Single particle, cryogenic electron microscopy (cryo-EM) experiments now routinely produce high-resolution data for large proteins and their complexes. Building an atomic model into a cryo-EM density map is challenging, particularly when no structure for the target protein is known a priori. Existing protocols for this type of task often rely on significant human intervention and can take hours to many days to produce an output. Here, we present a fully automated, template-free model building approach that is based entirely on neural networks. We use a graph convolutional network (GCN) to generate an embedding from a set of rotamer-based amino acid identities and candidate 3-dimensional C$\alpha$ locations. Starting from this embedding, we use a bidirectional long short-term memory (LSTM) module to order and label the candidate identities and atomic locations consistent with the input protein sequence to obtain a structural model. Our approach paves the way for determining protein structures from cryo-EM densities at a fraction of the time of existing approaches and without the need for human intervention.

K-12 classrooms consistently integrate collaboration as part of their learning experiences. However, owing to large classroom sizes, teachers do not have the time to properly assess each student and give them feedback. In this paper we propose using simple deep-learning-based machine learning models to automatically determine the overall collaboration quality of a group based on annotations of individual roles and individual level behavior of all the students in the group. We come across the following challenges when building these models: 1) Limited training data, 2) Severe class label imbalance. We address these challenges by using a controlled variant of Mixup data augmentation, a method for generating additional data samples by linearly combining different pairs of data samples and their corresponding class labels. Additionally, the label space for our problem exhibits an ordered structure. We take advantage of this fact and also explore using an ordinal-cross-entropy loss function and study its effects with and without Mixup.

Out-of-training-distribution (OOD) scenarios are a common challenge of learning agents at deployment, typically leading to arbitrary deductions and poorly-informed decisions. In principle, detection of and adaptation to OOD scenes can mitigate their adverse effects. In this paper, we highlight the limitations of current approaches to novel driving scenes and propose an epistemic uncertainty-aware planning method, called \emph{robust imitative planning} (RIP). Our method can detect and recover from some distribution shifts, reducing the overconfident and catastrophic extrapolations in OOD scenes. If the model's uncertainty is too great to suggest a safe course of action, the model can instead query the expert driver for feedback, enabling sample-efficient online adaptation, a variant of our method we term \emph{adaptive robust imitative planning} (AdaRIP). Our methods outperform current state-of-the-art approaches in the nuScenes \emph{prediction} challenge, but since no benchmark evaluating OOD detection and adaption currently exists to assess \emph{control}, we introduce an autonomous car novel-scene benchmark, \texttt{CARNOVEL}, to evaluate the robustness of driving agents to a suite of tasks with distribution shifts.

Graph Neural Networks (GNNs) are powerful and flexible neural networks that use the naturally sparse connectivity information of the data. GNNs represent this connectivity as sparse matrices, which have lower arithmetic intensity and thus higher communication costs compared to dense matrices, making GNNs harder to scale to high concurrencies than convolutional or fully-connected neural networks. We introduce a family of parallel algorithms for training GNNs and show that they can asymptotically reduce communication compared to previous parallel GNN training methods. We implement these algorithms, which are based on 1D, 1.5D, 2D, and 3D sparse-dense matrix multiplication, using torch.distributed on GPU-equipped clusters. Our algorithms optimize communication across the full GNN training pipeline. We train GNNs on over a hundred GPUs on multiple datasets, including a protein network with over a billion edges.

Video captioning has shown impressive progress in recent years. One key reason of the performance improvements made by existing methods lie in massive paired video-sentence data, but collecting such strong annotation, i.e., high-quality sentences, is time-consuming and laborious. It is the fact that there now exist an amazing number of videos with weak annotation that only contains semantic concepts such as actions and objects. In this paper, we investigate using weak annotation instead of strong annotation to train a video captioning model. To this end, we propose a progressive visual reasoning method that progressively generates fine sentences from weak annotations by inferring more semantic concepts and their dependency relationships for video captioning. To model concept relationships, we use dependency trees that are spanned by exploiting external knowledge from large sentence corpora. Through traversing the dependency trees, the sentences are generated to train the captioning model. Accordingly, we develop an iterative refinement algorithm that refines sentences via spanning dependency trees and fine-tunes the captioning model using the refined sentences in an alternative training manner. Experimental results demonstrate that our method using weak annotation is very competitive to the state-of-the-art methods using strong annotation.

The ability to reason abstractly about events as they unfold is a defining feature of human intelligence. We know instinctively that crying and writing are means of communicating, and that a panda falling from a tree and a plane landing are variations on descending. Organizing the world into abstract categories does not come easily to computers, but in recent years researchers have inched closer by training machine learning models on words and images infused with structural information about the world, and how objects, animals, and actions relate. In a new study at the European Conference on Computer Vision this month, researchers unveiled a hybrid language-vision model that can compare and contrast a set of dynamic events captured on video to tease out the high-level concepts connecting them. Their model did as well as or better than humans at two types of visual reasoning tasks--picking the video that conceptually best completes the set, and picking the video that doesn't fit.

With all that is going on in the world these days, is it frivolous to talk about weather prediction? Asked in the 21st century, this is bound to be a rhetorical question. Today, no lengthy justification is needed as to why prediction of atmospheric states is vital: Due to global warming, frequency and intensity of severe weather conditions – droughts, wildfires, hurricanes, heatwaves – have risen and will continue to rise. And while accurate forecasts don't change those events per se, they constitute essential information in mitigating their consequences. This goes for atmospheric forecasts on all scales: from so-called "nowcasting" (operating on a range of about six hours), over medium-range (three to five days) and sub-seasonal (weekly/monthly), to climate forecasts (concerned with years and decades).

In the big data domain, researchers need to ensure that conclusions are consistently verifiable. But that can be particularly challenging in medicine because physicians themselves aren't always sure about disease diagnoses and treatment plans. To investigate how machine learning research has historically handled medical uncertainties, scientists at the University of Texas at Dallas; the University of California, San Francisco; the National University of Singapore; and over half a dozen other institutions conducted a meta-survey of studies over the past 30 years. They found that uncertainty arising from imprecise measurements, missing values, and other errors was common among data and models but that the problems could potentially be addressed with deep learning techniques. The coauthors sought to quantify the prevalence of two types of uncertainty in the studies: structural uncertainty and uncertainty in model parameters.

Researchers at Google have developed a new deep-learning model called BigBird that allows Transformer neural networks to process sequences up to 8x longer than previously possible. Networks based on this model achieved new state-of-the-art performance levels on natural-language processing (NLP) and genomics tasks. The team described the model and a set of experiments in a paper published on arXiv. BigBird is a new self-attention model that reduces the neural-network complexity of Transformers, allowing for training and inference using longer input sequences. By increasing sequence length up to 8x, the team was able to achieve new state-of-the-art performance on several NLP tasks, including question-answering and document summarization.

Recently, researchers from DeepMind and McGill University proposed new approaches to speed up the solution of complex reinforcement learning problems. They mainly introduced a divide and conquer approach to reinforcement learning (RL), which is combined with deep learning to scale up the potentials of the agents. For a few years now, reinforcement learning has been providing a conceptual framework in order to address several fundamental problems. This algorithm has been utilised in several applications, such as to model robots, simulate artificial limbs, developing self-driving cars, play games like poker, Go, and more. Also, the recent combination of reinforcement learning with deep learning added several impressive achievements and is found to be a promising approach to tackle important sequential decision-making problems that are currently intractable.