Vital importance has necessity to be attached to cooperation in multi-agent environments, as a result of which some reinforcement learning algorithms combined with graph neural networks have been proposed to understand the mutual interplay between agents. However, highly complicated and dynamic multi-agent environments require more ingenious graph neural networks, which can comprehensively represent not only the graph topology structure but also evolution process of the structure due to agents emerging, disappearing and moving. To tackle these difficulties, we propose Gumbel Sinkhorn graph attention reinforcement learning, where a graph attention network highly represents the underlying graph topology structure of the multi-agent environment, and can adapt to the dynamic topology structure of graph better with the help of Gumbel Sinkhorn network by learning latent permutations. Empirically, simulation results show how our proposed graph reinforcement learning methodology outperforms existing methods in the PettingZoo multi-agent environment by learning latent permutations.

Contextual multi-armed bandits are classical models in reinforcement learning for sequential decision-making associated with individual information. A widely-used policy for bandits is Thompson Sampling, where samples from a data-driven probabilistic belief about unknown parameters are used to select the control actions. For this computationally fast algorithm, performance analyses are available under full context-observations. However, little is known for problems that contexts are not fully observed. We propose a Thompson Sampling algorithm for partially observable contextual multi-armed bandits, and establish theoretical performance guarantees. Technically, we show that the regret of the presented policy scales logarithmically with time and the number of arms, and linearly with the dimension. Further, we establish rates of learning unknown parameters, and provide illustrative numerical analyses.

Goal-conditioned reinforcement learning (RL) can solve tasks in a wide range of domains, including navigation and manipulation, but learning to reach distant goals remains a central challenge to the field. Learning to reach such goals is particularly hard without any offline data, expert demonstrations, and reward shaping. In this paper, we propose an algorithm to solve the distant goal-reaching task by using search at training time to automatically generate a curriculum of intermediate states. Our algorithm, Classifier-Planning (C-Planning), frames the learning of the goal-conditioned policies as expectation maximization: the E-step corresponds to planning an optimal sequence of waypoints using graph search, while the M-step aims to learn a goal-conditioned policy to reach those waypoints. Unlike prior methods that combine goal-conditioned RL with graph search, ours performs search only during training and not testing, significantly decreasing the compute costs of deploying the learned policy. Empirically, we demonstrate that our method is more sample efficient than prior methods. Moreover, it is able to solve very long horizons manipulation and navigation tasks, tasks that prior goal-conditioned methods and methods based on graph search fail to solve.

The demand for explainable machine learning (ML) models has been growing rapidly in recent years. Amongst the methods proposed to associate ML model predictions with human-understandable rationale, counterfactual explanations are one of the most popular. They consist of post-hoc rules derived from counterfactual examples (CFs), i.e., modified versions of input samples that result in alternative output responses from the predictive model to be explained. However, existing CF generation strategies either exploit the internals of specific models (e.g., random forests or neural networks), or depend on each sample's neighborhood, which makes them hard to be generalized for more complex models and inefficient for larger datasets. In this work, we aim to overcome these limitations and introduce a model-agnostic algorithm to generate optimal counterfactual explanations. Specifically, we formulate the problem of crafting CFs as a sequential decision-making task and then find the optimal CFs via deep reinforcement learning (DRL) with discrete-continuous hybrid action space. Differently from other techniques, our method is easily applied to any black-box model, as this resembles the environment that the DRL agent interacts with. In addition, we develop an algorithm to extract explainable decision rules from the DRL agent's policy, so as to make the process of generating CFs itself transparent. Extensive experiments conducted on several datasets have shown that our method outperforms existing CF generation baselines.

Microservice-based architectures enable different aspects of web applications to be created and updated independently, even after deployment. Associated technologies such as service mesh provide application-level fault resilience through attribute configurations that govern the behavior of request-response service -- and the interactions among them -- in the presence of failures. While this provides tremendous flexibility, the configured values of these attributes -- and the relationships among them -- can significantly affect the performance and fault resilience of the overall application. Furthermore, it is impossible to determine the best and worst combinations of attribute values with respect to fault resiliency via testing, due to the complexities of the underlying distributed system and the many possible attribute value combinations. In this paper, we present a model-based reinforcement learning workflow towards service mesh fault resiliency. Our approach enables the prediction of the most significant fault resilience behaviors at a web application-level, scratching from single service to aggregated multi-service management with efficient agent collaborations.

