This paper proposes a novel approach to system optimization leveraging Multi-Agent Reinforcement Learning (MARL). The core concept is to decompose complex systems into multiple intelligent agents, each responsible for controlling a specific subsystem. These agents then collaboratively learn optimal control strategies through a MARL framework. Traditional system optimization methods often rely heavily on expert knowledge and manual tuning, which can be time-consuming and limited in scope. MARSL offers a dynamic and adaptive solution, allowing systems to learn and optimize themselves based on real-time feedback and interactions. This approach is particularly well-suited for systems with high dimensionality, non-linear dynamics, and complex interactions between components. The paper outlines the theoretical foundations of MARSL, details the implementation considerations, and presents a conceptual framework for its application. We aim to demonstrate that MARL can significantly enhance the optimization capabilities of complex systems compared to conventional methods.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper explores the application of neuromorphic computing principles, specifically utilizing Spike-Timing Dependent Plasticity (STDP), for enhancing reinforcement learning (RL) performance. Traditional deep learning approaches often struggle with sparse reward environments, requiring extensive training and substantial computational resources. We propose a novel framework where a reinforcement learning agent is implemented on a neuromorphic platform, leveraging STDP to dynamically adjust synaptic weights based on the precise timing of pre- and post-synaptic spikes. This approach allows the network to learn directly from the temporal structure of the environment, potentially leading to more efficient learning and improved performance in scenarios with limited or delayed rewards. The core claim is that this combination offers a significant advantage over conventional deep learning architectures, particularly in the context of sparse reward reinforcement learning. We detail the theoretical underpinnings of STDP and its relevance to RL, and outline a conceptual architecture for such a system. Future research directions are also discussed, focusing on the practical implementation and scaling of this approach.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Rooftop solar lets households generate their own electricity, but the benefits are unevenly shared and home batteries are expensive. This thesis develops a system that lets neighbours trade electricity, and reductions in their energy use, directly with one another, while sharing a common pool of storage instead of buying individual batteries. To coordinate many homes at once, it uses artificial intelligence to run the calculations far faster while keeping each household's data private. On a model network, it lowers grid reliance, eases strain on power lines, and reduces bills. This makes local clean-energy sharing cheaper, fairer, and faster.
Ezzeldin Ayman Ibrahim Ismail· Monash University· 0 citations
This paper introduces a novel approach to data structure design leveraging Self-Organizing Constraint Networks (SOCN). SOCN networks dynamically reorganize themselves based on observed data patterns, offering a highly adaptable solution for a wide range of applications. We propose a reinforcement learning framework to train an agent that learns to restructure the constraint network, resulting in a system capable of automatically optimizing data structure performance. The core mechanism focuses on learning the optimal rearrangement of constraint nodes and edges to maximize efficiency and resilience. The system's adaptability stems from its continuous learning process, making it a significant advancement over static constraint network approaches. This work presents a comprehensive investigation of the system's behavior, demonstrating its effectiveness through simulations and a preliminary experimental evaluation.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
This study proposes a multi-objective optimization framework for task allocation and path planning in transport-oriented multi-robot systems. The framework explicitly considers heterogeneous robot capabilities and load capacities while jointly minimizing task completion time and overall energy consumption. A hierarchical architecture is adopted, consisting of two stages. In the upper layer, the NSGA-II algorithm evaluates task allocation strategies and constructs a Pareto-optimal solution space, enabling decision-makers to select solutions according to optimization preferences or operational constraints. In the lower layer, deep neural networks and reinforcement learning are employed for multi-agent learning to generate collision-free paths for the assigned tasks. This hierarchical design enables capability-aware task allocation while providing flexibility to accommodate optimization priorities. Simulation and experimental results demonstrate that the proposed framework effectively addresses complex scenarios involving task dependencies, improves path-learning efficiency and task allocation performance, and provides multiple interpretable trade-off solutions without compromising single-objective performance. These results highlight the framework’s effectiveness, scalability, and practical applicability to real-world multi-robot transportation tasks.
