In a study published on the preprint server Arxiv.org, DeepMind researchers describe a reinforcement learning algorithm-generating technique that discovers what to predict and how to learn it by interacting with environments. They claim the generated algorithms perform well on a range of challenging Atari video games, achieving "non-trivial" performance indicative of the technique's generalizability. Reinforcement learning algorithms -- algorithms that enable software agents to learn in environments by trial and error using feedback -- update an agent's parameters according to one of several rules. These rules are usually discovered through years of research, and automating their discovery from data could lead to more efficient algorithms, or algorithms better adapted to specific environments. DeepMind's solution is a meta-learning framework that jointly discovers what a particular agent should predict and how to use the predictions for policy improvement.

In offline reinforcement learning (RL), the goal is to learn a highly rewarding policy based solely on a dataset of historical interactions with the environment. The ability to train RL policies offline can greatly expand the applicability of RL, its data efficiency, and its experimental velocity. Prior work in offline RL has been confined almost exclusively to model-free RL approaches. In this work, we present MOReL, an algorithmic framework for model-based offline RL. This framework consists of two steps: (a) learning a pessimistic MDP (P-MDP) using the offline dataset; and (b) learning a near-optimal policy in this P-MDP. The learned P-MDP has the property that for any policy, the performance in the real environment is approximately lower-bounded by the performance in the P-MDP. This enables it to serve as a good surrogate for purposes of policy evaluation and learning, and overcome common pitfalls of model-based RL like model exploitation. Theoretically, we show that MOReL is minimax optimal (up to log factors) for offline RL. Through experiments, we show that MOReL matches or exceeds state-of-the-art results in widely studied offline RL benchmarks. Moreover, the modular design of MOReL enables future advances in its components (e.g. generative modeling, uncertainty estimation, planning etc.) to directly translate into advances for offline RL.

In this paper, three recently introduced reinforcement learning (RL) methods are used to generate human-interpretable policies for the cart-pole balancing benchmark. The novel RL methods learn human-interpretable policies in the form of compact fuzzy controllers and simple algebraic equations. The representations as well as the achieved control performances are compared with two classical controller design methods and three non-interpretable RL methods. All eight methods utilize the same previously generated data batch and produce their controller offline - without interaction with the real benchmark dynamics. The experiments show that the novel RL methods are able to automatically generate well-performing policies which are at the same time human-interpretable. Furthermore, one of the methods is applied to automatically learn an equation-based policy for a hardware cart-pole demonstrator by using only human-player-generated batch data. The solution generated in the first attempt already represents a successful balancing policy, which demonstrates the methods applicability to real-world problems.

The field of cybersecurity has mostly been a cat-and-mouse game with the discovery of new attacks leading the way. To take away an attacker's advantage of reconnaissance, researchers have proposed proactive defense methods such as Moving Target Defense (MTD). To find good movement strategies, researchers have modeled MTD as leader-follower games between the defender and a cyber-adversary. We argue that existing models are inadequate in sequential settings when there is incomplete information about a rational adversary and yield sub-optimal movement strategies. Further, while there exists an array of work on learning defense policies in sequential settings for cyber-security, they are either unpopular due to scalability issues arising out of incomplete information or tend to ignore the strategic nature of the adversary simplifying the scenario to use single-agent reinforcement learning techniques. To address these concerns, we propose (1) a unifying game-theoretic model, called the Bayesian Stackelberg Markov Games (BSMGs), that can model uncertainty over attacker types and the nuances of an MTD system and (2) a Bayesian Strong Stackelberg Q-learning (BSS-Q) approach that can, via interaction, learn the optimal movement policy for BSMGs within a reasonable time. We situate BSMGs in the landscape of incomplete-information Markov games and characterize the notion of Strong Stackelberg Equilibrium (SSE) in them. We show that our learning approach converges to an SSE of a BSMG and then highlight that the learned movement policy (1) improves the state-of-the-art in MTD for web-application security and (2) converges to an optimal policy in MTD domains with incomplete information about adversaries even when prior information about rewards and transitions is absent.

Deep counterfactual value networks combined with continual resolving provide a way to conduct depth-limited search in imperfect-information games. However, since their introduction in the DeepStack poker AI, deep counterfactual value networks have not seen widespread adoption. In this paper we introduce several improvements to deep counterfactual value networks, as well as counterfactual regret minimization, and analyze the effects of each change. We combined these improvements to create the poker AI Supremus. We show that while a reimplementation of DeepStack loses head-to-head against the strong benchmark agent Slumbot, Supremus successfully beats Slumbot by an extremely large margin and also achieves a lower exploitability than DeepStack against a local best response. Together, these results show that with our key improvements, deep counterfactual value networks can achieve state-of-the-art performance.

