This paper investigates the application of collective emergence theory to multi-agent reinforcement learning (MARL). The core claim is that leveraging principles of collective emergence – specifically localized rules and nonlinear interactions – can guide multi-agent systems to spontaneously generate novel behaviors and problem-solving capabilities within complex environments. Traditional MARL often focuses on optimizing individual agent performance, potentially limiting the system's overall intelligence. This work proposes a framework where the environment and reward functions are designed to explicitly encourage emergent group behavior. The key mechanism involves translating the theoretical concepts of "local rules" and "nonlinear interactions" into practical design choices for MARL. We explore how these interactions can lead to emergent coordination and innovative solutions, representing a shift towards understanding and harnessing the power of collective intelligence in reinforcement learning. The results, though theoretical in this initial presentation, demonstrate a potential pathway for developing more robust and adaptive multi-agent systems.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper explores a novel approach to discovering and validating fundamental physical laws using Multi-Agent Reinforcement Learning (MARL). The core idea is to leverage the interactions and competition among multiple agents within a simulated physical environment to autonomously identify and verify potential laws governing the system. Each agent is trained using a reinforcement learning algorithm, receiving rewards for behaviors that align with observed physical patterns. This approach offers a dynamic and automated method for uncovering hidden relationships and constraints, potentially leading to the discovery of new physical laws or the refinement of existing ones. The presented framework emphasizes exploration and utilizes agent interactions to drive the learning process. This research addresses the limitations of traditional physics discovery methods by introducing a system capable of adapting to complex environments and identifying patterns that might be missed by human observation. The key innovations lie in the application of MARL to this domain and the design of a reward structure that effectively guides agents toward discovering physical laws. ---
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper introduces Algorithmic Topology, a novel approach to computational geometry centered on the design of self-consistent, self-repairing topology algorithms. Traditional methods often struggle with complex constructions, requiring significant manual intervention. Our work proposes a paradigm shift towards adaptive reinforcement learning, enabling the algorithm to dynamically refine its topology, minimizing human effort. We define a framework for constructing geometric structures through a process of continuous refinement, mimicking evolutionary processes. The core claim is that this approach offers a significant advancement in computational geometry, allowing for the creation of arbitrarily shaped structures with minimal human input. We detail the algorithm's components, the underlying reinforcement learning mechanism, and the theoretical foundations supporting its efficacy. This research explores the potential of this method for generating intricate designs, providing a foundation for automated design and optimization within computational geometry.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper introduces a novel reinforcement learning algorithm, termed Adaptive Topological Network Structures (ATNS), designed to enhance network structure optimization within complex environments. Traditional reinforcement learning often relies on predefined reward functions, limiting adaptability. The algorithm leverages the inherent topology of a dynamically adjusted network as a core reward signal, enabling automated network restructuring. This approach promises improved performance compared to existing methods, particularly in environments with intricate dependencies and variable rewards. The core mechanism centers around a feedback loop that dynamically adjusts the network topology based on observed rewards, fostering a more robust and adaptable learning process. We present a comprehensive evaluation of the algorithm across a range of simulated scenarios, demonstrating its effectiveness in optimizing network configurations for specific tasks.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
Reinforcement learning (RL) has demonstrated remarkable success in various domains, yet its application is often hindered by the computational complexity associated with large state spaces. Traditional dynamic programming algorithms, such as Value Iteration and Policy Iteration, suffer severely from the curse of dimensionality, rendering them impractical for problems with a vast number of states. This paper proposes a novel approach to address this challenge by integrating hierarchical dynamic programming with learned abstraction layers. We decompose the state space into a hierarchy of abstraction levels, employing autoencoders to learn low-dimensional representations (embeddings) at each level. These embeddings facilitate efficient state aggregation and enable the application of dynamic programming within the hierarchical structure. The core idea is to reduce the problem size by representing the state space with a hierarchy of compact, learned representations, thereby mitigating the computational burden. This approach offers a scalable solution for RL problems with large state spaces, presenting a significant advancement in the field.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Quantum topology is a rapidly developing field with the potential to revolutionize quantum computing and information processing. It leverages the unique properties of spacetime to create topologically protected quantum states, offering enhanced resilience against noise and decoherence. The accurate modeling of quantum topology is crucial for its successful implementation. However, current parameter estimation techniques often rely on traditional methods, which can be computationally expensive and may not fully capture the complex dynamics governing these systems. This paper introduces a novel self-adaptive parameter adjustment strategy for quantum topology model parameters, aiming to address these limitations and significantly enhance model accuracy and performance. The proposed strategy dynamically adjusts parameters based on observed model behavior, optimizing for both fidelity and stability. We explore a novel approach leveraging a Bayesian optimization framework combined with a reinforcement learning component to iteratively refine parameter values. Our results demonstrate a substantial improvement in model accuracy, particularly in the context of generating complex topological structures, compared to static parameter settings. Furthermore, the proposed method exhibits improved robustness and adaptability to varying input conditions. The design incorporates a mechanism for self-monitoring and error correction, ensuring the model remains well-tuned over extended simulations. This work represents a significant advancement in parameter estimation techniques tailored specifically to the challenges posed by quantum topology modeling. This results in improved model performance and opens up new avenues for research and development in this exciting field.