Knowledge distillation (KD) enhances the performance of a student network by allowing it to learn the knowledge transferred from a teacher network incrementally. Existing methods dynamically adjust the temperature to enable the student network to adapt to the varying learning difficulties at different learning stages of KD. KD is a continuous process, but when adjusting the temperature, these methods consider only the immediate benefits of the operation in the current learning phase and fail to take into account its future returns. To address this issue, we formulate the adjustment of temperature as a sequential decision-making task and propose a method based on reinforcement learning, termed RLKD. Importantly, we design a novel state representation to enable the agent to make more informed action (i.e. instance temperature adjustment). To handle the problem of delayed rewards in our method due to the KD setting, we explore an instance reward calibration approach. In addition,we devise an efficient exploration strategy that enables the agent to learn valuable instance temperature adjustment policy more efficiently. Our framework can serve as a plug-and-play technique to be inserted into various KD methods easily, and we validate its effectiveness on both image classification and object detection tasks. Our project is at https://www.zayx.me/ITKD.github.io/.
Policy Mirror Descent (PMD) is a powerful and theoretically sound methodology for sequential decision-making. However, it is not directly applicable to Reinforcement Learning (RL) due to the inaccessibility of explicit action-value functions. We address this challenge by introducing a novel approach based on learning a world model of the environment using conditional mean embeddings. Leveraging tools from operator theory we derive a closed-form expression of the action-value function in terms of the world model via simple matrix operations. Combining these estimators with PMD leads to POWR, a new RL algorithm for which we prove convergence rates to the global optimum. Preliminary experiments in finite and infinite state settings support the effectiveness of our method
This paper tackles the problem of mitigating catastrophic risk (which is risk with very low frequency but very high severity) in the context of a sequential decision making process. This problem is particularly challenging due to the scarcity of observations in the far tail of the distribution of cumulative costs (negative rewards). A policy gradient algorithm is developed, that we call POTPG. It is based on approximations of the tail risk derived from extreme value theory. Numerical experiments highlight the out-performance of our method over common benchmarks, relying on the empirical distribution. An application to financial risk management, more precisely to the dynamic hedging of a financial option, is presented.
High-frequency trading (HFT) that executes algorithmic trading in short time scales, has recently occupied the majority of cryptocurrency market. Besides traditional quantitative trading methods, reinforcement learning (RL) has become another appealing approach for HFT due to its terrific ability of handling high-dimensional financial data and solving sophisticated sequential decision-making problems, \emph{e.g.,} hierarchical reinforcement learning (HRL) has shown its promising performance on second-level HFT by training a router to select only one sub-agent from the agent pool to execute the current transaction. However, existing RL methods for HFT still have some defects: 1) standard RL-based trading agents suffer from the overfitting issue, preventing them from making effective policy adjustments based on financial context; 2) due to the rapid changes in market conditions, investment decisions made by an individual agent are usually one-sided and highly biased, which might lead to significant loss in extreme markets. To tackle these problems, we propose a novel Memory Augmented Context-aware Reinforcement learning method On HFT, \emph{a.k.a.} MacroHFT, which consists of two training phases: 1) we first train multiple types of sub-agents with the market data decomposed according to various financial indicators, specifically market trend and volatility, where each agent owns a conditional adapter to adjust its trading policy according to market conditions; 2) then we train a hyper-agent to mix the decisions from these sub-agents and output a consistently profitable meta-policy to handle rapid market fluctuations, equipped with a memory mechanism to enhance the capability of decision-making. Extensive experiments on various cryptocurrency markets demonstrate that MacroHFT can achieve state-of-the-art performance on minute-level trading tasks.
