Many real-world problems are characterized by a large number of observations, confounding and spurious correlations, partially observable states, and distal, dynamic rewards with hierarchical reward structures. Such conditions make it hard for both animal and machines to learn complex skills. The learning process requires discovering what is important and what can be ignored, how the reward function is structured, and how to reuse knowledge across different tasks that share common properties. For these reasons, the application of standard reinforcement learning (RL) algorithms (Sutton and Barto, 2018) to solve structured problems is often not effective. Limitations of current RL algorithms include the problem of exploration with sparse rewards (Pathak et al., 2017), dealing with partially observable Markov decision problems (POMDP) (Ladosz et al., 2021), coping with large amounts of confounding stimuli (Thrun, 2000; Kim et al., 2019), and reusing skills for efficiently learning multiple task in a lifelong learning setting (Mendez and Eaton, 2020). Standard reinforcement learning algorithms are best suited when the problem can be formulated as a single-task problem in observable Markov decision problem (MDP). Under these assumptions, with complete observability and with static and frequent rewards, deep reinforcement learning (DRL) (Mnih et al., 2015; Li, 2017) has gained popularity due to the ability to learn an approximated Q-value function directly from raw pixel data in the Atari 2600 platform. This and similar algorithms stack multiple frames to derive states of an MDP, and use a basic ɛ-greedy exploration policy. In more complex cases with partial observability and sparse rewards, extensions have been proposed to include more advanced exploration techniques (Ladosz et al., 2022), e.g.

We introduce CriticSMC, a new algorithm for planning as inference built from a composition of sequential Monte Carlo with learned Soft-Q function heuristic factors. These heuristic factors, obtained from parametric approximations of the marginal likelihood ahead, more effectively guide SMC towards the desired target distribution, which is particularly helpful for planning in environments with hard constraints placed sparsely in time. Compared with previous work, we modify the placement of such heuristic factors, which allows us to cheaply propose and evaluate large numbers of putative action particles, greatly increasing inference and planning efficiency. CriticSMC is compatible with informative priors, whose density function need not be known, and can be used as a model-free control algorithm. Our experiments on collision avoidance in a high-dimensional simulated driving task show that CriticSMC significantly reduces collision rates at a low computational cost while maintaining realism and diversity of driving behaviors across vehicles and environment scenarios.

The standard assumption in reinforcement learning (RL) is that agents observe feedback for their actions immediately. However, in practice feedback is often observed in delay. This paper studies online learning in episodic Markov decision process (MDP) with unknown transitions, adversarially changing costs, and unrestricted delayed bandit feedback. More precisely, the feedback for the agent in episode $k$ is revealed only in the end of episode $k + d^k$, where the delay $d^k$ can be changing over episodes and chosen by an oblivious adversary. We present the first algorithms that achieve near-optimal $\sqrt{K + D}$ regret, where $K$ is the number of episodes and $D = \sum_{k=1}^K d^k$ is the total delay, significantly improving upon the best known regret bound of $(K + D)^{2/3}$.

The Cauchy-Schwarz (CS) divergence was developed by Pr\'{i}ncipe et al. in 2000. In this paper, we extend the classic CS divergence to quantify the closeness between two conditional distributions and show that the developed conditional CS divergence can be simply estimated by a kernel density estimator from given samples. We illustrate the advantages (e.g., the rigorous faithfulness guarantee, the lower computational complexity, the higher statistical power, and the much more flexibility in a wide range of applications) of our conditional CS divergence over previous proposals, such as the conditional KL divergence and the conditional maximum mean discrepancy. We also demonstrate the compelling performance of conditional CS divergence in two machine learning tasks related to time series data and sequential inference, namely the time series clustering and the uncertainty-guided exploration for sequential decision making.

In this paper, we investigate Nash-regret minimization in congestion games, a class of games with benign theoretical structure and broad real-world applications. We first propose a centralized algorithm based on the optimism in the face of uncertainty principle for congestion games with (semi-)bandit feedback, and obtain finite-sample guarantees. Then we propose a decentralized algorithm via a novel combination of the Frank-Wolfe method and G-optimal design. By exploiting the structure of the congestion game, we show the sample complexity of both algorithms depends only polynomially on the number of players and the number of facilities, but not the size of the action set, which can be exponentially large in terms of the number of facilities. We further define a new problem class, Markov congestion games, which allows us to model the non-stationarity in congestion games. We propose a centralized algorithm for Markov congestion games, whose sample complexity again has only polynomial dependence on all relevant problem parameters, but not the size of the action set.

