High-quality dialogue is crucial for e-commerce customer service, yet traditional intent-based systems struggle with dynamic, multi-turn interactions. We present MindFlow+, a self-evolving dialogue agent that learns domain-specific behavior by combining large language models (LLMs) with imitation learning and offline reinforcement learning (RL). MindFlow+ introduces two data-centric mechanisms to guide learning: tool-augmented demonstration construction, which exposes the model to knowledge-enhanced and agentic (ReAct-style) interactions for effective tool use; and reward-conditioned data modeling, which aligns responses with task-specific goals using reward signals. To evaluate the model's role in response generation, we introduce the AI Contribution Ratio, a novel metric quantifying AI involvement in dialogue. Experiments on real-world e-commerce conversations show that MindFlow+ outperforms strong baselines in contextual relevance, flexibility, and task accuracy. These results demonstrate the potential of combining LLMs tool reasoning, and reward-guided learning to build domain-specialized, context-aware dialogue systems.

Reinforcement learning has been widely applied to robotic control, but effective policy learning under partial observability remains a major challenge, especially in high-dimensional tasks like humanoid locomotion. To date, no prior work has demonstrated stable training of humanoid policies with incomplete state information in the benchmark Gymnasium Humanoid-v4 environment. The objective in this environment is to walk forward as fast as possible without falling, with rewards provided for staying upright and moving forward, and penalties incurred for excessive actions and external contact forces. This research presents the first successful instance of learning under partial observability in this environment. The learned policy achieves performance comparable to state-of-the-art results with full state access, despite using only one-third to two-thirds of the original states. Moreover, the policy exhibits adaptability to robot properties, such as variations in body part masses. The key to this success is a novel history encoder that processes a fixed-length sequence of past observations in parallel. Integrated into a standard model-free algorithm, the encoder enables performance on par with fully observed baselines. We hypothesize that it reconstructs essential contextual information from recent observations, thereby enabling robust decision-making.

-- Efficient exploration in multi-agent reinforcement learning (MARL) is a challenging problem when receiving only a team reward, especially in environments with sparse rewards. A powerful method to mitigate this issue involves crafting dense individual rewards to guide the agents toward efficient exploration. However, individual rewards generally rely on manually engineered shaping-reward functions that lack high-order intelligence, thus it behaves ineffectively than humans regarding learning and generalization in complex problems. T o tackle these issues, we combine the above two paradigms and propose a novel framework, LIGHT (Learning Individual Intrinsic reward via Incorporating Generalized Human experTise), which can integrate human knowledge into MARL algorithms in an end-to-end manner . LIGHT guides each agent to avoid unnecessary exploration by considering both individual action distribution and human expertise preference distribution. Then, LIGHT designs individual intrinsic rewards for each agent based on actionable representational transformation relevant to Q-learning so that the agents align their action preferences with the human expertise while maximizing the joint action value. Experimental results demonstrate the superiority of our method over representative baselines regarding performance and better knowledge reusability across different sparse-reward tasks on challenging scenarios. Cooperative multi-agent reinforcement learning (MARL) is an important branch in the field of artificial intelligence (AI), playing a crucial role in sequential challenging decision-making problems, such as in autonomous driving [1], sensor networks [2], [3] and robotics control [4]. Centralized training with decentralized execution (CTDE) paradigm has gained substantial attention in cooperative MARL that aims to facilitate agent cooperation by providing global state information during training and executing only based on local observations during execution [5], [6], [7].

Foundation models compress a large amount of information in a single, large neural network, which can then be queried for individual tasks. There are strong parallels between this widespread framework and offline goal-conditioned reinforcement learning algorithms: a universal value function is trained on a large number of goals, and the policy is evaluated on a single goal in each test episode. Extensive research in foundation models has shown that performance can be substantially improved through test-time training, specializing the model to the current goal. We find similarly that test-time offline reinforcement learning on experience related to the test goal can lead to substantially better policies at minimal compute costs. We propose a novel self-supervised data selection criterion, which selects transitions from an offline dataset according to their relevance to the current state and quality with respect to the evaluation goal. We demonstrate across a wide range of high-dimensional loco-navigation and manipulation tasks that fine-tuning a policy on the selected data for a few gradient steps leads to significant performance gains over standard offline pre-training. Our goal-conditioned test-time training (GC-TTT) algorithm applies this routine in a receding-horizon fashion during evaluation, adapting the policy to the current trajectory as it is being rolled out. Finally, we study compute allocation at inference, demonstrating that, at comparable costs, GC-TTT induces performance gains that are not achievable by scaling model size.

This study focuses on the development of a simulation-driven reinforcement learning (RL) framework for optimizing routing decisions in complex queueing network systems, with a particular emphasis on manufacturing and communication applications. Recognizing the limitations of traditional queueing methods, which often struggle with dynamic, uncertain environments, we propose a robust RL approach leveraging Deep Deterministic Policy Gradient (DDPG) combined with Dyna-style planning (Dyna-DDPG). The framework includes a flexible and configurable simulation environment capable of modeling diverse queueing scenarios, disruptions, and unpredictable conditions. Our enhanced Dyna-DDPG implementation incorporates separate predictive models for next-state transitions and rewards, significantly improving stability and sample efficiency. Comprehensive experiments and rigorous evaluations demonstrate the framework's capability to rapidly learn effective routing policies that maintain robust performance under disruptions and scale effectively to larger network sizes. Additionally, we highlight strong software engineering practices employed to ensure reproducibility and maintainability of the framework, enabling practical deployment in real-world scenarios.

