High-frequency trading (HFT) has transformed modern financial markets, making reliable short-term price forecasting models essential. In this study, we present a novel approach to mid-price forecasting using Level 1 limit order book (LOB) data from NASDAQ, focusing on 100 U.S. stocks from the S&P 500 index during the period from September to November 2022. Expanding on our previous work with Radial Basis Function Neural Networks (RBFNN), which leveraged automated feature importance techniques based on mean decrease impurity (MDI) and gradient descent (GD), we introduce the Adaptive Learning Policy Engine (ALPE) - a reinforcement learning (RL)-based agent designed for batch-free, immediate mid-price forecasting. ALPE incorporates adaptive epsilon decay to dynamically balance exploration and exploitation, outperforming a diverse range of highly effective machine learning (ML) and deep learning (DL) models in forecasting performance.

As neural computation is revolutionizing the field of Artificial Intelligence (AI), rethinking the ideal neural hardware is becoming the next frontier. Fast and reliable von Neumann architecture has been the hosting platform for neural computation. Although capable, its separation of memory and computation creates the bottleneck for the energy efficiency of neural computation, contrasting the biological brain. The question remains: how can we efficiently combine memory and computation, while exploiting the physics of the substrate, to build intelligent systems? In this thesis, I explore an alternative way with memristive devices for neural computation, where the unique physical dynamics of the devices are used for inference, learning and routing. Guided by the principles of gradient-based learning, we selected functions that need to be materialized, and analyzed connectomics principles for efficient wiring. Despite non-idealities and noise inherent in analog physics, I will provide hardware evidence of adaptability of local learning to memristive substrates, new material stacks and circuit blocks that aid in solving the credit assignment problem and efficient routing between analog crossbars for scalable architectures.

Offline preference-based reinforcement learning (PbRL) provides an effective way to overcome the challenges of designing reward and the high costs of online interaction. However, since labeling preference needs real-time human feedback, acquiring sufficient preference labels is challenging. To solve this, this paper proposes a offLine prEference-bAsed RL with high Sample Efficiency (LEASE) algorithm, where a learned transition model is leveraged to generate unlabeled preference data. Considering the pretrained reward model may generate incorrect labels for unlabeled data, we design an uncertainty-aware mechanism to ensure the performance of reward model, where only high confidence and low variance data are selected. Moreover, we provide the generalization bound of reward model to analyze the factors influencing reward accuracy, and demonstrate that the policy learned by LEASE has theoretical improvement guarantee. The developed theory is based on state-action pair, which can be easily combined with other offline algorithms. The experimental results show that LEASE can achieve comparable performance to baseline under fewer preference data without online interaction.

Multi-Agent Reinforcement Learning involves agents that learn together in a shared environment, leading to emergent dynamics sensitive to initial conditions and parameter variations. A Dynamical Systems approach, which studies the evolution of multi-component systems over time, has uncovered some of the underlying dynamics by constructing deterministic approximation models of stochastic algorithms. In this work, we demonstrate that even in the simplest case of independent Q-learning with a Boltzmann exploration policy, significant discrepancies arise between the actual algorithm and previous approximations. We elaborate why these models actually approximate interesting variants rather than the original incremental algorithm. To explain the discrepancies, we introduce a new discrete-time approximation model that explicitly accounts for agents' update frequencies within the learning process and show that its dynamics fundamentally differ from the simplified dynamics of prior models. We illustrate the usefulness of our approach by applying it to the question of spontaneous cooperation in social dilemmas, specifically the Prisoner's Dilemma as the simplest case study. We identify conditions under which the learning behaviour appears as long-term stable cooperation from an external perspective. However, our model shows that this behaviour is merely a metastable transient phase and not a true equilibrium, making it exploitable. We further exemplify how specific parameter settings can significantly exacerbate the moving target problem in independent learning. Through a systematic analysis of our model, we show that increasing the discount factor induces oscillations, preventing convergence to a joint policy. These oscillations arise from a supercritical Neimark-Sacker bifurcation, which transforms the unique stable fixed point into an unstable focus surrounded by a stable limit cycle.

Q-learning is a powerful tool for network control and policy optimization in wireless networks, but it struggles with large state spaces. Recent advancements, like multi-environment mixed Q-learning (MEMQ), improves performance and reduces complexity by integrating multiple Q-learning algorithms across multiple related environments so-called digital cousins. However, MEMQ is designed for centralized single-agent networks and is not suitable for decentralized or multi-agent networks. To address this challenge, we propose a novel multi-agent MEMQ algorithm for partially decentralized wireless networks with multiple mobile transmitters (TXs) and base stations (BSs), where TXs do not have access to each other's states and actions. In uncoordinated states, TXs act independently to minimize their individual costs. In coordinated states, TXs use a Bayesian approach to estimate the joint state based on local observations and share limited information with leader TX to minimize joint cost. The cost of information sharing scales linearly with the number of TXs and is independent of the joint state-action space size. The proposed scheme is 50% faster than centralized MEMQ with only a 20% increase in average policy error (APE) and is 25% faster than several advanced decentralized Q-learning algorithms with 40% less APE. The convergence of the algorithm is also demonstrated.

