Whittle index policy is a heuristic to the intractable restless multi-armed bandits (RMAB) problem. Although it is provably asymptotically optimal, finding Whittle indices remains difficult. In this paper, we present Neural-Q-Whittle, a Whittle index based Q-learning algorithm for RMAB with neural network function approximation, which is an example of nonlinear two-timescale stochastic approximation with Q-function values updated on a faster timescale and Whittle indices on a slower timescale. Despite the empirical success of deep Q-learning, the non-asymptotic convergence rate of Neural-Q-Whittle, which couples neural networks with two-timescale Q-learning largely remains unclear. This paper provides a finite-time analysis of Neural-Q-Whittle, where data are generated from a Markov chain, and Q-function is approximated by a ReLU neural network. Our analysis leverages a Lyapunov drift approach to capture the evolution of two coupled parameters, and the nonlinearity in value function approximation further requires us to characterize the approximation error. Combing these provide Neural-Q-Whittle with $\mathcal{O}(1/k^{2/3})$ convergence rate, where $k$ is the number of iterations.

What can be learned about causality and experimentation from passive data? This question is salient given recent successes of passively-trained language models in interactive domains such as tool use. Passive learning is inherently limited. However, we show that purely passive learning can in fact allow an agent to learn generalizable strategies for determining and using causal structures, as long as the agent can intervene at test time. We formally illustrate that learning a strategy of first experimenting, then seeking goals, can allow generalization from passive learning in principle. We then show empirically that agents trained via imitation on expert data can indeed generalize at test time to infer and use causal links which are never present in the training data; these agents can also generalize experimentation strategies to novel variable sets never observed in training. We then show that strategies for causal intervention and exploitation can be generalized from passive data even in a more complex environment with high-dimensional observations, with the support of natural language explanations. Explanations can even allow passive learners to generalize out-of-distribution from perfectly-confounded training data. Finally, we show that language models, trained only on passive next-word prediction, can generalize causal intervention strategies from a few-shot prompt containing examples of experimentation, together with explanations and reasoning. These results highlight the surprising power of passive learning of active causal strategies, and may help to understand the behaviors and capabilities of language models.

Constraint programming is known for being an efficient approach for solving combinatorial problems. Important design choices in a solver are the branching heuristics, which are designed to lead the search to the best solutions in a minimum amount of time. However, developing these heuristics is a time-consuming process that requires problem-specific expertise. This observation has motivated many efforts to use machine learning to automatically learn efficient heuristics without expert intervention. To the best of our knowledge, it is still an open research question. Although several generic variable-selection heuristics are available in the literature, the options for a generic value-selection heuristic are more scarce. In this paper, we propose to tackle this issue by introducing a generic learning procedure that can be used to obtain a value-selection heuristic inside a constraint programming solver. This has been achieved thanks to the combination of a deep Q-learning algorithm, a tailored reward signal, and a heterogeneous graph neural network architecture. Experiments on graph coloring, maximum independent set, and maximum cut problems show that our framework is able to find better solutions close to optimality without requiring a large amounts of backtracks while being generic.

Perceptive deep reinforcement learning (DRL) has lead to many recent breakthroughs for complex AI systems leveraging image-based input data. Applications of these results range from super-human level video game agents to dexterous, physically intelligent robots. However, training these perceptive DRL-enabled systems remains incredibly compute and memory intensive, often requiring huge training datasets and large experience replay buffers. This poses a challenge for the next generation of field robots that will need to be able to learn on the edge in order to adapt to their environments. In this paper, we begin to address this issue through differentially encoded observation spaces. By reinterpreting stored image-based observations as a video, we leverage lossless differential video encoding schemes to compress the replay buffer without impacting training performance. We evaluate our approach with three state-of-the-art DRL algorithms and find that differential image encoding reduces the memory footprint by as much as 14.2x and 16.7x across tasks from the Atari 2600 benchmark and the DeepMind Control Suite (DMC) respectively. These savings also enable large-scale perceptive DRL that previously required paging between flash and RAM to be run entirely in RAM, improving the latency of DMC tasks by as much as 32%.

We study the representation complexity of model-based and model-free reinforcement learning (RL) in the context of circuit complexity. We prove theoretically that there exists a broad class of MDPs such that their underlying transition and reward functions can be represented by constant depth circuits with polynomial size, while the optimal $Q$-function suffers an exponential circuit complexity in constant-depth circuits. By drawing attention to the approximation errors and building connections to complexity theory, our theory provides unique insights into why model-based algorithms usually enjoy better sample complexity than model-free algorithms from a novel representation complexity perspective: in some cases, the ground-truth rule (model) of the environment is simple to represent, while other quantities, such as $Q$-function, appear complex. We empirically corroborate our theory by comparing the approximation error of the transition kernel, reward function, and optimal $Q$-function in various Mujoco environments, which demonstrates that the approximation errors of the transition kernel and reward function are consistently lower than those of the optimal $Q$-function. To the best of our knowledge, this work is the first to study the circuit complexity of RL, which also provides a rigorous framework for future research.

