Success in racing requires a unique combination of vehicle setup, understanding of the racetrack, and human expertise. Since building and testing many different vehicle configurations in the real world is prohibitively expensive, high-fidelity simulation is a critical part of racecar development. However, testing different vehicle configurations still requires expert human input in order to evaluate their performance on different racetracks. In this work, we present the first steps towards an autonomous test driver, trained using deep reinforcement learning, capable of evaluating changes in vehicle setup on racing performance while driving at the level of the best human drivers. In addition, the autonomous driver model can be tuned to exhibit more human-like behavioral patterns by incorporating imitation learning into the RL training process. This extension permits the possibility of driver-specific vehicle setup optimization.

Advances in tracking technologies have spurred the rapid growth of large-scale trajectory data. Building a compact collection of pathlets, referred to as a trajectory pathlet dictionary, is essential for supporting mobility-related applications. Existing methods typically adopt a top-down approach, generating numerous candidate pathlets and selecting a subset, leading to high memory usage and redundant storage from overlapping pathlets. To overcome these limitations, we propose a bottom-up strategy that incrementally merges basic pathlets to build the dictionary, reducing memory requirements by up to 24,000 times compared to baseline methods. The approach begins with unit-length pathlets and iteratively merges them while optimizing utility, which is defined using newly introduced metrics of trajectory loss and representability. We develop a deep reinforcement learning framework, PathletRL, which utilizes Deep Q-Networks (DQN) to approximate the utility function, resulting in a compact and efficient pathlet dictionary. Experiments on both synthetic and real-world datasets demonstrate that our method outperforms state-of-the-art techniques, reducing the size of the constructed dictionary by up to 65.8%. Additionally, our results show that only half of the dictionary pathlets are needed to reconstruct 85% of the original trajectory data. Building on PathletRL, we introduce PathletRL++, which extends the original model by incorporating a richer state representation and an improved reward function to optimize decision-making during pathlet merging. These enhancements enable the agent to gain a more nuanced understanding of the environment, leading to higher-quality pathlet dictionaries. PathletRL++ achieves even greater dictionary size reduction, surpassing the performance of PathletRL, while maintaining high trajectory representability.

Offline reinforcement learning (RL) enables policy learning from static datasets, without active environment interaction, making it ideal for high-stakes applications like autonomous driving and robot manipulation [Levine et al., 2020, Ma et al., 2022, Wang et al., 2024a]. A key challenge in offline RL is managing the discrepancy between the learned policy and the behaviour policy that generated the dataset. Small discrepancies can hinder policy improvement, while large discrepancies push the learned policy into uncharted areas, causing significant extrapolation errors and poor generalisation [Fujimoto et al., 2019, Yang et al., 2023]. Addressing these challenges, existing research has proposed various solutions. Conservative approaches penalise actions that stray into out-of-distribution (OOD) regions [Yu et al., 2020, Kumar et al., 2020], while others regularise the policy by minimising its divergence from the behaviour policy, ensuring better fidelity to the dataset [Fujimoto and Gu, 2021, Wu et al., 2019].

Multi-agent reinforcement learning (MARL), as a theoretical framework for modeling agent behavior in complex game environments, has become a significant area of research [42, 37]. Unlike traditional game theory, MARL typically allows agents to learn strategies through repeated interactions to achieve equilibrium [34]. By relaxing the assumptions of agent rationality and independence, MARL can learn strategies efficiently with arbitrary environments and opponents [10, 20, 17]. Current applications of MARL in game environments are primarily focused on zero-sum games (fully competitive) [10, 41] and fully cooperative games [12, 38], since the behavioral preferences of opponent agents in these environments are relatively easy to predict. Nevertheless, the environments in practical applications, e.g., economic markets, robotics and distributed control, may have multiple equilibrium [16, 40], and opponent agents may not exhibit clear preferences for different strategies, thus agents need to learn strategies in general-sum games [8, 7]. The Prisoner's dilemma [3, 14] is a classic example of the tension between mutual cooperation leading to a win-win situation and focusing solely on self-interest leading to a lose-lose situation. Therefore, modeling and shaping the behavior of opponent agents is the main challenge for the application of MARL in these environments [11]. Recent advancements in MARL have introduced opponent modeling and shaping techniques that allow agents to learn not just their own strategies, but also to predict and influence the strategies of the opponent, such as [10, 20, 36]. These methods show promise in improving the efficiency of strategy learning by incorporating the behavior of other agents into the learning process.

Despite significant advancements in deep reinforcement learning (DRL)-based autonomous driving policies, these policies still exhibit vulnerability to adversarial attacks. This vulnerability poses a formidable challenge to the practical deployment of these policies in autonomous driving. Designing effective adversarial attacks is an indispensable prerequisite for enhancing the robustness of these policies. In view of this, we present a novel stealthy and efficient adversarial attack method for DRL-based autonomous driving policies. Specifically, we introduce a DRL-based adversary designed to trigger safety violations (e.g., collisions) by injecting adversarial samples at critical moments. We model the attack as a mixed-integer optimization problem and formulate it as a Markov decision process. Then, we train the adversary to learn the optimal policy for attacking at critical moments without domain knowledge. Furthermore, we introduce attack-related information and a trajectory clipping method to enhance the learning capability of the adversary. Finally, we validate our method in an unprotected left-turn scenario across different traffic densities. The experimental results show that our method achieves more than 90% collision rate within three attacks in most cases. Furthermore, our method achieves more than 130% improvement in attack efficiency compared to the unlimited attack method.