Most multi-agent reinforcement learning (MARL) methods are limited in the scale of problems they can handle. With increasing numbers of agents, the number of training iterations required to find the optimal behaviors increases exponentially due to the exponentially growing joint state and action spaces. This paper tackles this limitation by introducing a scalable MARL method called Distributed multi-Agent Reinforcement Learning with One-hop Neighbors (DARL1N). DARL1N is an off-policy actor-critic method that addresses the curse of dimensionality by restricting information exchanges among the agents to one-hop neighbors when representing value and policy functions. Each agent optimizes its value and policy functions over a one-hop neighborhood, significantly reducing the learning complexity, yet maintaining expressiveness by training with varying neighbor numbers and states. This structure allows us to formulate a distributed learning framework to further speed up the training procedure. Distributed computing systems, however, contain straggler compute nodes, which are slow or unresponsive due to communication bottlenecks, software or hardware problems. To mitigate the detrimental straggler effect, we introduce a novel coded distributed learning architecture, which leverages coding theory to improve the resilience of the learning system to stragglers. Comprehensive experiments show that DARL1N significantly reduces training time without sacrificing policy quality and is scalable as the number of agents increases. Moreover, the coded distributed learning architecture improves training efficiency in the presence of stragglers.

In Reinforcement Learning (RL) theory, we impose restrictive assumptions to design an algorithm with provably sublinear regret. Common assumptions, like linear or RKHS models, and Gaussian or log-concave posteriors over the models, do not explain practical success of RL across a wider range of distributions and models. Thus, we study how to design RL algorithms with sublinear regret for isoperimetric distributions, specifically the ones satisfying the Log-Sobolev Inequality (LSI). LSI distributions include the standard setups of RL and others, such as many non-log-concave and perturbed distributions. First, we show that the Posterior Sampling-based RL (PSRL) yields sublinear regret if the data distributions satisfy LSI under some mild additional assumptions. Also, when we cannot compute or sample from an exact posterior, we propose a Langevin sampling-based algorithm design: LaPSRL. We show that LaPSRL achieves order optimal regret and subquadratic complexity per episode. Finally, we deploy LaPSRL with a Langevin sampler -- SARAH-LD, and test it for different bandit and MDP environments. Experimental results validate the generality of LaPSRL across environments and its competitive performance with respect to the baselines.

In the context of inverse reinforcement learning (IRL) with a single expert, adversarial inverse reinforcement learning (AIRL) serves as a foundational approach to providing comprehensive and transferable task descriptions. However, AIRL faces practical performance challenges, primarily stemming from the framework's overly idealized decomposability condition, the unclear proof regarding the potential equilibrium in reward recovery, or questionable robustness in high-dimensional environments. This paper revisits AIRL in \textbf{high-dimensional scenarios where the state space tends to infinity}. Specifically, we first establish a necessary and sufficient condition for reward transferability by examining the rank of the matrix derived from subtracting the identity matrix from the transition matrix. Furthermore, leveraging random matrix theory, we analyze the spectral distribution of this matrix, demonstrating that our rank criterion holds with high probability even when the transition matrices are unobservable. This suggests that the limitations on transfer are not inherent to the AIRL framework itself, but are instead related to the training variance of the reinforcement learning algorithms employed within it. Based on this insight, we propose a hybrid framework that integrates on-policy proximal policy optimization in the source environment with off-policy soft actor-critic in the target environment, leading to significant improvements in reward transfer effectiveness.

This work presents a novel communication framework for decentralized multi-agent systems operating in dynamic network environments. Integrated into a multi-agent reinforcement learning system, the framework is designed to enhance decision-making by optimizing the network's collective knowledge through efficient communication. Key contributions include adapting a static network packet-routing scenario to a dynamic setting with node failures, incorporating a graph attention network layer in a recurrent message-passing framework, and introducing a multi-round communication targeting mechanism. This approach enables an attention-based aggregation mechanism to be successfully trained within a sparse-reward, dynamic network packet-routing environment using only reinforcement learning. Experimental results show improvements in routing performance, including a 9.5 percent increase in average rewards and a 6.4 percent reduction in communication overhead compared to a baseline system. The study also examines the ethical and legal implications of deploying such systems in critical infrastructure and military contexts, identifies current limitations, and suggests potential directions for future research.

This paper investigates a novel nonlinear update rule based on temporal difference (TD) errors in reinforcement learning (RL). The update rule in the standard RL states that the TD error is linearly proportional to the degree of updates, treating all rewards equally without no bias. On the other hand, the recent biological studies revealed that there are nonlinearities in the TD error and the degree of updates, biasing policies optimistic or pessimistic. Such biases in learning due to nonlinearities are expected to be useful and intentionally leftover features in biological learning. Therefore, this research explores a theoretical framework that can leverage the nonlinearity between the degree of the update and TD errors. To this end, we focus on a control as inference framework, since it is known as a generalized formulation encompassing various RL and optimal control methods. In particular, we investigate the uncomputable nonlinear term needed to be approximately excluded in the derivation of the standard RL from control as inference. By analyzing it, Weber-Fechner law (WFL) is found, namely, perception (a.k.a. the degree of updates) in response to stimulus change (a.k.a. TD error) is attenuated by increase in the stimulus intensity (a.k.a. the value function). To numerically reveal the utilities of WFL on RL, we then propose a practical implementation using a reward-punishment framework and modifying the definition of optimality. Analysis of this implementation reveals that two utilities can be expected i) to increase rewards to a certain level early, and ii) to sufficiently suppress punishment. We finally investigate and discuss the expected utilities through simulations and robot experiments. As a result, the proposed RL algorithm with WFL shows the expected utilities that accelerate the reward-maximizing startup and continue to suppress punishments during learning.