step the robot must take to complete the task. Then it trains a ‘student’ machine-learning model that learns abstract ideas about when and how to execute each skill it needs.

Generating chit-chat including laughs, yawns, ‘ums,’ and other nonverbal cues from raw audio
Generative Spoken Language Model (GSLM) discovers structured content by addressing raw audio signals, without any labels or text. It enables NLP models to capture the expressivity of oral language. Today’s AI systems fail to capture these rich, expressive signals because they learn only from written text. Today, we’ve open-sourced the Textless Python Library, which machine learning practitioners can build experiments on top of GSLM components.

Grading Complex Interactive Coding Programs with Reinforcement Learning
NeurIPS 2021 paper explores challenges in treating interactive coding assignment grading as game playing. Can the same algorithms that master Atari games help us grade these game assignments? We introduce the Play to Grade Challenge to students who develop games as part of a programming assignment. Massive Online Coding Education has reached striking success over the past decade. As a non-profit organization, Code.org claims to have reached over 60 million learners across the world.

An optimized solution for face recognition
The human brain seems to care a lot about faces. It’s dedicated a specific area to identifying them, and the neurons there are so good at their job that most of us can readily recognize thousands of individuals. With artificial intelligence, computers can now recognize faces with a similar efficiency. The finding suggests that the millions of years of evolution that have shaped circuits in the human brain have optimized our system for facial recognition.

Does this artificial intelligence think like a human?
Researchers at MIT and IBM Research have created a method that enables a user to aggregate, sort, and rank these individual explanations to rapidly analyze a machine-learning model’s behavior. Their technique, called Shared Interest, incorporates quantifiable metrics that compare how well a model’s reasoning matches that of a human. The paper will be presented at the Conference on Human Factors in Computing Systems.

How does Astro localize itself in an ever-changing home?
Amazon’s Astro’s Intelligent Motion system relies on visual simultaneous localization and mapping. V-SLAM enables a robot to use visual data to simultaneously construct a map of its environment and determine its position on that map. Astro then processes the visual features, estimated sensor poses, and loop-closure information and optimizes it to obtain a global motion trajectory and map.

Molecular contrastive learning of representations via graph neural networks
MolCLR (Molecular Contrastive Learning of Representations via Graph Neural Networks) is self-supervised learning framework that leverages large unlabelled data (~10 million unique molecules). MolCLRs learn to embed molecules into representations that can distinguish chemically reasonable molecular similarities. The framework significantly improves the performance of graph-neural-network encoders on various molecular property benchmarks including classification and regression tasks.

Leave a Reply

Your email address will not be published. Required fields are marked *