Recent work in machine learning shows that deep neural networks can be used to solve a wide variety of inverse problems arising in computational imaging. We explore the central prevailing themes of this emerging area and present a taxonomy that can be used to categorize different problems and reconstruction methods. Our taxonomy is organized along two central axes: (1) whether or not a forward model is known and to what extent it is used in training and testing, and (2) whether or not the learning is supervised or unsupervised, i.e., whether or not the training relies on access to matched ground truth image and measurement pairs. We also discuss the trade-offs associated with these different reconstruction approaches, caveats and common failure modes, plus open problems and avenues for future work.

Democratizing access to artificial intelligence (AI) requires competitions that promote the development of sample-efficient learning, as well as ensure the reproducibility and generalizability of results. Sample efficiency is important because practitioners with limited compute resources cannot readily utilize algorithms that require a massive number of samples. The complexity of these stateof-the-art methods is outpacing advancements in computation. Moreover, as methods and domains become more specialized, learning procedures become more fragile: often undocumented modifications can inhibit reproducible results and seeds are chosen to reflect the optimal performance of a given solution [Henderson et al., 2018]. Because the focus of traditional research challenges is the development of new techniques in a particular field, these challenges seek to reward participants for novel solutions. However, submissions with the best performance on the (often highly specified) task tend leverage domain knowledge that is not broadly applicable, leading challenges to open separate tracks where submissions are subjectively evaluated on research novelty [Pavlov et al., 2018]. To encourage participants to develop methods with reproducible and robust training behavior, we propose a challenge paradigm where competitors are evaluated directly on the performance of their learning procedures rather than pre-trained agents. Since competition organizers retrain submissions in a controlled setting they can guarantee reproducibility, and - by retraining submissions using a held-out test set - help ensure generalization of submissions past the environments on which they were trained.

In this paper, we present the first comprehensive categorization of essential commonsense knowledge for answering the Winograd Schema Challenge (WSC). For each of the questions, we invite annotators to first provide reasons for making correct decisions and then categorize them into six major knowledge categories. By doing so, we better understand the limitation of existing methods (i.e., what kind of knowledge cannot be effectively represented or inferred with existing methods) and shed some light on the commonsense knowledge that we need to acquire in the future for better commonsense reasoning. Moreover, to investigate whether current WSC models can understand the commonsense or they simply solve the WSC questions based on the statistical bias of the dataset, we leverage the collected reasons to develop a new task called WinoWhy, which requires models to distinguish plausible reasons from very similar but wrong reasons for all WSC questions. Experimental results prove that even though pre-trained language representation models have achieved promising progress on the original WSC dataset, they are still struggling at WinoWhy. Further experiments show that even though supervised models can achieve better performance, the performance of these models can be sensitive to the dataset distribution. WinoWhy and all codes are available at: https://github.com/HKUST-KnowComp/WinoWhy.

Under increasing economic and environmental pressure, airlines are constantly seeking new technologies and optimizing flight operations to reduce fuel consumption. However, the current policy on fuel loading, which has a significant impact on aircraft weight, leaves room for improvement. Excess fuel is loaded by dispatchers and(or) pilots to ensure safety because of fuel consumption uncertainties, primarily caused by flight time uncertainties, which cannot be predicted by current Flight Planning Systems (FPS). In this paper, we develop a novel spatial weighted recurrent neural network model to provide better flight time predictions by capturing air traffic information at a national scale based on multiple data sources, including Automatic Dependent Surveillance - Broadcast, Meteorological Airdrome Reports, and airline records. In this model, we adopt recurrent neural network layers to extract spatiotemporal correlations between features utilizing the repetitive traffic patterns and interacting elements in aviation traffic networks. A spatial weighted layer is introduced to learn origin-destination (OD) specific features, and a two-step training procedure is introduced to integrate individual OD models into one model for a national air traffic network. This model was trained and tested using one year of historical data from real operations. Results show that our model can provide a more accurate flight time predictions than the FPS and the LASSO methods, especially for flights with extreme delays. We also show that with the improved flight time prediction, fuel loading can be optimized to reduce fuel consumption by 0.83% for an example airline's fleet without increasing the fuel depletion risk.