This thesis presents the results of a comprehensive research project focused on applying Reinforcement Learning (RL) to the problem of market making in financial markets. Market makers (MMs) play a fundamental role in providing liquidity, yet face significant challenges arising from inventory risk, competition, and non-stationary market dynamics. This research explores how RL, particularly Deep Reinforcement Learning (DRL), can be employed to develop autonomous, adaptive, and profitable market making strategies. The study begins by formulating the MM task as a reinforcement learning problem, designing agents capable of operating in both single-agent and multi-agent settings within a simulated financial environment. It then addresses the complex issue of inventory management using two complementary approaches: reward engineering and Multi-Objective Reinforcement Learning (MORL). While the former uses dynamic reward shaping to guide behavior, the latter leverages Pareto front optimization to explicitly balance competing objectives. To address the problem of non-stationarity, the research introduces POW-dTS, a novel policy weighting algorithm based on Discounted Thompson Sampling. This method allows agents to dynamically select and combine pretrained policies, enabling continual adaptation to shifting market conditions. The experimental results demonstrate that the proposed RL-based approaches significantly outperform traditional and baseline algorithmic strategies across various performance metrics. Overall, this research thesis contributes new methodologies and insights for the design of robust, efficient, and adaptive market making agents, reinforcing the potential of RL to transform algorithmic trading in complex financial systems.

This work presents a novel co-design strategy that integrates trajectory planning and control to handle STL-based tasks in autonomous robots. The method consists of two phases: $(i)$ learning spatio-temporal motion primitives to encapsulate the inherent robot-specific constraints and $(ii)$ constructing an STL-compliant motion plan from these primitives. Initially, we employ reinforcement learning to construct a library of control policies that perform trajectories described by the motion primitives. Then, we map motion primitives to spatio-temporal characteristics. Subsequently, we present a sampling-based STL-compliant motion planning strategy tailored to meet the STL specification. The proposed model-free approach, which generates feasible STL-compliant motion plans across various environments, is validated on differential-drive and quadruped robots across various STL specifications. Demonstration videos are available at https://tinyurl.com/m6zp7rsm.

We introduce Delphos, a deep reinforcement learning framework for assisting the discrete choice model specification process. Unlike traditional approaches that treat model specification as a static optimisation problem, Delphos represents a paradigm shift: it frames this specification challenge as a sequential decision-making problem, formalised as a Markov Decision Process. In this setting, an agent learns to specify well-performing model candidates by choosing a sequence of modelling actions - such as selecting variables, accommodating both generic and alternative-specific taste parameters, applying non-linear transformations, and including interactions with covariates - and interacting with a modelling environment that estimates each candidate and returns a reward signal. Specifically, Delphos uses a Deep Q-Network that receives delayed rewards based on modelling outcomes (e.g., log-likelihood) and behavioural expectations (e.g., parameter signs), and distributes rewards across the sequence of actions to learn which modelling decisions lead to well-performing candidates. We evaluate Delphos on both simulated and empirical datasets, varying the size of the modelling space and the reward function. To assess the agent's performance in navigating the model space, we analyse the learning curve, the distribution of Q-values, occupancy metrics, and Pareto fronts. Our results show that the agent learns to adaptively explore strategies to identify well-performing models across search spaces, even without prior domain knowledge. It efficiently explores large modelling spaces, concentrates its search in high-reward regions, and suggests candidates that define Pareto frontiers balancing model fit and behavioural plausibility. These findings highlight the potential of this novel adaptive, learning-based framework to assist in the model specification process.

The growing popularity of large language models has raised concerns regarding the potential to misuse AI-generated text (AIGT). It becomes increasingly critical to establish an excellent AIGT detection method with high generalization and robustness. However, existing methods either focus on model generalization or concentrate on robustness. The unified mechanism, to simultaneously address the challenges of generalization and robustness, is less explored. In this paper, we argue that robustness can be view as a specific form of domain shift, and empirically reveal an intrinsic mechanism for model generalization of AIGT detection task. Then, we proposed a novel AIGT detection method (DP-Net) via dynamic perturbations introduced by a reinforcement learning with elaborated reward and action. Experimentally, extensive results show that the proposed DP-Net significantly outperforms some state-of-the-art AIGT detection methods for generalization capacity in three cross-domain scenarios. Meanwhile, the DP-Net achieves best robustness under two text adversarial attacks. The code is publicly available at https://github.com/CAU-ISS-Lab/AIGT-Detection-Evade-Detection/tree/main/DP-Net.

This paper aims to develop effective A/B testing solutions across various industries, including internet companies such as Google, LinkedIn, X, and Meta, e-commerce platforms like Amazon, and two-sided marketplaces such as Airbnb. A/B testing has become the gold standard in these companies for policy evaluation and product deployment. For example, on traditional portal websites, it is common to assess a new version of a webpage (B) against the existing one (A) by randomly assigning visitors to either variant and then comparing an outcome of interest - such as the click through rate - to determine whether B outperforms A. A motivating application considered in this paper is the development of A/B testing solutions for large-scale fleet management in ride-sharing platforms, such as Uber and Lyft in the United States, and Didi Chuxing in China. The widespread adoption of smartphones and ride-sharing apps has enabled these companies to revolutionize and reshape urban transportation (Alonso-Mora et al., 2017; Hagiu and Wright, 2019). Ride-sharing platform is a typical two-sided market that enables efficient interactions between passengers and drivers (Rysman, 2009), as well as a complex spatio-temporal ecosystem (Wang and Y ang, 2019). Specifically, the demand and supply of this two-sided market can be measured by the numbers of call orders and the total drivers' online time in a city. These variables exhibit strong temporal patterns (see Figure 1 for a visualization), and interact with each other over time and location.