Reward Machines provide an automaton-inspired structure for specifying instructions, safety constraints, and other temporally extended reward-worthy behaviour. By exposing the underlying structure of a reward function, they enable the decomposition of an RL task, leading to impressive gains in sample efficiency. Although Reward Machines and similar formal specifications have a rich history of application towards sequential decision-making problems, they critically rely on a ground-truth interpretation of the domain-specific vocabulary that forms the building blocks of the reward function--such ground-truth interpretations are elusive in the real world due in part to partial observability and noisy sensing. In this work, we explore the use of Reward Machines for Deep RL in noisy and uncertain environments. We characterize this problem as a POMDP and propose a suite of RL algorithms that exploit task structure under uncertain interpretation of the domain-specific vocabulary. Through theory and experiments, we expose pitfalls in naive approaches to this problem while simultaneously demonstrating how task structure can be successfully leveraged under noisy interpretations of the vocabulary.
Partially Observable Markov Decision Processes (POMDPs) are powerful models for sequential decision making under transition and observation uncertainties. This paper studies the challenging yet important problem in POMDPs known as the (indefinite-horizon) Maximal Reachability Probability Problem (MRPP), where the goal is to maximize the probability of reaching some target states. This is also a core problem in model checking with logical specifications and is naturally undiscounted (discount factor is one). Inspired by the success of point-based methods developed for discounted problems, we study their extensions to MRPP. Specifically, we focus on trial-based heuristic search value iteration techniques and present a novel algorithm that leverages the strengths of these techniques for efficient exploration of the belief space (informed search via value bounds) while addressing their drawbacks in handling loops for indefinite-horizon problems. The algorithm produces policies with two-sided bounds on optimal reachability probabilities. We prove convergence to an optimal policy from below under certain conditions. Experimental evaluations on a suite of benchmarks show that our algorithm outperforms existing methods in almost all cases in both probability guarantees and computation time.
Reinforcement Learning (RL) has achieved great success in sequential decision-making problems, but often at the cost of a large number of agent-environment interactions. To improve sample efficiency, methods like Reinforcement Learning from Expert Demonstrations (RLED) introduce external expert demonstrations to facilitate agent exploration during the learning process. In practice, these demonstrations, which are often collected from human users, are costly and hence often constrained to a limited amount. How to select the best set of human demonstrations that is most beneficial for learning therefore becomes a major concern. This paper presents EARLY (Episodic Active Learning from demonstration querY), an algorithm that enables a learning agent to generate optimized queries of expert demonstrations in a trajectory-based feature space. Based on a trajectory-level estimate of uncertainty in the agent's current policy, EARLY determines the optimized timing and content for feature-based queries. By querying episodic demonstrations as opposed to isolated state-action pairs, EARLY improves the human teaching experience and achieves better learning performance. We validate the effectiveness of our method in three simulated navigation tasks of increasing difficulty. The results show that our method is able to achieve expert-level performance for all three tasks with convergence over 30\% faster than other baseline methods when demonstrations are generated by simulated oracle policies. The results of a follow-up pilot user study (N=18) further validate that our method can still maintain a significantly better convergence in the case of human expert demonstrators while achieving a better user experience in perceived task load and consuming significantly less human time.
This paper studies policy evaluation with multiple data sources, especially in scenarios that involve one experimental dataset with two arms, complemented by a historical dataset generated under a single control arm. We propose novel data integration methods that linearly integrate base policy value estimators constructed based on the experimental and historical data, with weights optimized to minimize the mean square error (MSE) of the resulting combined estimator. We further apply the pessimistic principle to obtain more robust estimators, and extend these developments to sequential decision making. Theoretically, we establish non-asymptotic error bounds for the MSEs of our proposed estimators, and derive their oracle, efficiency and robustness properties across a broad spectrum of reward shift scenarios. Numerical experiments and real-data-based analyses from a ridesharing company demonstrate the superior performance of the proposed estimators.