Portfolio management is an important yet challenging task in AI for FinTech, which aims to allocate investors' budgets among different assets to balance the risk and return of an investment. In this study, we propose a general Multi-objectIve framework with controLLable rIsk for pOrtfolio maNagement (MILLION), which consists of two main phases, i.e., return-related maximization and risk control. Specifically, in the return-related maximization phase, we introduce two auxiliary objectives, i.e., return rate prediction, and return rate ranking, combined with portfolio optimization to remit the overfitting problem and improve the generalization of the trained model to future markets. Subsequently, in the risk control phase, we propose two methods, i.e., portfolio interpolation and portfolio improvement, to achieve fine-grained risk control and fast risk adaption to a user-specified risk level. For the portfolio interpolation method, we theoretically prove that the risk can be perfectly controlled if the to-be-set risk level is in a proper interval. In addition, we also show that the return rate of the adjusted portfolio after portfolio interpolation is no less than that of the min-variance optimization, as long as the model in the reward maximization phase is effective. Furthermore, the portfolio improvement method can achieve greater return rates while keeping the same risk level compared to portfolio interpolation. Extensive experiments are conducted on three real-world datasets. The results demonstrate the effectiveness and efficiency of the proposed framework.

In this paper, we tackle the challenging problem of delayed rewards in reinforcement learning (RL). While Proximal Policy Optimization (PPO) has emerged as a leading Policy Gradient method, its performance can degrade under delayed rewards. We introduce two key enhancements to PPO: a hybrid policy architecture that combines an offline policy (trained on expert demonstrations) with an online PPO policy, and a reward shaping mechanism using Time Window Temporal Logic (TWTL). The hybrid architecture leverages offline data throughout training while maintaining PPO's theoretical guarantees. Building on the monotonic improvement framework of Trust Region Policy Optimization (TRPO), we prove that our approach ensures improvement over both the offline policy and previous iterations, with a bounded performance gap of $(2\varsigma\gamma\alpha^2)/(1-\gamma)^2$, where $\alpha$ is the mixing parameter, $\gamma$ is the discount factor, and $\varsigma$ bounds the expected advantage. Additionally, we prove that our TWTL-based reward shaping preserves the optimal policy of the original problem. TWTL enables formal translation of temporal objectives into immediate feedback signals that guide learning. We demonstrate the effectiveness of our approach through extensive experiments on an inverted pendulum and a lunar lander environments, showing improvements in both learning speed and final performance compared to standard PPO and offline-only approaches.

Diffusion models achieve superior generation quality but suffer from slow generation speed due to the iterative nature of denoising. In contrast, consistency models, a new generative family, achieve competitive performance with significantly faster sampling. These models are trained either through consistency distillation, which leverages pretrained diffusion models, or consistency training/tuning directly from raw data. In this work, we propose a novel framework for understanding consistency models by modeling the denoising process of the diffusion model as a Markov Decision Process (MDP) and framing consistency model training as the value estimation through Temporal Difference~(TD) Learning. More importantly, this framework allows us to analyze the limitations of current consistency training/tuning strategies. Built upon Easy Consistency Tuning (ECT), we propose Stable Consistency Tuning (SCT), which incorporates variance-reduced learning using the score identity. SCT leads to significant performance improvements on benchmarks such as CIFAR-10 and ImageNet-64. On ImageNet-64, SCT achieves 1-step FID 2.42 and 2-step FID 1.55, a new SoTA for consistency models.

Mean field type games (MFTGs) describe Nash equilibria between large coalitions: each coalition consists of a continuum of cooperative agents who maximize the average reward of their coalition while interacting non-cooperatively with a finite number of other coalitions. Although the theory has been extensively developed, we are still lacking efficient and scalable computational methods. Here, we develop reinforcement learning methods for such games in a finite space setting with general dynamics and reward functions. We start by proving that MFTG solution yields approximate Nash equilibria in finite-size coalition games. We then propose two algorithms. The first is based on quantization of mean-field spaces and Nash Q-learning. We provide convergence and stability analysis. We then propose a deep reinforcement learning algorithm, which can scale to larger spaces. Numerical experiments in 5 environments with mean-field distributions of dimension up to $200$ show the scalability and efficiency of the proposed method.

Recent work in language modeling has raised the possibility of self-improvement, where a language models evaluates and refines its own generations to achieve higher performance without external feedback. It is impossible for this self-improvement to create information that is not already in the model, so why should we expect that this will lead to improved capabilities? We offer a new perspective on the capabilities of self-improvement through a lens we refer to as sharpening. Motivated by the observation that language models are often better at verifying response quality than they are at generating correct responses, we formalize self-improvement as using the model itself as a verifier during post-training in order to ``sharpen'' the model to one placing large mass on high-quality sequences, thereby amortizing the expensive inference-time computation of generating good sequences. We begin by introducing a new statistical framework for sharpening in which the learner aims to sharpen a pre-trained base policy via sample access, and establish fundamental limits. Then we analyze two natural families of self-improvement algorithms based on SFT and RLHF. We find that (i) the SFT-based approach is minimax optimal whenever the initial model has sufficient coverage, but (ii) the RLHF-based approach can improve over SFT-based self-improvement by leveraging online exploration, bypassing the need for coverage. Finally, we empirically validate the sharpening mechanism via inference-time and amortization experiments. We view these findings as a starting point toward a foundational understanding that can guide the design and evaluation of self-improvement algorithms.