Goal-conditioned reinforcement learning (GCRL) has a wide range of potential real-world applications, including manipulation and navigation problems in robotics. Especially in such robotics tasks, sample efficiency is of the utmost importance for GCRL since, by default, the agent is only rewarded when it reaches its goal. While several methods have been proposed to improve the sample efficiency of GCRL, one relatively under-studied approach is the design of neural architectures to support sample efficiency. In this work, we introduce a novel neural architecture for GCRL that achieves significantly better sample efficiency than the commonly-used monolithic network architecture. The key insight is that the optimal action-value function Q^*(s, a, g) must satisfy the triangle inequality in a specific sense. Furthermore, we introduce the metric residual network (MRN) that deliberately decomposes the action-value function Q(s,a,g) into the negated summation of a metric plus a residual asymmetric component. MRN provably approximates any optimal action-value function Q^*(s,a,g), thus making it a fitting neural architecture for GCRL. We conduct comprehensive experiments across 12 standard benchmark environments in GCRL. The empirical results demonstrate that MRN uniformly outperforms other state-of-the-art GCRL neural architectures in terms of sample efficiency.

We evaluate the state estimation performance of the RL-ROE for systems governed by the Burgers equation and Navier-Stokes equations. For each system, we first compute various solution trajectories corresponding to different physical parameter values, which we use to construct the ROM and train the RL-ROE following the procedure outlined in Section 2.4. The trained RL-ROE is finally deployed online and compared against a time-dependent Kalman filter constructed from the same ROM, which we refer to as KF-ROE. The KF-ROE is given by equations (3a) and (4), with the calculation of the time-varying Kalman gain detailed in Appendix C of the supplementary materials. Before proceeding to the results, we discuss our choice of baseline. The ensemble Kalman filter and 4D-Var are two estimation techniques for high-dimensional systems such as those governed by PDEs (Lorenc, 2003). Although they are commonly employed for data assimilation in numerical weather prediction, they require large computational resources since they involve repeated solutions of the high-dimensional dynamics (1). Thus, they are not applicable in the context of embedded control systems, whose limited resources call for an inexpensive model such as the ROM (2). Since the ROM that we consider has linear dynamics, extensions of the Kalman filter for nonlinear dynamics such as the extended or unscented Kalman filters (Wan & Van Der Merwe, 2000; Julier & Uhlmann, 2004) are not relevant, and the vanilla Kalman filter remains the best choice of baseline.

Extracting summaries from long documents can be regarded as sentence classification using the structural information of the documents. How to use such structural information to summarize a document is challenging. In this paper, we propose GoSum, a novel graph and reinforcement learning based extractive model for long-paper summarization. In particular, GoSum encodes sentence states in reinforcement learning by building a heterogeneous graph for each input document at different discourse levels. An edge in the graph reflects the discourse hierarchy of a document for restraining the semantic drifts across section boundaries. We evaluate GoSum on two datasets of scientific articles summarization: PubMed and arXiv. The experimental results have demonstrated that GoSum achieve state-of-the-art results compared with strong baselines of both extractive and abstractive models. The ablation studies further validate that the performance of our GoSum benefits from the use of discourse information.

In Model-based Reinforcement Learning (MBRL), model learning is critical since an inaccurate model can bias policy learning via generating misleading samples. However, learning an accurate model can be difficult since the policy is continually updated and the induced distribution over visited states used for model learning shifts accordingly. Prior methods alleviate this issue by quantifying the uncertainty of model-generated samples. However, these methods only quantify the uncertainty passively after the samples were generated, rather than foreseeing the uncertainty before model trajectories fall into those highly uncertain regions. The resulting low-quality samples can induce unstable learning targets and hinder the optimization of the policy. Moreover, while being learned to minimize one-step prediction errors, the model is generally used to predict for multiple steps, leading to a mismatch between the objectives of model learning and model usage. To this end, we propose \emph{Plan To Predict} (P2P), an MBRL framework that treats the model rollout process as a sequential decision making problem by reversely considering the model as a decision maker and the current policy as the dynamics. In this way, the model can quickly adapt to the current policy and foresee the multi-step future uncertainty when generating trajectories. Theoretically, we show that the performance of P2P can be guaranteed by approximately optimizing a lower bound of the true environment return. Empirical results demonstrate that P2P achieves state-of-the-art performance on several challenging benchmark tasks.

Why do we not profit from our long-existing classical robotics knowledge and look for some alternative way for data collection? The situation ignoring all existing methods might be such a waste. This article argues that a dataset created using a classical robotics algorithm is a crucial part of future development. This developed classic algorithm has a perfect domain adaptation and generalization property, and most importantly, collecting datasets based on them is quite easy. It is well known that current robot skill-learning approaches perform exceptionally badly in the unseen domain, and their performance against adversarial attacks is quite limited as long as they do not have a very exclusive big dataset. Our experiment is the initial steps of using a dataset created by classical robotics codes. Our experiment investigated possible trajectory collection based on classical robotics. It addressed some advantages and disadvantages and pointed out other future development ideas.