Finetuning language agents with reasoning-action trajectories is effective, but obtaining these trajectories from human annotations or stronger models is costly and sometimes impractical. In this paper, we investigate the use of self-training in language agents, which can generate supervision from the agent itself, offering a promising alternative without relying on human or stronger model demonstrations. Self-training, however, requires high-quality model-generated samples, which are hard to obtain for challenging language agent tasks. To address this, we present Reflection-Reinforced Self-Training (Re-ReST), which uses a \textit{reflector} to refine low-quality generated samples during self-training. The reflector takes the agent's output and feedback from an external environment (e.g., unit test results in code generation) to produce improved samples. This technique enhances the quality of inferior samples and efficiently enriches the self-training dataset with higher-quality samples. We conduct extensive experiments on open-source language agents across tasks, including multi-hop question answering, sequential decision-making, code generation, visual question answering, and text-to-image generation. The results demonstrate the effectiveness of self-training and Re-ReST in language agent tasks, with self-training improving baselines by 7.6\% on HotpotQA and 28.4\% on AlfWorld, and Re-ReST further boosting performance by 2.0\% and 14.1\%, respectively. Our studies also confirm the efficiency of using a reflector to generate high-quality samples for self-training. Moreover, we demonstrate a method to employ reflection during inference without ground-truth feedback, addressing the limitation of previous reflection work. Our code is released at https://github.com/PlusLabNLP/Re-ReST.
In traditional federated learning, a single global model cannot perform equally well for all clients. Therefore, the need to achieve the client-level fairness in federated system has been emphasized, which can be realized by modifying the static aggregation scheme for updating the global model to an adaptive one, in response to the local signals of the participating clients. Our work reveals that existing fairness-aware aggregation strategies can be unified into an online convex optimization framework, in other words, a central server's sequential decision making process. To enhance the decision making capability, we propose simple and intuitive improvements for suboptimal designs within existing methods, presenting AAggFF. Considering practical requirements, we further subdivide our method tailored for the cross-device and the cross-silo settings, respectively. Theoretical analyses guarantee sublinear regret upper bounds for both settings: $\mathcal{O}(\sqrt{T \log{K}})$ for the cross-device setting, and $\mathcal{O}(K \log{T})$ for the cross-silo setting, with $K$ clients and $T$ federation rounds. Extensive experiments demonstrate that the federated system equipped with AAggFF achieves better degree of client-level fairness than existing methods in both practical settings. Code is available at https://github.com/vaseline555/AAggFF
Sequential decision-making algorithms such as reinforcement learning (RL) in real-world scenarios inevitably face environments with partial observability. This paper scrutinizes the effectiveness of a popular architecture, namely Transformers, in Partially Observable Markov Decision Processes (POMDPs) and reveals its theoretical limitations. We establish that regular languages, which Transformers struggle to model, are reducible to POMDPs. This poses a significant challenge for Transformers in learning POMDP-specific inductive biases, due to their lack of inherent recurrence found in other models like RNNs. This paper casts doubt on the prevalent belief in Transformers as sequence models for RL and proposes to introduce a point-wise recurrent structure. The Deep Linear Recurrent Unit (LRU) emerges as a well-suited alternative for Partially Observable RL, with empirical results highlighting the sub-optimal performance of the Transformer and considerable strength of LRU.
Clustering-based nearest neighbor search is an effective method in which points are partitioned into geometric shards to form an index, with only a few shards searched during query processing to find a set of top-$k$ vectors. Even though the search efficacy is heavily influenced by the algorithm that identifies the shards to probe, it has received little attention in the literature. This work bridges that gap by studying routing in clustering-based maximum inner product search. We unpack existing routers and notice the surprising contribution of optimism. We then take a page from the sequential decision making literature and formalize that insight following the principle of ``optimism in the face of uncertainty.'' In particular, we present a framework that incorporates the moments of the distribution of inner products within each shard to estimate the maximum inner product. We then present an instance of our algorithm that uses only the first two moments to reach the same accuracy as state-of-the-art routers such as ScaNN by probing up to $50\%$ fewer points on benchmark datasets. Our algorithm is also space-efficient: we design a sketch of the second moment whose size is independent of the number of points and requires $\mathcal{O}(1)$ vectors per shard.