Imitation Learning from Observation (ILfO) is a setting in which a learner tries to imitate the behavior of an expert, using only observational data and without the direct guidance of demonstrated actions. In this paper, we re-examine the use of optimal transport for IL, in which a reward is generated based on the Wasserstein distance between the state trajectories of the learner and expert. We show that existing methods can be simplified to generate a reward function without requiring learned models or adversarial learning. Unlike many other state-of-the-art methods, our approach can be integrated with any RL algorithm, and is amenable to ILfO. We demonstrate the effectiveness of this simple approach on a variety of continuous control tasks and find that it surpasses the state of the art in the IlfO setting, achieving expert-level performance across a range of evaluation domains even when observing only a single expert trajectory without actions.

We theoretically explore the relationship between sample-efficiency and adaptivity in reinforcement learning. An algorithm is sample-efficient if it uses a number of queries $n$ to the environment that is polynomial in the dimension $d$ of the problem. Adaptivity refers to the frequency at which queries are sent and feedback is processed to update the querying strategy. To investigate this interplay, we employ a learning framework that allows sending queries in $K$ batches, with feedback being processed and queries updated after each batch. This model encompasses the whole adaptivity spectrum, ranging from non-adaptive 'offline' ($K=1$) to fully adaptive ($K=n$) scenarios, and regimes in between. For the problems of policy evaluation and best-policy identification under $d$-dimensional linear function approximation, we establish $\Omega(\log \log d)$ lower bounds on the number of batches $K$ required for sample-efficient algorithms with $n = O(poly(d))$ queries. Our results show that just having adaptivity ($K>1$) does not necessarily guarantee sample-efficiency. Notably, the adaptivity-boundary for sample-efficiency is not between offline reinforcement learning ($K=1$), where sample-efficiency was known to not be possible, and adaptive settings. Instead, the boundary lies between different regimes of adaptivity and depends on the problem dimension.

The Quadratic Assignment Problem (QAP) is an NP-hard problem which has proven particularly challenging to solve: unlike other combinatorial problems like the traveling salesman problem (TSP), which can be solved to optimality for instances with hundreds or even thousands of locations using advanced integer programming techniques, no methods are known to exactly solve QAP instances of size greater than 30. Solving the QAP is nevertheless important because of its many critical applications, such as electronic wiring design and facility layout selection. We propose a method to solve the original Koopmans-Beckman formulation of the QAP using deep reinforcement learning. Our approach relies on a novel double pointer network, which alternates between selecting a location in which to place the next facility and a facility to place in the previous location. We train our model using A2C on a large dataset of synthetic instances, producing solutions with no instance-specific retraining necessary. Out of sample, our solutions are on average within 7.5% of a high-quality local search baseline, and even outperform it on 1.2% of instances.

Primary sclerosing cholangitis (PSC) is a rare disease wherein altered bile acid metabolism contributes to sustained liver injury. This paper introduces REMEDI, a framework that captures bile acid dynamics and the body's adaptive response during PSC progression that can assist in exploring treatments. REMEDI merges a differential equation (DE)-based mechanistic model that describes bile acid metabolism with reinforcement learning (RL) to emulate the body's adaptations to PSC continuously. An objective of adaptation is to maintain homeostasis by regulating enzymes involved in bile acid metabolism. These enzymes correspond to the parameters of the DEs. REMEDI leverages RL to approximate adaptations in PSC, treating homeostasis as a reward signal and the adjustment of the DE parameters as the corresponding actions. On real-world data, REMEDI generated bile acid dynamics and parameter adjustments consistent with published findings. Also, our results support discussions in the literature that early administration of drugs that suppress bile acid synthesis may be effective in PSC treatment.

Classical pixel-based Visual Servoing (VS) approaches offer high accuracy but suffer from a limited convergence area due to optimization nonlinearity. Modern deep learning-based VS methods overcome traditional vision issues but lack scalability, requiring training on limited scenes. This paper proposes a hybrid VS strategy utilizing Deep Reinforcement Learning (DRL) and optimal control to enhance both convergence area and scalability. The DRL component of our approach separately handles representation and policy learning to enhance scalability, generalizability, learning efficiency and ease domain adaptation. Moreover, the optimal control part ensures high end-point accuracy. Our method showcases remarkable achievements in terms of high convergence rates and minimal end-positioning errors using a 7-DOF manipulator. Importantly, it exhibits scalability across more than 1000 distinct scenes. Furthermore, we demonstrate its capacity for generalization to previously unseen datasets. Lastly, we illustrate the real-world applicability of our approach, highlighting its adaptability through single-shot domain transfer learning in environments with noise and occlusions. Real-robot experiments can be found at \url{https://sites.google.com/view/vsls}.