Sheng-Hsiang Luo, Rongshun Chen· International Journal of Aut...· 0 citations
This paper investigates the development of an adaptive quantum state measurement protocol designed to enhance information extraction efficiency from quantum states while mitigating the detrimental effects of decoherence. The core idea revolves around employing a reinforcement learning algorithm to dynamically adjust the measurement basis, based on the observed quantum state characteristics and feedback signals generated during the measurement process. Real-time adjustments to measurement device parameters via quantum control techniques further optimize the measurement outcome. The proposed method addresses the limitations of traditional fixed-basis measurements, which are highly susceptible to decoherence. Through adaptive basis selection, this approach aims to maximize information extraction and minimize interference with the quantum state. The theoretical framework and underlying principles are presented, outlining the key components and their interrelationship. The potential impact of this technique on quantum information processing is discussed.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper presents a novel approach to symbolic regression, termed Dynamic Symbolic Regression with Evolutionary Reinforcement Learning (DSR-ERL). The core idea is to leverage reinforcement learning (RL) to dynamically control an evolutionary algorithm, specifically optimizing the mutation and crossover operators within the symbolic regression process. Traditional symbolic regression methods often suffer from issues like premature convergence and inefficient exploration of the expression space. Our system addresses these challenges by allowing an RL agent to learn a policy that adapts the evolutionary strategy based on the quality of the generated symbolic expressions. The agent observes the fitness (evaluation) of the expression and adjusts the evolutionary operators to favor solutions with higher fitness. We demonstrate the effectiveness of DSR-ERL through a theoretical analysis and a conceptual framework, outlining the key components and their interactions. The system offers the potential to discover more accurate and efficient mathematical models from data, particularly in complex domains. The key innovation lies in the adaptive control provided by the RL agent, leading to a dynamic and potentially superior symbolic regression process. We provide a detailed description of the system's architecture, including the RL agent, the evolutionary algorithm, and the interface between them. Furthermore, we discuss the challenges and future directions of this research.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper proposes a novel hierarchical reinforcement learning (HRL) framework that leverages intrinsic motivation, meta-learning, and a hierarchical architecture to address the limitations of traditional HRL methods concerning exploration and generalization. The core idea is to integrate these three components to create a more adaptive and efficient learning system. At each level of the hierarchy, intrinsic motivation, specifically novelty seeking, encourages exploration. Simultaneously, meta-learning dynamically adjusts the learning rate and policy updates, enabling rapid adaptation to diverse tasks. We demonstrate the effectiveness of this approach through a theoretical analysis and a conceptual framework, outlining the key components and their interactions. The resulting system offers improved learning speed and robustness compared to standard HRL techniques, particularly when facing complex and varied environments. This work lays the groundwork for future research in developing truly adaptable and intelligent hierarchical agents.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper introduces a novel self-adaptive graph theory algorithm designed to optimize the performance of machine learning algorithms through dynamic graph structure adjustment. Traditional graph theory algorithms often operate on static graphs, failing to adequately leverage the inherent variability in data and task characteristics. This algorithm employs a feedback loop that continuously analyzes and modifies the graph's structure and node connections based on observed data, resulting in enhanced machine learning capabilities. The core mechanism revolves around a reinforcement learning-inspired approach, iteratively refining the network topology to minimize loss functions and maximize model performance. The paper details the algorithm's design, implementation, and preliminary results demonstrating its effectiveness in a specific machine learning benchmark.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper presents a novel approach to designing neuro-morphic networks capable of adapting to dynamic and complex input data. The core idea revolves around mimicking the inherent plasticity and evolving connectivity observed in biological neural networks. We introduce a reinforcement learning-based algorithm that continuously monitors input data statistics and network performance, dynamically adjusting the synaptic strengths and connection weights between neurons to optimize response speed and accuracy. The system leverages biologically-inspired neuron models, such as pulse and dynamic membrane models, to simulate neuronal behavior accurately. Unlike conventional neuro-morphic networks which often rely on fixed hardware architectures, our approach focuses on software-based adaptation, offering a more flexible and efficient solution. The key innovation lies in the integration of reinforcement learning and bio-inspired models, creating a network that learns and adapts in real-time. This adaptive architecture demonstrates potential for applications in sensor data processing, robotics, and pattern recognition where input data characteristics change over time. The overall framework addresses the limitations of static neuro-morphic designs by providing a dynamic and intelligent network capable of handling unpredictable environments.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper addresses the challenge of optimizing multi-agent collaborative learning (MCL) by introducing a dynamic topology structure optimization framework. Traditional MCL methods often rely on fixed topologies, which may not be optimal for varying task demands and evolving agent states. We propose a novel approach that leverages reinforcement learning (RL) to dynamically adjust the topology of a multi-agent system, enhancing both efficiency and overall learning performance. The core idea is to model the agent-agent communication and task dependencies as a graph and use RL to learn optimal edge weights and connectivity patterns within this graph. The system adapts to changing conditions by modifying the strengths of connections between agents, adjusting communication frequencies, and potentially adding or removing connections entirely. This dynamic adjustment enables the system to focus computational resources on critical interactions and effectively distribute tasks, ultimately leading to improved convergence rates and better solutions. We outline the key components of the framework, including the state representation, action space, reward function, and the RL algorithm employed. Experimental results (simulated) demonstrate the effectiveness of the proposed approach compared to static topology MCL.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This final project examines how adaptive neurocognitive mechanisms related to threat detection, uncertainty processing, anticipation, reward learning, social comparison, belonging, metacognition, and behavioral regulation interact with characteristics of contemporary environments. The study adopts a theoretical and integrative literature review approach and proposes the Adaptive Mechanism–Environmental Amplification Model (AMEA), a hypothesis-generating conceptual framework describing how repeated environmental signals may increase the frequency, persistence, or regulatory importance of otherwise adaptive neurocognitive mechanisms. Rather than assuming that contemporary technologies or environments are intrinsically harmful, the model emphasizes the interaction among environmental architecture, perceived relevance, learning and reinforcement, individual differences, perceived control, metacognitive capacity, and behavioral flexibility. The study discusses potential pathways involving threat amplification, uncertainty-driven information seeking, reward-based checking, social comparison, external validation, belief reinforcement, and decision-making under uncertainty, while explicitly considering alternative explanations, limitations, and conditions under which the proposed model could be empirically challenged. Keywords: Neuroscience; Human Behavior; Uncertainty; Stress; Reward; Social Comparison; Belonging; Metacognition; Decision-Making; Behavioral Regulation; Digital Environments; AMEA.
Thiago Campos Silva· Zenodo (CERN European Organi...· 0 citations
A new machine-learning framework aims to improve the success rate of computational protein design while moving away from results that reproduce sequences found in nature.
MIT News · Artificial Intelligence· news.mit.eduAug 24, 2026
A new method for surgically removing training examples from a model reveals that as datasets grow, the link between what a model learns and what it produces dissolves.
