We present a method to incrementally generate complete 2D or 3D scenes with the following properties: (a) it is globally consistent at each step according to a learned scene prior, (b) real observations of a scene can be incorporated while observing global consistency, (c) unobserved regions can be hallucinated locally in consistence with previous observations, hallucinations and global priors, and (d) hallucinations are statistical in nature, i.e., different scenes can be generated from the same observations. To achieve this, we model the virtual scene, where an active agent at each step can either perceive an observed part of the scene or generate a local hallucination. The latter can be interpreted as the agent's expectation at this step through the scene and can be applied to autonomous navigation. It can otherwise sample entirely imagined scenes from prior distributions. Besides autonomous agents, applications include problems where large data is required for building robust real-world applications, but few samples are available. We demonstrate efficacy on various 2D as well as 3D data.

Recent breakthroughs in AI for multi-agent games like Go, Poker, and Dota, have seen great strides in recent years. Yet none of these games address the real-life challenge of cooperation in the presence of unknown and uncertain teammates. This challenge is a key game mechanism in hidden role games. Here we develop the DeepRole algorithm, a multi-agent reinforcement learning agent that we test on "The Resistance: Avalon", the most popular hidden role game. DeepRole combines counterfactual regret minimization (CFR) with deep value networks trained through self-play.

We consider the networked multi-agent reinforcement learning (MARL) problem in a fully decentralized setting, where agents learn to coordinate to achieve joint success. This problem is widely encountered in many areas including traffic control, distributed control, and smart grids. We assume each agent is located at a node of a communication network and can exchange information only with its neighbors. Using softmax temporal consistency, we derive a primal-dual decentralized optimization method and obtain a principled and data-efficient iterative algorithm named {\em value propagation}. We prove a non-asymptotic convergence rate of $\mathcal{O}(1/T)$ with nonlinear function approximation.

Adversarial training (AT) has become a popular choice for training robust networks. However, by virtue of its formulation, AT tends to sacrifice clean accuracy heavily in favor of robustness. Furthermore, AT with a large perturbation budget can cause models to get stuck at poor local minima and behave like a constant function, always predicting the same class. To address the above concerns we propose Adversarial Training with Early Stopping (ATES). The design of ATES is guided by principles from curriculum learning that emphasizes on starting "easy" and gradually ramping up on the "difficulty" of training. We do so by early stopping the adversarial example generation step in AT, progressively increasing difficulty of the samples the network trains on. This stabilizes network training even for large perturbation budgets and allows the network to operate at a better clean accuracy versus robustness trade-off curve compared to AT. Functionally, this leads to a significant improvement in both clean accuracy and robustness for ATES models.

We consider training machine learning models using Training data located on multiple private and geographically-scattered servers with different privacy settings. Due to the distributed nature of the data, communicating with all collaborating private data owners simultaneously may prove challenging or altogether impossible. In this paper, we develop differentially-private asynchronous algorithms for collaboratively training machine-learning models on multiple private datasets. The asynchronous nature of the algorithms implies that a central learner interacts with the private data owners one-on-one whenever they are available for communication without needing to aggregate query responses to construct gradients of the entire fitness function. Therefore, the algorithm efficiently scales to many data owners. We define the cost of privacy as the difference between the fitness of a privacy-preserving machine-learning model and the fitness of trained machine-learning model in the absence of privacy concerns. We prove that we can forecast the performance of the proposed privacy-preserving asynchronous algorithms. We demonstrate that the cost of privacy has an upper bound that is inversely proportional to the combined size of the training datasets squared and the sum of the privacy budgets squared. We validate the theoretical results with experiments on financial and medical datasets. The experiments illustrate that collaboration among more than 10 data owners with at least 10,000 records with privacy budgets greater than or equal to 1 results in a superior machine-learning model in comparison to a model trained in isolation on only one of the datasets, illustrating the value of collaboration and the cost of the privacy. The number of the collaborating datasets can be lowered if the privacy budget is higher.

Due to the availability of huge amounts of data and processing abilities, current artificial intelligence (AI) systems are effective at solving complex tasks. However, despite the success of AI in different areas, the problem of designing AI systems that can truly mimic human cognitive capabilities such as artificial general intelligence, remains largely open. Consequently, many emerging cross-device AI applications will require a transition from traditional centralized learning systems towards large-scale distributed AI systems that can collaboratively perform multiple complex learning tasks. In this paper, we propose a novel design philosophy called democratized learning (Dem-AI) whose goal is to build large-scale distributed learning systems that rely on the self-organization of distributed learning agents that are well-connected, but limited in learning capabilities. Correspondingly, inspired from the societal groups of humans, the specialized groups of learning agents in the proposed Dem-AI system are selforganized in a hierarchical structure to collectively perform learning tasks more efficiently. As such, the Dem-AI learning system can evolve and regulate itself based on the underlying duality of two processes that we call specialized and generalized processes. In this regard, we present a reference design as a guideline to realize future Dem-AI systems, inspired by various interdisciplinary fields. Accordingly, we introduce four underlying mechanisms in the design such as plasticity-stability transition mechanism, self-organizing hierarchical structuring, specialized learning, and generalization. Finally, we establish possible extensions and new challenges for the existing learning approaches to provide better scalable, flexible, and more powerful learning systems with the new setting of Dem-AI.