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper investigates the potential of utilizing biofeedback signals, specifically electroencephalogram (EEG) and electrocardiogram (ECG) data, as inputs to design adaptive and personalized machine learning algorithms. The core concept involves translating biofeedback signals into adjustable parameters within a machine learning model. This process leverages reinforcement learning or genetic algorithms to iteratively optimize these parameters, thereby enhancing the algorithm's performance and tailoring it to individual user characteristics. The research aims to address the limitations of traditional machine learning approaches by incorporating real-time physiological data, leading to more dynamic and effective algorithms. The presented framework offers a novel approach to algorithm design, promising improved efficiency and accuracy in various applications. This work focuses solely on the theoretical design and conceptual implementation, without experimental validation.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
ABSTRACT Background Traditional rehabilitation medicine, primarily dependent on qualitative clinical assessment and static therapeutic protocols, faces significant challenges in scalability, objectivity, and dynamic adaptability. The integration of Artificial Intelligence (AI) is catalyzing a paradigm shift from “experience‐driven” to “data‐driven” precision rehabilitation. Objective This narrative review delineates the current landscape of AI innovations in rehabilitation, evaluates their clinical integration across the patient lifecycle, and identifies the socio‐technical barriers to widespread adoption. Methods We narratively synthesized recent advancements in four foundational technological pillars: Computer Vision (CV) for markerless motion capture, Reinforcement Learning (RL) for intention‐aware robotics, Digital Twins (DT) for prognostic simulation, and Explainable AI (XAI) for clinical decision support. Results Our analysis reveals that AI‐driven models enhance rehabilitative efficiency by providing highly objective functional assessments, demonstrating high accuracy in specific controlled validation datasets. Clinical evidence suggests that AI‐integrated interventions can potentially reduce certain motor recovery cycles by up to 20%–30% through real‐time assist‐as‐needed (AAN) paradigms. Furthermore, the deployment of AI‐mediated remote monitoring and virtual assistants has demonstrated up to a 25% relative improvement in patient adherence post‐discharge based on selected pilot studies, effectively bridging the “rehabilitation gap” between hospital and home. Conclusion While AI offers transformative potential for personalized and accessible care, its maturation depends on overcoming challenges related to data heterogeneity, algorithmic “black‐box” distrust, and systemic interoperability. We propose a multidisciplinary roadmap to establish unified regulatory frameworks and standardized APIs. Ultimately, the transition to AI‐augmented rehabilitation is highly promising for achieving equitable and evidence‐based functional recovery in the era of digital medicine.
Hao Wu, Yali Yang, Li Wang et al.· Journal of Evaluation in Cli...· 0 citations
We document Organismo Vivo, a multi-agent trading framework combining reinforcement learning agents with encoder-based regime detection, validated through walk-forward testing, preregistration, and adversarial auditing.
Augusto Toso· Zenodo (CERN European Organi...· 0 citations
This paper explores a novel approach to program self-optimization, termed "Entropy-Driven Program Self-Optimization (EDPSO)." The core concept revolves around a program's ability to monitor and adapt its own execution based on the inherent information entropy within its processes. The system employs an "entropy-aware" module to continuously measure the entropy during program execution. This entropy value then acts as a feedback signal, guiding the adjustment of the program's code structure. Genetic algorithms and/or reinforcement learning are utilized to optimize the code, driven by the dynamic entropy feedback. The key innovation lies in the program's intrinsic understanding of its own performance, shifting away from solely relying on external metrics. The proposed EDPSO framework presents a potentially powerful methodology for enhancing program efficiency and adaptability, particularly in complex and evolving environments. This paper details the architecture, the entropy measurement methodology, and the optimization algorithms utilized within EDPSO, outlining a pathway towards truly self-optimizing software.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper introduces Dynamic Topological Dependency Learning (TDTL), a novel approach to knowledge representation and reasoning that leverages reinforcement learning to dynamically construct and adapt an internal knowledge graph based on observed data and inference results. Unlike traditional methods that rely on pre-defined topologies or manual annotation, TDTL autonomously learns data dependencies through a 'topological optimization' algorithm. The system begins with a simple, undirected graph and iteratively modifies it based on prediction errors, guided by a reward signal. A 'topological regularization' mechanism is incorporated to prevent over-complexity and maintain graph connectivity. The core claim is that a system can automatically build and adjust its internal knowledge representation's topology to reflect dynamic data dependencies without pre-defined topologies or manual labels. This represents a significant departure from existing knowledge graph construction and reasoning techniques. ---
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Automated theorem proving (ATP) aims to develop systems capable of mechanically proving mathematical theorems. Despite significant advancements, ATP systems often struggle with complex reasoning tasks, largely due to the inherent difficulty in representing and executing logical deduction rules. This work proposes a novel approach to ATP that integrates neural networks to provide guidance during the proof process. The core idea is to train a neural network to suggest promising proof steps and identify relevant theorems, essentially acting as an "intelligent assistant" for the ATP system. This guidance mechanism is expected to improve the efficiency and effectiveness of ATP, particularly in tackling challenging mathematical problems. The presented framework utilizes a reinforcement learning approach, where the neural network learns to optimize the proof strategy based on the current state of the proof and the available theorems. The system is evaluated conceptually, outlining the architecture and training process, and highlighting potential improvements. Further research will focus on developing and refining the network architecture, exploring different training strategies, and integrating the guidance mechanism with existing ATP systems.
Jincheng Zhang· 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.
