Foundation models require fine-tuning to ensure their generative outputs align with intended results for specific tasks. Automating this fine-tuning process is challenging, as it typically needs human feedback that can be expensive to acquire. We present AutoRefine, a method that leverages reinforcement learning for targeted fine-tuning, utilizing direct feedback from measurable performance improvements in specific downstream tasks. We demonstrate the method for a problem arising in algorithmic hiring platforms where linguistic biases influence a recommendation system. In this setting, a generative model seeks to rewrite given job specifications to receive more diverse candidate matches from a recommendation engine which matches jobs to candidates. Our model detects and regulates biases in job descriptions to meet diversity and fairness criteria. The experiments on a public hiring dataset and a real-world hiring platform showcase how large language models can assist in identifying and mitigation biases in the real world.

In recent years, meta-reinforcement learning (meta-RL) algorithm has been proposed to improve sample efficiency in the field of decision-making and control, enabling agents to learn new knowledge from a small number of samples. However, most research uses the Gaussian distribution to extract task representation, which is poorly adapted to tasks that change in non-stationary environment. To address this problem, we propose a novel meta-reinforcement learning method by leveraging Gaussian mixture model and the transformer network to construct task inference model. The Gaussian mixture model is utilized to extend the task representation and conduct explicit encoding of tasks. Specifically, the classification of tasks is encoded through transformer network to determine the Gaussian component corresponding to the task. By leveraging task labels, the transformer network is trained using supervised learning. We validate our method on MuJoCo benchmarks with non-stationary and multi-task environments. Experimental results demonstrate that the proposed method dramatically improves sample efficiency and accurately recognizes the classification of the tasks, while performing excellently in the environment.

Deep Reinforcement Learning (DRL) has been extensively used to address portfolio optimization problems. The DRL agents acquire knowledge and make decisions through unsupervised interactions with their environment without requiring explicit knowledge of the joint dynamics of portfolio assets. Among these DRL algorithms, the combination of actor-critic algorithms and deep function approximators is the most widely used DRL algorithm. Here, we find that training the DRL agent using the actor-critic algorithm and deep function approximators may lead to scenarios where the improvement in the DRL agent's risk-adjusted profitability is not significant. We propose that such situations primarily arise from the following two problems: sparsity in positive reward and the curse of dimensionality. These limitations prevent DRL agents from comprehensively learning asset price change patterns in the training environment. As a result, the DRL agents cannot explore the dynamic portfolio optimization policy to improve the risk-adjusted profitability in the training process. To address these problems, we propose a novel multi-agent Hierarchical Deep Reinforcement Learning (HDRL) algorithmic framework in this research. Under this framework, the agents work together as a learning system for portfolio optimization. Specifically, by designing an auxiliary agent that works together with the executive agent for optimal policy exploration, the learning system can focus on exploring the policy with higher risk-adjusted return in the action space with positive return and low variance. In this way, we can overcome the issue of the curse of dimensionality and improve the training efficiency in the positive reward sparse environment.

This paper reports on learning a reward map for social navigation in dynamic environments where the robot can reason about its path at any time, given agents' trajectories and scene geometry. Humans navigating in dense and dynamic indoor environments often work with several implied social rules. A rule-based approach fails to model all possible interactions between humans, robots, and scenes. We propose a novel Smooth Maximum Entropy Deep Inverse Reinforcement Learning (S-MEDIRL) algorithm that can extrapolate beyond expert demos to better encode scene navigability from few-shot demonstrations. The agent learns to predict the cost maps reasoning on trajectory data and scene geometry. The agent samples a trajectory that is then executed using a local crowd navigation controller. We present results in a photo-realistic simulation environment, with a robot and a human navigating a narrow crossing scenario. The robot implicitly learns to exhibit social behaviors such as yielding to oncoming traffic and avoiding deadlocks. We compare the proposed approach to the popular model-based crowd navigation algorithm ORCA and a rule-based agent that exhibits yielding.

In reinforcement learning (RL), continuing tasks refer to tasks where the agent-environment interaction is ongoing and can not be broken down into episodes. These tasks are suitable when environment resets are unavailable, agent-controlled, or predefined but where all rewards-including those beyond resets-are critical. These scenarios frequently occur in real-world applications and can not be modeled by episodic tasks. While modern deep RL algorithms have been extensively studied and well understood in episodic tasks, their behavior in continuing tasks remains underexplored. To address this gap, we provide an empirical study of several well-known deep RL algorithms using a suite of continuing task testbeds based on Mujoco and Atari environments, highlighting several key insights concerning continuing tasks. Using these testbeds, we also investigate the effectiveness of a method for improving temporal-difference-based RL algorithms in continuing tasks by centering rewards, as introduced by Naik et al. (2024). While their work primarily focused on this method in conjunction with Q-learning, our results extend their findings by demonstrating that this method is effective across a broader range of algorithms, scales to larger tasks, and outperforms two other reward-centering approaches.

