This paper investigates posterior sampling algorithms for competitive reinforcement learning (RL) in the context of general function approximations. Focusing on zero-sum Markov games (MGs) under two critical settings, namely self-play and adversarial learning, we first propose the self-play and adversarial generalized eluder coefficient (GEC) as complexity measures for function approximation, capturing the exploration-exploitation trade-off in MGs. Based on self-play GEC, we propose a model-based self-play posterior sampling method to control both players to learn Nash equilibrium, which can successfully handle the partial observability of states. Furthermore, we identify a set of partially observable MG models fitting MG learning with the adversarial policies of the opponent. Incorporating the adversarial GEC, we propose a model-based posterior sampling method for learning adversarial MG with potential partial observability. We further provide low regret bounds for proposed algorithms that can scale sublinearly with the proposed GEC and the number of episodes $T$. To the best of our knowledge, we for the first time develop generic model-based posterior sampling algorithms for competitive RL that can be applied to a majority of tractable zero-sum MG classes in both fully observable and partially observable MGs with self-play and adversarial learning.

Recently, partial Bayesian neural networks (pBNNs), which only consider a subset of the parameters to be stochastic, were shown to perform competitively with full Bayesian neural networks. However, pBNNs are often multi-modal in the latent-variable space and thus challenging to approximate with parametric models. To address this problem, we propose an efficient sampling-based training strategy, wherein the training of a pBNN is formulated as simulating a Feynman--Kac model. We then describe variations of sequential Monte Carlo samplers that allow us to simultaneously estimate the parameters and the latent posterior distribution of this model at a tractable computational cost. We show on various synthetic and real-world datasets that our proposed training scheme outperforms the state of the art in terms of predictive performance.

Sequential latent variable models (SLVMs) are essential tools in statistics and machine learning, with applications ranging from healthcare to neuroscience. As their flexibility increases, analytic inference and model learning can become challenging, necessitating approximate methods. Here we introduce neural adaptive smoothing via twisting (NAS-X), a method that extends reweighted wake-sleep (RWS) to the sequential setting by using smoothing sequential Monte Carlo (SMC) to estimate intractable posterior expectations. Combining RWS and smoothing SMC allows NAS-X to provide low-bias and low-variance gradient estimates, and fit both discrete and continuous latent variable models. We illustrate the theoretical advantages of NAS-X over previous methods and explore these advantages empirically in a variety of tasks, including a challenging application to mechanistic models of neuronal dynamics. These experiments show that NAS-X substantially outperforms previous VI- and RWS-based methods in inference and model learning, achieving lower parameter error and tighter likelihood bounds.

Inverse optimal control can be used to characterize behavior in sequential decision-making tasks. Most existing work, however, is limited to fully observable or linear systems, or requires the action signals to be known. Here, we introduce a probabilistic approach to inverse optimal control for partially observable stochastic non-linear systems with unobserved action signals, which unifies previous approaches to inverse optimal control with maximum causal entropy formulations. Using an explicit model of the noise characteristics of the sensory and motor systems of the agent in conjunction with local linearization techniques, we derive an approximate likelihood function for the model parameters, which can be computed within a single forward pass. We present quantitative evaluations on stochastic and partially observable versions of two classic control tasks and two human behavioral tasks. Importantly, we show that our method can disentangle perceptual factors and behavioral costs despite the fact that epistemic and pragmatic actions are intertwined in sequential decision-making under uncertainty, such as in active sensing and active learning. The proposed method has broad applicability, ranging from imitation learning to sensorimotor neuroscience.

Learning from Demonstration (LfD) [1, 2] is a set of supervised learning methods where a teacher (often, but not always, a human) demonstrates a task, and a student (usually a robot) uses this information to learn to perform the same task. Some LfD methods cope with small morphological mismatches between the teacher and student [3, 4] (e.g., five-fingered hand to two-fingered gripper). However, they typically fail for a large mismatch (e.g., bimanual human demonstration to a robot arm with one gripper). The key difference is that to reproduce the transition from a demonstration state to the next, no single student action suffices - a sequence of actions may be needed. Supervised methods are appealing where demonstration-free methods [5] do not converge or underperform [6] and purely analytical approaches are computationally infeasible [7, 8]. In such settings, human demonstrations of complex tasks are often readily available e.g., it is straightforward for a human to show a robot how to fold a cloth. An LfD-based imitation learning approach is appealing in such settings provided we allow the human demonstrator to use their body in the way they find most convenient (e.g., using two hands to hang a cloth on a clothesline to dry). This requirement induces a potentially large morphology mismatch - we want to learn and execute complex tasks with deformable objects on a single manipulator robot using natural human demonstrations. We propose a framework, Morphological Adaptation in Imitation Learning (MAIL), to bridge this mismatch.