Heated debates continue over the best autonomous driving framework. The classic modular pipeline is widely adopted in the industry owing to its great interpretability and stability, whereas the end-to-end paradigm has demonstrated considerable simplicity and learnability along with the rise of deep learning. We introduce a new modularized end-to-end reinforcement learning framework (ModEL) for autonomous driving, which combines the merits of both previous approaches. The autonomous driving stack of ModEL is decomposed into perception, planning, and control module, leveraging scene understanding, end-to-end reinforcement learning, and PID control respectively. Furthermore, we build a fully functional autonomous vehicle to deploy this framework. Through extensive simulation and real-world experiments, our framework has shown great generalizability to various complicated scenarios and outperforms the competing baselines.

A key component of understanding hand-object interactions is the ability to identify the active object -- the object that is being manipulated by the human hand -- despite the occlusion induced by hand-object interactions. Based on the observation that hand appearance is a strong indicator of the location and size of the active object, we set up our active object detection method as a sequential decision-making process that is conditioned on the location and appearance of the hands. The key innovation of our approach is the design of the active object detection policy that uses an internal representation called the Relational Box Field, which allows for every pixel to regress an improved location of an active object bounding box, essentially giving every pixel the ability to vote for a better bounding box location. The policy is trained using a hybrid imitation learning and reinforcement learning approach, and at test time, the policy is used repeatedly to refine the bounding box location of the active object. We perform experiments on two large-scale datasets: 100DOH and MECCANO, improving AP50 performance by 8% and 30%, respectively, over the state of the art.

Undesired bias afflicts both human and algorithmic decision making, and may be especially prevalent when information processing trade-offs incentivize the use of heuristics. One primary example is \textit{statistical discrimination} -- selecting social partners based not on their underlying attributes, but on readily perceptible characteristics that covary with their suitability for the task at hand. We present a theoretical model to examine how information processing influences statistical discrimination and test its predictions using multi-agent reinforcement learning with various agent architectures in a partner choice-based social dilemma. As predicted, statistical discrimination emerges in agent policies as a function of both the bias in the training population and of agent architecture. All agents showed substantial statistical discrimination, defaulting to using the readily available correlates instead of the outcome relevant features. We show that less discrimination emerges with agents that use recurrent neural networks, and when their training environment has less bias. However, all agent algorithms we tried still exhibited substantial bias after learning in biased training populations.

Reinforcement learning (RL) experiments have notoriously high variance, and minor details can have disproportionately large effects on measured outcomes. This is problematic for creating reproducible research and also serves as an obstacle for real-world applications, where safety and predictability are paramount. In this paper, we investigate causes for this perceived instability. To allow for an in-depth analysis, we focus on a specifically popular setup with high variance -- continuous control from pixels with an actor-critic agent. In this setting, we demonstrate that variance mostly arises early in training as a result of poor "outlier" runs, but that weight initialization and initial exploration are not to blame. We show that one cause for early variance is numerical instability which leads to saturating nonlinearities. We investigate several fixes to this issue and find that one particular method is surprisingly effective and simple -- normalizing penultimate features. Addressing the learning instability allows for larger learning rates, and significantly decreases the variance of outcomes. This demonstrates that the perceived variance in RL is not necessarily inherent to the problem definition and may be addressed through simple architectural modifications.

We present Logical Optimal Actions (LOA), an action decision architecture of reinforcement learning applications with a neuro-symbolic framework which is a combination of neural network and symbolic knowledge acquisition approach for natural language interaction games. The demonstration for LOA experiments consists of a web-based interactive platform for text-based games and visualization for acquired knowledge for improving interpretability for trained rules. This demonstration also provides a comparison module with other neuro-symbolic approaches as well as non-symbolic state-of-the-art agent models on the same text-based games. Our LOA also provides open-sourced implementation in Python for the reinforcement learning environment to facilitate an experiment for studying neuro-symbolic agents. Code: https://github.com/ibm/loa