Deep multi-agent reinforcement learning (MARL) holds the promise of automating many real-world cooperative robotic manipulation and transportation tasks. Nevertheless, decentralised cooperative robotic control has received less attention from the deep reinforcement learning community, as compared to single-agent robotics and multi-agent games with discrete actions. To address this gap, this paper introduces Multi-Agent Mujoco, an easily extensible multi-agent benchmark suite for robotic control in continuous action spaces. The benchmark tasks are diverse and admit easily configurable partially observable settings. Inspired by the success of single-agent continuous value-based algorithms in robotic control, we also introduce COMIX, a novel extension to a common discrete action multi-agent $Q$-learning algorithm. We show that COMIX significantly outperforms state-of-the-art MADDPG on a partially observable variant of a popular particle environment and matches or surpasses it on Multi-Agent Mujoco. Thanks to this new benchmark suite and method, we can now pose an interesting question: what is the key to performance in such settings, the use of value-based methods instead of policy gradients, or the factorisation of the joint $Q$-function? To answer this question, we propose a second new method, FacMADDPG, which factors MADDPG's critic. Experimental results on Multi-Agent Mujoco suggest that factorisation is the key to performance.

A novel deep multi-agent reinforcement learning framework is proposed to identify and resolve conflicts among a variable number of aircraft in a high-density, stochastic, and dynamic sector in en route airspace. Currently the sector capacity is limited by human air traffic controller's cognitive limitation. In order to scale up to a high-density airspace, in this work we investigate the feasibility of a new concept (autonomous separation assurance) and a new approach (multi-agent reinforcement learning) to push the sector capacity above human cognitive limitation. We propose the concept of using distributed vehicle autonomy to ensure separation, instead of a centralized sector air traffic controller. Our proposed framework utilizes an actor-critic model, Proximal Policy Optimization (PPO) that we customize to incorporate an attention network. By using the attention network, we are able to encode the information from a variable number of intruder aircraft into a fixed length vector and allow the agents to learn which intruder aircraft's information is critical to achieve the optimal performance. This allows the agents to have access to variable aircraft information in the sector in a scalable, efficient approach to achieve high traffic throughput under uncertainty. The agents are trained using a centralized learning, decentralized execution scheme where one neural network is learned and shared by all agents in the environment. To validate the proposed framework, we designed three challenging case studies in the BlueSky air traffic control environment. Numerical results show the proposed framework significantly reduces the offline training time without sacrificing performance.

In this paper, we present new results on the fair and efficient allocation of indivisible goods to agents that have monotone, submodular, non-additive valuation functions over bundles. Despite their simple structure, these agent valuations are a natural model for several real-world domains. We show that, if such a valuation function has binary marginal gains, a socially optimal (i.e. utilitarian social welfare-maximizing) allocation that achieves envy-freeness up to one item (EF1) exists and is computationally tractable. We also prove that the Nash welfare-maximizing and the leximin allocations both exhibit this fairness-efficiency combination, by showing that they can be achieved by minimizing any symmetric strictly convex function over utilitarian optimal outcomes. To the best of our knowledge, this is the first valuation function class not subsumed by additive valuations for which it has been established that an allocation maximizing Nash welfare is EF1. Moreover, for a subclass of these valuation functions based on maximum (unweighted) bipartite matching, we show that a leximin allocation can be computed in polynomial time.

Interest in semi-autonomous systems (SAS) is growing rapidly as a paradigm to deploy autonomous systems in domains that require occasional reliance on humans. This paradigm allows service robots or autonomous vehicles to operate at varying levels of autonomy and offer safety in situations that require human judgment. We propose an introspective model of autonomy that is learned and updated online through experience and dictates the extent to which the agent can act autonomously in any given situation. We define a competence-aware system (CAS) that explicitly models its own proficiency at different levels of autonomy and the available human feedback. A CAS learns to adjust its level of autonomy based on experience to maximize overall efficiency, factoring in the cost of human assistance. We analyze the convergence properties of CAS and provide experimental results for robot delivery and autonomous driving domains that demonstrate the benefits of the approach.