Reinforcement learning presents an attractive paradigm to reason about several distinct aspects of sequential decision making, such as specifying complex goals, planning future observations and actions, and critiquing their utilities. However, the combined integration of these capabilities poses competing algorithmic challenges in retaining maximal expressivity while allowing for flexibility in modeling choices for efficient learning and inference. We present Decision Stacks, a generative framework that decomposes goal-conditioned policy agents into 3 generative modules. These modules simulate the temporal evolution of observations, rewards, and actions via independent generative models that can be learned in parallel via teacher forcing. Our framework guarantees both expressivity and flexibility in designing individual modules to account for key factors such as architectural bias, optimization objective and dynamics, transferrability across domains, and inference speed. Our empirical results demonstrate the effectiveness of Decision Stacks for offline policy optimization for several MDP and POMDP environments, outperforming existing methods and enabling flexible generative decision making.

The dynamic pricing problem has been extensively studied under the \textbf{stream} model: A stream of customers arrives sequentially, each with an independently and identically distributed valuation. However, this formulation is not entirely reflective of the real world. In many scenarios, high-valuation customers tend to make purchases earlier and leave the market, leading to a \emph{shift} in the valuation distribution. Thus motivated, we consider a model where a \textbf{pool} of $n$ non-strategic unit-demand customers interact repeatedly with the seller. Each customer monitors the price intermittently according to an independent Poisson process and makes a purchase if the observed price is lower than her \emph{private} valuation, whereupon she leaves the market permanently. We present a minimax \emph{optimal} algorithm that efficiently computes a non-adaptive policy which guarantees a $1/k$ fraction of the optimal revenue, given any set of $k$ prices. Moreover, we present an adaptive \emph{learn-then-earn} policy based on a novel \emph{debiasing} approach, and prove an $\tilde O(kn^{3/4})$ regret bound. We further improve the bound to $\tilde O(k^{3/4} n^{3/4})$ using martingale concentration inequalities.

Given its substantial contribution of 40\% to global power consumption, the built environment has received increasing attention to serve as a source of flexibility to assist the modern power grid. In that respect, previous research mainly focused on energy management of individual buildings. In contrast, in this paper, we focus on aggregated control of a set of residential buildings, to provide grid supporting services, that eventually should include ancillary services. In particular, we present a real-life pilot study that studies the effectiveness of reinforcement-learning (RL) in coordinating the power consumption of 8 residential buildings to jointly track a target power signal. Our RL approach relies solely on observed data from individual households and does not require any explicit building models or simulators, making it practical to implement and easy to scale. We show the feasibility of our proposed RL-based coordination strategy in a real-world setting. In a 4-week case study, we demonstrate a hierarchical control system, relying on an RL-based ranking system to select which households to activate flex assets from, and a real-time PI control-based power dispatch mechanism to control the selected assets. Our results demonstrate satisfactory power tracking, and the effectiveness of the RL-based ranks which are learnt in a purely data-driven manner.

Reinforcement learning (RL) is a powerful tool for finding optimal policies in sequential decision processes. However, deep RL methods suffer from two weaknesses: collecting the amount of agent experience required for practical RL problems is prohibitively expensive, and the learned policies exhibit poor generalization on tasks outside of the training distribution. To mitigate these issues, we introduce automaton distillation, a form of neuro-symbolic transfer learning in which Q-value estimates from a teacher are distilled into a low-dimensional representation in the form of an automaton. We then propose two methods for generating Q-value estimates: static transfer, which reasons over an abstract Markov Decision Process constructed based on prior knowledge, and dynamic transfer, where symbolic information is extracted from a teacher Deep Q-Network (DQN). The resulting Q-value estimates from either method are used to bootstrap learning in the target environment via a modified DQN loss function. We list several failure modes of existing automaton-based transfer methods and demonstrate that both static and dynamic automaton distillation decrease the time required to find optimal policies for various decision tasks.

The space environment around the Earth is becoming increasingly populated by both active spacecraft and space debris. To avoid potential collision events, significant improvements in Space Situational Awareness (SSA) activities and Collision Avoidance (CA) technologies are allowing the tracking and maneuvering of spacecraft with increasing accuracy and reliability. However, these procedures still largely involve a high level of human intervention to make the necessary decisions. For an increasingly complex space environment, this decision-making strategy is not likely to be sustainable. Therefore, it is important to successfully introduce higher levels of automation for key Space Traffic Management (STM) processes to ensure the level of reliability needed for navigating a large number of spacecraft. These processes range from collision risk detection to the identification of the appropriate action to take and the execution of avoidance maneuvers. This work proposes an implementation of autonomous CA decision-making capabilities on spacecraft based on Reinforcement Learning (RL) techniques. A novel methodology based on a Partially Observable Markov Decision Process (POMDP) framework is developed to train the Artificial Intelligence (AI) system on board the spacecraft, considering epistemic and aleatory uncertainties. The proposed framework considers imperfect monitoring information about the status of the debris in orbit and allows the AI system to effectively learn stochastic policies to perform accurate Collision Avoidance Maneuvers (CAMs). The objective is to successfully delegate the decision-making process for autonomously implementing a CAM to the spacecraft without human intervention. This approach would allow for a faster response in the decision-making process and for highly decentralized operations.