Transformer-based trajectory optimization methods have demonstrated exceptional performance in offline Reinforcement Learning (offline RL). Yet, it poses challenges due to substantial parameter size and limited scalability, which is particularly critical in sequential decision-making scenarios where resources are constrained such as in robots and drones with limited computational power. Mamba, a promising new linear-time sequence model, offers performance on par with transformers while delivering substantially fewer parameters on long sequences. As it remains unclear whether Mamba is compatible with trajectory optimization, this work aims to conduct comprehensive experiments to explore the potential of Decision Mamba (dubbed DeMa) in offline RL from the aspect of data structures and essential components with the following insights: (1) Long sequences impose a significant computational burden without contributing to performance improvements since DeMa's focus on sequences diminishes approximately exponentially. Consequently, we introduce a Transformer-like DeMa as opposed to an RNN-like DeMa. (2) For the components of DeMa, we identify the hidden attention mechanism as a critical factor in its success, which can also work well with other residual structures and does not require position embedding. Extensive evaluations demonstrate that our specially designed DeMa is compatible with trajectory optimization and surpasses previous methods, outperforming Decision Transformer (DT) with higher performance while using 30\% fewer parameters in Atari, and exceeding DT with only a quarter of the parameters in MuJoCo.
In sequential decision-making problems involving sensitive attributes like race and gender, reinforcement learning (RL) agents must carefully consider long-term fairness while maximizing returns. Recent works have proposed many different types of fairness notions, but how unfairness arises in RL problems remains unclear. In this paper, we address this gap in the literature by investigating the sources of inequality through a causal lens. We first analyse the causal relationships governing the data generation process and decompose the effect of sensitive attributes on long-term well-being into distinct components. We then introduce a novel notion called dynamics fairness, which explicitly captures the inequality stemming from environmental dynamics, distinguishing it from those induced by decision-making or inherited from the past. This notion requires evaluating the expected changes in the next state and the reward induced by changing the value of the sensitive attribute while holding everything else constant. To quantitatively evaluate this counterfactual concept, we derive identification formulas that allow us to obtain reliable estimations from data. Extensive experiments demonstrate the effectiveness of the proposed techniques in explaining, detecting, and reducing inequality in reinforcement learning. We publicly release code at https://github.com/familyld/InsightFair.
We study the problem of online sequential decision-making given auxiliary demonstrations from experts who made their decisions based on unobserved contextual information. These demonstrations can be viewed as solving related but slightly different problems than what the learner faces. This setting arises in many application domains, such as self-driving cars, healthcare, and finance, where expert demonstrations are made using contextual information, which is not recorded in the data available to the learning agent. We model the problem as zero-shot meta-reinforcement learning with an unknown distribution over the unobserved contextual variables and a Bayesian regret minimization objective, where the unobserved variables are encoded as parameters with an unknown prior. We propose the Experts-as-Priors algorithm (ExPerior), an empirical Bayes approach that utilizes expert data to establish an informative prior distribution over the learner's decision-making problem. This prior distribution enables the application of any Bayesian approach for online decision-making, such as posterior sampling. We demonstrate that our strategy surpasses existing behaviour cloning, online, and online-offline baselines for multi-armed bandits, Markov decision processes (MDPs), and partially observable MDPs, showcasing the broad reach and utility of ExPerior in using expert demonstrations across different decision-making setups.
We study a sequential decision-making problem for a profit-maximizing operator of an autonomous mobility-on-demand system. Optimizing a central operator's vehicle-to-request dispatching policy requires efficient and effective fleet control strategies. To this end, we employ a multi-agent Soft Actor-Critic algorithm combined with weighted bipartite matching. We propose a novel vehicle-based algorithm architecture and adapt the critic's loss function to appropriately consider coordinated actions. Furthermore, we extend our algorithm to incorporate rebalancing capabilities. Through numerical experiments, we show that our approach outperforms state-of-the-art benchmarks by up to 12.9% for dispatching and up to 38.9% with integrated rebalancing.