In this work, we build upon the offline reinforcement learning algorithm TD7[Fujimoto et al., 2024],which incorporates State-Action Learned Embeddings (SALE) and a prioritized experience replay buffer (LAP), propose a model-free actor-critic algorithm that integrates ensemble Q-networks and a gradient diversity penalty from EDAC[An et al., 2021]. The ensemble Q-networks introduce penalties to guide the actor network toward in-distribution actions, effectively addressing the challenge of out-of-distribution actions. Meanwhile, the gradient diversity penalty encourages diverse Q-value gradients, further suppressing overestimation for out-of-distribution actions. Additionally, our method retains an adjustable behavior cloning (BC) term that directs the actor network toward dataset actions during early training stages, while gradually reducing its influence as the precision of the Q-ensemble improves. These enhancements work synergistically to improve the stability and precision of the training. Experimental results on the D4RL MuJoCo benchmarks demonstrate that our algorithm achieves higher convergence speed, stability, and performance compared to existing methods.

This Research proposes a Novel Reinforcement Learning (RL) model to optimise malware forensics investigation during cyber incident response. It aims to improve forensic investigation efficiency by reducing false negatives and adapting current practices to evolving malware signatures. The proposed RL framework leverages techniques such as Q-learning and the Markov Decision Process (MDP) to train the system to identify malware patterns in live memory dumps, thereby automating forensic tasks. The RL model is based on a detailed malware workflow diagram that guides the analysis of malware artefacts using static and behavioural techniques as well as machine learning algorithms. Furthermore, it seeks to address challenges in the UK justice system by ensuring the accuracy of forensic evidence. We conduct testing and evaluation in controlled environments, using datasets created with Windows operating systems to simulate malware infections. The experimental results demonstrate that RL improves malware detection rates compared to conventional methods, with the RL model's performance varying depending on the complexity and learning rate of the environment. The study concludes that while RL offers promising potential for automating malware forensics, its efficacy across diverse malware types requires ongoing refinement of reward systems and feature extraction methods.

Double reinforcement learning (DRL) enables statistically efficient inference on the value of a policy in a nonparametric Markov Decision Process (MDP) given trajectories generated by another policy. However, this approach necessarily requires stringent overlap between the state distributions, which is often violated in practice. To relax this requirement and extend DRL, we study efficient inference on linear functionals of the $Q$-function (of which policy value is a special case) in infinite-horizon, time-invariant MDPs under semiparametric restrictions on the $Q$-function. These restrictions can reduce the overlap requirement and lower the efficiency bound, yielding more precise estimates. As an important example, we study the evaluation of long-term value under domain adaptation, given a few short trajectories from the new domain and restrictions on the difference between the domains. This can be used for long-term causal inference. Our method combines flexible estimates of the $Q$-function and the Riesz representer of the functional of interest (e.g., the stationary state density ratio for policy value) and is automatic in that we do not need to know the form of the latter - only the functional we care about. To address potential model misspecification bias, we extend the adaptive debiased machine learning (ADML) framework of \citet{van2023adaptive} to construct nonparametrically valid and superefficient estimators that adapt to the functional form of the $Q$-function. As a special case, we propose a novel adaptive debiased plug-in estimator that uses isotonic-calibrated fitted $Q$-iteration - a new calibration algorithm for MDPs - to circumvent the computational challenges of estimating debiasing nuisances from min-max objectives.

``Distribution shift'' is the main obstacle to the success of offline reinforcement learning. A learning policy may take actions beyond the behavior policy's knowledge, referred to as Out-of-Distribution (OOD) actions. The Q-values for these OOD actions can be easily overestimated. As a result, the learning policy is biased by using incorrect Q-value estimates. One common approach to avoid Q-value overestimation is to make a pessimistic adjustment. Our key idea is to penalize the Q-values of OOD actions associated with high uncertainty. In this work, we propose Q-Distribution Guided Q-Learning (QDQ), which applies a pessimistic adjustment to Q-values in OOD regions based on uncertainty estimation. This uncertainty measure relies on the conditional Q-value distribution, learned through a high-fidelity and efficient consistency model. Additionally, to prevent overly conservative estimates, we introduce an uncertainty-aware optimization objective for updating the Q-value function. The proposed QDQ demonstrates solid theoretical guarantees for the accuracy of Q-value distribution learning and uncertainty measurement, as well as the performance of the learning policy. QDQ consistently shows strong performance on the D4RL benchmark and achieves significant improvements across many tasks.

The fuzzy job shop scheduling problem (FJSSP) emerges as an innovative extension to the job shop scheduling problem (JSSP), incorporating a layer of uncertainty that aligns the problem more closely with the complexities of real-world manufacturing environments. This improvement increases the computational complexity of deriving the solution while improving its applicability. In the domain of deterministic scheduling, neural combinatorial optimization (NCO) has recently demonstrated remarkable efficacy. However, its application to the realm of fuzzy scheduling has been relatively unexplored. This paper aims to bridge this gap by investigating the feasibility of employing neural networks to assimilate and process fuzzy information for the resolution of FJSSP, thereby leveraging the advancements in NCO to enhance fuzzy scheduling methodologies. To achieve this, we approach the FJSSP as a generative task and introduce an expectation-maximization algorithm-based autoregressive model (EMARM) to address it. During training, our model alternates between generating scheduling schemes from given instances (E-step) and adjusting the autoregressive model weights based on these generated schemes (M-step). This novel methodology effectively navigates around the substantial hurdle of obtaining ground-truth labels, which is a prevalent issue in NCO frameworks. In testing, the experimental results demonstrate the superior capability of EMARM in addressing the FJSSP, showcasing its effectiveness and potential for practical applications in fuzzy scheduling.