Quality control is an essential process in manufacturing to make the product defect-free as well as to meet customer needs. The automation of this process is important to maintain high quality along with the high manufacturing throughput. With recent developments in deep learning and computer vision technologies, it has become possible to detect various features from the images with near-human accuracy. However, many of these approaches are data intensive. Training and deployment of such a system on manufacturing floors may become expensive and time-consuming. The need for large amounts of training data is one of the limitations of the applicability of these approaches in real-world manufacturing systems. In this work, we propose the application of a Siamese convolutional neural network to do one-shot recognition for such a task. Our results demonstrate how one-shot learning can be used in quality control of steel by identification of defects on the steel surface. This method can significantly reduce the requirements of training data and can also be run in real-time.

Expanding the scope of graph-based, deep-learning models to noncovalent protein-ligand interactions has earned increasing attention in structure-based drug design. Modeling the protein-ligand interactions with graph neural networks (GNNs) has experienced difficulties in the conversion of protein-ligand complex structures into the graph representation and left questions regarding whether the trained models properly learn the appropriate noncovalent interactions. Here, we proposed a GNN architecture, denoted as InteractionNet, which learns two separated molecular graphs, being covalent and noncovalent, through distinct convolution layers. We also analyzed the InteractionNet model with an explainability technique, i.e., layer-wise relevance propagation, for examination of the chemical relevance of the model's predictions. Separation of the covalent and noncovalent convolutional steps made it possible to evaluate the contribution of each step independently and analyze the graph-building strategy for noncovalent interactions. We applied InteractionNet to the prediction of protein-ligand binding affinity and showed that our model successfully predicted the noncovalent interactions in both performance and relevance in chemical interpretation.

Neural language models are becoming the prevailing methodology for the tasks of query answering, text classification, disambiguation, completion and translation. Commonly comprised of hundreds of millions of parameters, these neural network models offer state-of-the-art performance at the cost of interpretability; humans are no longer capable of tracing and understanding how decisions are being made. The attention mechanism, introduced initially for the task of translation, has been successfully adopted for other language-related tasks. We propose AttViz, an online toolkit for exploration of self-attention---real values associated with individual text tokens. We show how existing deep learning pipelines can produce outputs suitable for AttViz, offering novel visualizations of the attention heads and their aggregations with minimal effort, online. We show on examples of news segments how the proposed system can be used to inspect and potentially better understand what a model has learned (or emphasized).

Microsoft and Intel have a novel approach to classifying malware: visualizing it. They're collaborating on STAMINA (Static Malware-as-Image Network Analysis), a project that turns rogue code into grayscale images so that a deep learning system can study them. The approach converts the binary form of an input file into a simple stream of pixels, and turns that into a picture with dimensions that vary depending on aspects like file size. A trained neural network then determines what (if anything) has infected the file. ZDNet noted that the AI is trained on the huge amount of data Microsoft has collected from Windows Defenders installations. The technology doesn't need full-size, pixel-by-pixel recreations of viruses, which makes sense when large malware could easily translate to gigantic pictures.

New and Updated Deep Learning Institute Courses Launched, With More Training Delivery Partners Added. Starting this month, NVIDIA's Deep Learning Institute is offering instructor-led workshops that are delivered remotely via a virtual classroom. DLI provides hands-on training in AI, accelerated computing and accelerated data science to help developers, data scientists and other professionals solve their most challenging problems. These in-depth classes are taught by experts in their respective fields, delivering industry-leading technical knowledge to drive breakthrough results for individuals and organizations. DLI has already trained more than 200,000 developers globally and is growing quickly to bridge the digital skills gap worldwide.

A Bengaluru-based Artificial Intelligence (AI) firm has come up with "designs" of several chemical molecules that may help stop SARS-CoV2 virus, which causes Covid-19 infection from multiplying in an infected person. The firm, which used deep learning technology to identify structures of molecules that would inhibit a critical enzyme of the virus – called 3CLpro, is already in talks with an Indian pharma company, which could synthesise the molecules for testing, if the deal materialises. "We are not in a position to disclose the name of the pharma firm yet," said Vikram Jha, CEO of Pucho Technology Information Limited, the five-year-old firm. The firm, which has office in Bengaluru has been involved in developing an AI platform that offers search and information services to people who have otherwise access to the Internet in languages they are comfortable with when the lockdown happened. "We were ready with our platform but could not launch because of the lockdown. So, we decided to direct our energies towards the Covid-19 fight in a meaningful way using our expertise in AI. That was when two of our engineers specialising in deep learning suggested they could use neural network technology to look for generating potential lead compounds that can target the viral enzyme, 3CLpro," said Jha.