Probabilistic inference in high-dimensional state-space models is computationally challenging. For many spatiotemporal systems, however, prior knowledge about the dependency structure of state variables is available. We leverage this structure to develop a computationally efficient approach to state estimation and learning in graph-structured state-space models with (partially) unknown dynamics and limited historical data. Building on recent methods that combine ideas from deep learning with principled inference in Gaussian Markov random fields (GMRF), we reformulate graph-structured state-space models as Deep GMRFs defined by simple spatial and temporal graph layers. This results in a flexible spatiotemporal prior that can be learned efficiently from a single time sequence via variational inference. Under linear Gaussian assumptions, we retain a closed-form posterior, which can be sampled efficiently using the conjugate gradient method, scaling favorably compared to classical Kalman filter based approaches.

In this paper, we introduce Randomized Q-learning (RandQL), a novel randomized model-free algorithm for regret minimization in episodic Markov Decision Processes (MDPs). To the best of our knowledge, RandQL is the first tractable model-free posterior sampling-based algorithm. We analyze the performance of RandQL in both tabular and non-tabular metric space settings. In tabular MDPs, RandQL achieves a regret bound of order $\widetilde{\mathcal{O}}(\sqrt{H^{5}SAT})$, where $H$ is the planning horizon, $S$ is the number of states, $A$ is the number of actions, and $T$ is the number of episodes. For a metric state-action space, RandQL enjoys a regret bound of order $\widetilde{\mathcal{O}}(H^{5/2} T^{(d_z+1)/(d_z+2)})$, where $d_z$ denotes the zooming dimension. Notably, RandQL achieves optimistic exploration without using bonuses, relying instead on a novel idea of learning rate randomization. Our empirical study shows that RandQL outperforms existing approaches on baseline exploration environments.

Restricted Boltzmann machines are energy models made of a visible and a hidden layer. We identify an effective energy function describing the zero-temperature landscape on the visible units and depending only on the tail behaviour of the hidden layer prior distribution. Studying the location of the local minima of such an energy function, we show that the ability of a restricted Boltzmann machine to reconstruct a random pattern depends indeed only on the tail of the hidden prior distribution. We find that hidden priors with strictly super-Gaussian tails give only a logarithmic loss in pattern retrieval, while an efficient retrieval is much harder with hidden units with strictly sub-Gaussian tails; if the hidden prior has Gaussian tails, the retrieval capability is determined by the number of hidden units (as in the Hopfield model).

We show that large language models (LLMs) can be adapted to be generalizable policies for embodied visual tasks. Our approach, called Large LAnguage model Reinforcement Learning Policy (LLaRP), adapts a pre-trained frozen LLM to take as input text instructions and visual egocentric observations and output actions directly in the environment. Using reinforcement learning, we train LLaRP to see and act solely through environmental interactions. We show that LLaRP is robust to complex paraphrasings of task instructions and can generalize to new tasks that require novel optimal behavior. In particular, on 1,000 unseen tasks it achieves 42% success rate, 1.7x the success rate of other common learned baselines or zero-shot applications of LLMs. Finally, to aid the community in studying language conditioned, massively multi-task, embodied AI problems we release a novel benchmark, Language Rearrangement, consisting of 150,000 training and 1,000 testing tasks for language-conditioned rearrangement. Video examples of LLaRP in unseen Language Rearrangement instructions are at https://llm-rl.github.io.

Automatic sign language recognition (SLR) is an important topic within the areas of human-computer interaction and machine learning. On the one hand, it poses a complex challenge that requires the intervention of various knowledge areas, such as video processing, image processing, intelligent systems and linguistics. On the other hand, robust recognition of sign language could assist in the translation process and the integration of hearing-impaired people, as well as the teaching of sign language for the hearing population. SLR systems usually employ Hidden Markov Models, Dynamic Time Warping or similar models to recognize signs. Such techniques exploit the sequential ordering of frames to reduce the number of hypothesis. This paper presents a general probabilistic model for sign classification that combines sub-classifiers based on different types of features such as position, movement and handshape. The model employs a bag-of-words approach in all classification steps, to explore the hypothesis that ordering is not essential for recognition. The proposed model achieved an accuracy rate of 97% on an Argentinian Sign Language dataset containing 64 classes of signs and 3200 samples, providing some evidence that indeed recognition without ordering is possible.

