Hierarchical DQN (h-DQN) is a two-level architecture of feedforward neural networks where the meta level selects goals and the lower level takes actions to achieve the goals. We show tasks that cannot be solved by h-DQN, exemplifying the limitation of this type of hierarchical framework (HF). We describe the recurrent hierarchical framework (RHF), generalizing architectures that use a recurrent neural network at the meta level. We analyze the expressiveness of HF and RHF using context-sensitive grammars. We show that RHF is more expressive than HF. We perform experiments comparing an implementation of RHF with two HF baselines; the results corroborate our theoretical findings.

Prior Networks are a class of models which yield interpretable measures of uncertainty and have been shown to outperform state-of-the-art ensemble approaches on a range of tasks. However, Prior Networks have so far been developed only for classification tasks. The properties of Regression Prior Networks are demonstrated on synthetic data, selected UCI datasets, and two monocular depth estimation tasks. They yield performance competitive with ensemble approaches. However, in order to improve the safety of AI systems (Amodei et al., 2016) and avoid costly mistakes in high-risk applications, such as self-driving cars, it is desirable for models to yield estimates of uncertainty in their predictions. Ensemble methods are known to yield both improved predictive performance and robust uncertainty estimates (Gal & Ghahramani, 2016; Lakshminarayanan et al., 2017; Maddox et al., 2019). Importantly, ensemble approaches allow interpretable measures of uncertainty to be derived via a mathematically consistent probabilistic framework. Specifically, the overall total uncertainty can be decomposed into data uncertainty, or uncertainty due to inherent noise in the data, and knowledge uncertainty, which is due to the model having limited uncertainty of the test data (Malinin, 2019). Uncertainty estimates derived from ensembles have been applied to the detection of misclassifications, out-of-domain inputs and adversarial attack detection (Carlini & Wagner, 2017; Smith & Gal, 2018), and active learning (Kirsch et al., 2019). Unfortunately, ensemble methods may be computationally expensive to train and are always expensive during inference.

Victoria Stodden (vcs@stodden.net) is a statistician and associate professor at the University of Illinois at Urbana-Champaign, IL, USA. This material is based upon work supported by National Science Foundation Award #1941443.

From foreign intervention in free elections to the rise of the American surveillance state, the Internet has transformed the relationship between the public and private sectors, especially democracy's public sphere. The global pandemic only further highlights the extent to which technological innovation is changing how we live, work, and play. What has too often gone unacknowledged is that the same revolution has produced a series of conflicts between our desires as consumers and our duties as citizens. Left unaddressed, the consequence is a moral vacuum that has become a threat to liberal democracy and human values. Surveillance in the Internet Age, whether by governments or companies, often relies on algorithmic searches of big data.

A second burst of interest in AI authorship broke out in the mid-1980s. Congress once again commissioned a study, this time from its Office of Technology Assessment (OTA), to address this and other controversial computer-related issues. OTA did not offer an answer to the question, perhaps in part because at that time, it was a "toy problem" because no commercially significant outputs of AI or other software programs had yet been generated.5 But deep learning and other AI breakthroughs have caused IP professionals to rethink the AI authorship issue.1,2 For example, The Next Rembrandt video features a group of art experts and computer scientists discussing how they collaborated to digitize many Rembrandt paintings, develop models of particular features of the paintings, and then create a Rembrandt-like portrait of a man with facial hair wearing a hat and looking to the right.6 The resulting AI-generated painting really does look like a Rembrandt.

Alongside the White House Office of Science and Technology Policy (OSTP), IBM announced in March that it would help coordinate an effort to provide hundreds of petaflops of compute to scientists researching the coronavirus. As part of the newly launched COVID-19 High Performance Computing (HPC) Consortium, IBM pledged to assist in evaluating proposals and to provide access to resources for projects that "make the most immediate impact." Much work remains, but some of the Consortium's most prominent members -- among them Microsoft, Intel, and Nvidia -- claim that progress is being made. Powerful computers allow researchers to undertake high volumes of calculations in epidemiology, bioinformatics, and molecular modeling, many of which would take months on traditional computing platforms (or years if done by hand). Moreover, because the computers are available in the cloud, they enable teams to collaborate from anywhere in the world. Insights generated by the experiments can help advance our understanding of key aspects of COVID-19, such as viral-human interaction, viral structure and function, small molecule design, drug repurposing, and patient trajectory and outcomes.

Following a proposal by the Conference of Presidents (president and chairs of political groups), plenary set out the responsibilities, numerical strength and term of office of three special committees and the tax subcommittee. Lists of members will be announced in a subsequent plenary session. After a series of special committees and a committee of inquiry, created to delve into the various tax leaks and scandals of recent years, Parliament today established a more permanent setup to shed light on the matter. The subcommittee on tax matters will be composed of 30 members. It will deal particularly with the fight against tax fraud, tax evasion and tax avoidance, as well as financial transparency for taxation purposes.

MIT researchers have created an AI tool capable of generating time-lapse videos that predict how human artists use their hands to create watercolor or digital paintings. The AI is trained using time-lapse videos of people making art on Vimeo and YouTube. The probabilistic model can synthesize and predict moments in the painting process from just a single image of an artwork. The network is meant to mimic the ability skilled human artists possess to see a piece of art and comprehend the series of brush strokes or steps a person took to put it together. There are often many possible ways to create a given painting.

Vedantu, a pioneer in LIVE online learning platform in India, is assisting the Ministry of Health (Govt of Karnataka) to impart COVID training and communication to doctors across the State. To assist the Karnataka State Government & Doctors in the ongoing battle against Covid-19, Vedantu has volunteered to offer its WAVE platform. This initiative is helping the ministry to reach out to 250 doctors across Karnataka and train them LIVE using Vedantu's platform to improve their understanding on the following – This training is led by senior officials – Dr. Ravi Kumar, Senior Regional Director at Ministry of Health & FW, Govt of India and Dr Mahamood Shariff, Research officer, NVBDCP, Directorate of H&FW, Govt of Karnataka. Speaking on the partnership, Dr. PC Jaffer (IAS), Secretary to Govt (Expenditure), Finance Department, Karnataka Govt Secretariat, said, "In the current times, taking the risk of face to face training was out of question for us. It was imperative to find a solution that mirrors, but also has the same impact as the face to face training. Hence, we decided to choose Vedantu's platform, given its rich and interactive platform capabilities, to train and converse online with our COVID warriors across the state. We are glad as our training sessions are happening smoothly now and we continue our fight against COVID. "It is Vedantu's great honour to support doctors who are in the front line of the fight against the pandemic.

Multi-agent reinforcement learning (MARL) requires coordination to efficiently solve certain tasks. Fully centralized control is often infeasible in such domains due to the size of joint action spaces. Coordination graph based formalization allows reasoning about the joint action based on the structure of interactions. However, they often require domain expertise in their design. This paper introduces the deep implicit coordination graph (DICG) architecture for such scenarios. DICG consists of a module for inferring the dynamic coordination graph structure which is then used by a graph neural network based module to learn to implicitly reason about the joint actions or values. DICG allows learning the tradeoff between full centralization and decentralization via standard actor-critic methods to significantly improve coordination for domains with large number of agents. We apply DICG to both centralized-training-centralized-execution and centralized-training-decentralized-execution regimes. We demonstrate that DICG solves the relative overgeneralization pathology in predatory-prey tasks as well as outperforms various MARL baselines on the challenging StarCraft II Multi-agent Challenge (SMAC) and traffic junction environments.