While reinforcement learning (RL) algorithms have been successfully applied across numerous sequential decision-making problems, their generalization to unforeseen testing environments remains a significant concern. In this paper, we study the problem of out-of-distribution (OOD) detection in RL, which focuses on identifying situations at test time that RL agents have not encountered in their training environments. We first propose a clarification of terminology for OOD detection in RL, which aligns it with the literature from other machine learning domains. We then present new benchmark scenarios for OOD detection, which introduce anomalies with temporal autocorrelation into different components of the agent-environment loop. We argue that such scenarios have been understudied in the current literature, despite their relevance to real-world situations. Confirming our theoretical predictions, our experimental results suggest that state-of-the-art OOD detectors are not able to identify such anomalies. To address this problem, we propose a novel method for OOD detection, which we call DEXTER (Detection via Extraction of Time Series Representations). By treating environment observations as time series data, DEXTER extracts salient time series features, and then leverages an ensemble of isolation forest algorithms to detect anomalies. We find that DEXTER can reliably identify anomalies across benchmark scenarios, exhibiting superior performance compared to both state-of-the-art OOD detectors and high-dimensional changepoint detectors adopted from statistics.
Background: Clinical diagnosis is typically reached by following a series of steps recommended by guidelines authored by colleges of experts. Accordingly, guidelines play a crucial role in rationalizing clinical decisions but suffer from limitations as they are built to cover the majority of the population and fail at covering patients with uncommon conditions. Moreover, their updates are long and expensive, making them unsuitable for emerging diseases and practices. Methods: Inspired by guidelines, we formulate the task of diagnosis as a sequential decision-making problem and study the use of Deep Reinforcement Learning (DRL) algorithms to learn the optimal sequence of actions to perform in order to obtain a correct diagnosis from Electronic Health Records (EHRs). We apply DRL on synthetic, but realistic EHRs and develop two clinical use cases: Anemia diagnosis, where the decision pathways follow the schema of a decision tree; and Systemic Lupus Erythematosus (SLE) diagnosis, which follows a weighted criteria score. We particularly evaluate the robustness of our approaches to noisy and missing data since these frequently occur in EHRs. Results: In both use cases, and in the presence of imperfect data, our best DRL algorithms exhibit competitive performance when compared to the traditional classifiers, with the added advantage that they enable the progressive generation of a pathway to the suggested diagnosis which can both guide and explain the decision-making process. Conclusion: DRL offers the opportunity to learn personalized decision pathways to diagnosis. We illustrate with our two use cases their advantages: they generate step-by-step pathways that are self-explanatory; and their correctness is competitive when compared to state-of-the-art approaches.
Decision Transformer, a promising approach that applies Transformer architectures to reinforcement learning, relies on causal self-attention to model sequences of states, actions, and rewards. While this method has shown competitive results, this paper investigates the integration of the Mamba framework, known for its advanced capabilities in efficient and effective sequence modeling, into the Decision Transformer architecture, focusing on the potential performance enhancements in sequential decision-making tasks. Our study systematically evaluates this integration by conducting a series of experiments across various decision-making environments, comparing the modified Decision Transformer, Decision Mamba, with its traditional counterpart. This work contributes to the advancement of sequential decision-making models, suggesting that the architecture and training methodology of neural networks can significantly impact their performance in complex tasks, and highlighting the potential of Mamba as a valuable tool for improving the efficacy of Transformer-based models in reinforcement learning scenarios.
For effective human-agent teaming, robots and other artificial intelligence (AI) agents must infer their human partner's abilities and behavioral response patterns and adapt accordingly. Most prior works make the unrealistic assumption that one or more teammates can act near-optimally. In real-world collaboration, humans and autonomous agents can be suboptimal, especially when each only has partial domain knowledge. In this work, we develop computational modeling and optimization techniques for enhancing the performance of suboptimal human-agent teams, where the human and the agent have asymmetric capabilities and act suboptimally due to incomplete environmental knowledge. We adopt an online Bayesian approach that enables a robot to infer people's willingness to comply with its assistance in a sequential decision-making game. Our user studies show that user preferences and team performance indeed vary with robot intervention styles, and our approach for mixed-initiative collaborations enhances objective team performance ($p<.001$) and subjective measures, such as user's trust ($p<.001$) and perceived likeability of the robot ($p<.001$).