In this paper, we investigate the probabilistic variants of the strategy logics ATL and ATL* under imperfect information. Specifically, we present novel decidability and complexity results when the model transitions are stochastic and agents play uniform strategies. That is, the semantics of the logics are based on multi-agent, stochastic transition systems with imperfect information, which combine two sources of uncertainty, namely, the partial observability agents have on the environment, and the likelihood of transitions to occur from a system state. Since the model checking problem is undecidable in general in this setting, we restrict our attention to agents with memoryless (positional) strategies. The resulting setting captures the situation in which agents have qualitative uncertainty of the local state and quantitative uncertainty about the occurrence of future events. We illustrate the usefulness of this setting with meaningful examples.

In real-world reinforcement learning (RL) systems, various forms of {\it impaired observability} can complicate matters. These situations arise when an agent is unable to observe the most recent state of the system due to latency or lossy channels, yet the agent must still make real-time decisions. This paper introduces a theoretical investigation into efficient RL in control systems where agents must act with delayed and missing state observations. We present algorithms and establish near-optimal regret upper and lower bounds, of the form $\tilde{\mathcal{O}}(\sqrt{{\rm poly}(H) SAK})$, for RL in the delayed and missing observation settings. Here $S$ and $A$ are the sizes of state and action spaces, $H$ is the time horizon and $K$ is the number of episodes. Despite impaired observability posing significant challenges to the policy class and planning, our results demonstrate that learning remains efficient, with the regret bound optimally depending on the state-action size of the original system. Additionally, we provide a characterization of the performance of the optimal policy under impaired observability, comparing it to the optimal value obtained with full observability. Numerical results are provided to support our theory.

Offline reinforcement learning (RL) aims to learn optimal policies from offline datasets, where the parameterization of policies is crucial but often overlooked. Recently, Diffsuion-QL significantly boosts the performance of offline RL by representing a policy with a diffusion model, whose success relies on a parametrized Markov Chain with hundreds of steps for sampling. However, Diffusion-QL suffers from two critical limitations. 1) It is computationally inefficient to forward and backward through the whole Markov chain during training. 2) It is incompatible with maximum likelihood-based RL algorithms (e.g., policy gradient methods) as the likelihood of diffusion models is intractable. Therefore, we propose efficient diffusion policy (EDP) to overcome these two challenges. EDP approximately constructs actions from corrupted ones at training to avoid running the sampling chain. We conduct extensive experiments on the D4RL benchmark. The results show that EDP can reduce the diffusion policy training time from 5 days to 5 hours on gym-locomotion tasks. Moreover, we show that EDP is compatible with various offline RL algorithms (TD3, CRR, and IQL) and achieves new state-of-the-art on D4RL by large margins over previous methods. Our code is available at https://github.com/sail-sg/edp.

Unsupervised pre-training methods utilizing large and diverse datasets have achieved tremendous success across a range of domains. Recent work has investigated such unsupervised pre-training methods for model-based reinforcement learning (MBRL) but is limited to domain-specific or simulated data. In this paper, we study the problem of pre-training world models with abundant in-the-wild videos for efficient learning of downstream visual control tasks. However, in-the-wild videos are complicated with various contextual factors, such as intricate backgrounds and textured appearance, which precludes a world model from extracting shared world knowledge to generalize better. To tackle this issue, we introduce Contextualized World Models (ContextWM) that explicitly separate context and dynamics modeling to overcome the complexity and diversity of in-the-wild videos and facilitate knowledge transfer between distinct scenes. Specifically, a contextualized extension of the latent dynamics model is elaborately realized by incorporating a context encoder to retain contextual information and empower the image decoder, which encourages the latent dynamics model to concentrate on essential temporal variations. Our experiments show that in-the-wild video pre-training equipped with ContextWM can significantly improve the sample efficiency of MBRL in various domains, including robotic manipulation, locomotion, and autonomous driving. Code is available at this repository: https://github.com/thuml/ContextWM.

Partially Observable Markov Decision Processes (POMDPs) are used to model environments where the full state cannot be perceived by an agent. As such the agent needs to reason taking into account the past observations and actions. However, simply remembering the full history is generally intractable due to the exponential growth in the history space. Maintaining a probability distribution that models the belief over what the true state is can be used as a sufficient statistic of the history, but its computation requires access to the model of the environment and is often intractable. While SOTA algorithms use Recurrent Neural Networks to compress the observation-action history aiming to learn a sufficient statistic, they lack guarantees of success and can lead to sub-optimal policies. To overcome this, we propose the Wasserstein Belief Updater, an RL algorithm that learns a latent model of the POMDP and an approximation of the belief update. Our approach comes with theoretical guarantees on the quality of our approximation ensuring that our outputted beliefs allow for learning the optimal value function.