We present the Battlesnake Challenge, a framework for multi-agent reinforcement learning with Human-In-the-Loop Learning (HILL). It is developed upon Battlesnake, a multiplayer extension of the traditional Snake game in which 2 or more snakes compete for the final survival. The Battlesnake Challenge consists of an offline module for model training and an online module for live competitions. We develop a simulated game environment for the offline multi-agent model training and identify a set of baseline heuristics that can be instilled to improve learning. Our framework is agent-agnostic and heuristics-agnostic such that researchers can design their own algorithms, train their models, and demonstrate in the online Battlesnake competition. We validate the framework and baseline heuristics with our preliminary experiments. Our results show that agents with the proposed HILL methods consistently outperform agents without HILL. Besides, heuristics of reward manipulation had the best performance in the online competition. We open source our framework at https://github.com/awslabs/sagemaker-battlesnake-ai.

A new generation of automated bin picking systems using deep learning is evolving to support increasing demand for e-commerce. To accommodate a wide variety of products, many automated systems include multiple gripper types and/or tool changers. However, for some objects, sequential grasp failures are common: when a computed grasp fails to lift and remove the object, the bin is often left unchanged; as the sensor input is consistent, the system retries the same grasp over and over, resulting in a significant reduction in mean successful picks per hour (MPPH). Based on an empirical study of sequential failures, we characterize a class of "sequential failure objects" (SFOs) -- objects prone to sequential failures based on a novel taxonomy. We then propose three non-Markov picking policies that incorporate memory of past failures to modify subsequent actions. Simulation experiments on SFO models and the EGAD dataset suggest that the non-Markov policies significantly outperform the Markov policy in terms of the sequential failure rate and MPPH. In physical experiments on 50 heaps of 12 SFOs the most effective Non-Markov policy increased MPPH over the Dex-Net Markov policy by 107%.

"This is a great opportunity to continue the synergistic virtuous circle' that has connected neuroscience and AI for decades." Artificial Neural networks occasionally get the bad rap for watering down the complexity of how a human brain works with over the top analogies. But, there is no denying the fact that popular algorithms were heavily inspired by how the natural systems work. Now, after three decades of innovation and inventions, AI as a domain has touched many functionalities of human cognition. From attention to memory to dreams, there is an active research space that is burgeoning with every passing day. Now a team of researchers at DeepMind are exploring the possibility of reverse engineering the results of algorithms to know more about cognitive functions.

Social network structure is one of the key determinants of human language evolution. Previous work has shown that the network of social interactions shapes decentralized learning in human groups, leading to the emergence of different kinds of communicative conventions. We examined the effects of social network organization on the properties of communication systems emerging in decentralized, multi-agent reinforcement learning communities. We found that the global connectivity of a social network drives the convergence of populations on shared and symmetric communication systems, preventing the agents from forming many local "dialects". Moreover, the agent's degree is inversely related to the consistency of its use of communicative conventions. These results show the importance of the basic properties of social network structure on reinforcement communication learning and suggest a new interpretation of findings on human convergence on word conventions.

The unprecedented progress in Artificial Intelligence (AI) [1-6], over the last decade, came alongside of multiple AI failures [7, 8] and cases of dual use [9] causing a realization [10] that it is not sufficient to create highly capable machines, but that it is even more important to make sure that intelligent machines are beneficial [11] for the humanity. This lead to the birth of the new subfield of research commonly known as AI Safety and Security [12] with hundreds of papers and books published annually on different aspects of the problem [13-31]. All such research is done under the assumption that the problem of controlling highly capable intelligent machines is solvable, which has not been established by any rigorous means. However, it is a standard practice in computer science to first show that a problem doesn't belong to a class of unsolvable problems [32, 33] before investing resources into trying to solve it or deciding what approaches to try. Unfortunately, to the best of our knowledge no mathematical proof or even rigorous argumentation has been published demonstrating that the AI control problem may be solvable, even in principle, much less in practice. Or as Gans puts it citing Bostrom: "Thusfar, AI researchers and philosophers have not been able to come up with methods of control that would ensure [bad] outcomes did not take place …" [34].