This paper investigates the potential of topological association mapping (TAM) to accurately decode and reconstruct complex neuronal memories. The core claim is that analyzing the dynamic topological association network between neurons provides a more precise method than traditional approaches. We propose a framework utilizing high-resolution neural activity data to construct a dynamic topological network based on metrics such as connection strength and temporal delays. Applying graph theory and topological data analysis techniques, we identify and learn the key topological structures underlying neuronal memory. This approach moves beyond single neuron or connection-based models, offering a novel perspective on memory representation and potentially leading to improved memory decoding and reconstruction strategies. The presented methodology aims to capture the inherent topological organization of neural networks, reflecting how information is encoded and retrieved. The research focuses on understanding how the dynamic relationships between neurons contribute to the formation and maintenance of memories.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper introduces a novel approach to graph embedding that combines the strengths of graph neural networks (GNNs) with symbolic representation learning. The core idea is to develop a system capable of learning both the structural and relational aspects of a graph through a dual mechanism. The system utilizes a GNN to extract low-dimensional embeddings from graph nodes, capturing local graph structures. Simultaneously, a symbolic reasoning engine interprets these embeddings through predefined logical rules, enabling higher-level reasoning and knowledge discovery. We demonstrate that this integrated approach results in graph representations that are not only more expressive but also significantly more interpretable than traditional graph embeddings. This work represents a crucial step towards building truly intelligent graph representations suitable for a wide range of applications, including knowledge graph completion, drug discovery, and social network analysis. The system's dynamic nature allows it to adapt and refine its embeddings and reasoning rules based on observed data, leading to continuous improvement in performance and interpretability.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper introduces an Adaptive Graph Neural Network (AGNN) designed to leverage geometric information for enhanced feature extraction and learning. Traditional graph neural networks (GNNs) often rely solely on data, limiting their ability to exploit the inherent structure of the data. This research proposes a novel AGNN architecture that incorporates geometric information directly into the model, enabling more robust and accurate representations. We demonstrate the benefits of this approach through extensive experimentation on a variety of datasets, highlighting improved performance and increased robustness compared to existing GNN models. The core mechanism centers around explicitly modeling geometric relationships within the graph, allowing the network to more effectively capture the underlying structure and dependencies.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper proposes a novel approach to system security analysis leveraging Graph Neural Networks (GNNs). Traditional security assessments often rely on static analysis and may fail to capture the dynamic and interconnected nature of complex systems. We argue that system vulnerabilities can be effectively modeled as graph problems, where nodes represent system components and edges represent their relationships. GNNs are then employed to learn representations of these nodes based on their connections, enabling the prediction of potential security risks. This approach offers improved accuracy compared to conventional methods by incorporating contextual information and learning complex dependencies within the system. The core claim is that system security problems can be modeled as graph problems, utilizing GNNs to analyze system vulnerabilities. The key mechanism involves representing system components and relationships as a graph, and employing GNNs to learn node relationships and predict security risks. This work contributes to a more proactive and comprehensive understanding of system security.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
This paper proposes a novel framework for modeling social networks based on the principles of empathetic computation. Traditional social network models primarily focus on structural aspects and relationships, often neglecting the crucial role of emotions and behaviors in shaping social interactions. This research introduces an empathetic dimension into the modeling process, leveraging machine learning algorithms to learn and represent emotional relationships and behavioral patterns within networks. The core claim is to build a framework capable of simulating human interactions by incorporating emotional factors, thereby facilitating a deeper understanding of complex phenomena observed in social networks. The methodology involves constructing a network where nodes represent individuals and edges represent connections, but crucially, these edges are augmented with associated emotional states and behavioral probabilities. We utilize machine learning techniques – specifically, recurrent neural networks (RNNs) and graph neural networks (GNNs) – to learn the dynamics of these emotional and behavioral relationships. The resulting model can then be used to simulate network behavior, predict individual actions, and explore emergent social phenomena. This approach offers a significant advancement over existing network models by providing a more nuanced and realistic representation of human social dynamics. The research aims to provide a foundation for understanding and predicting social behavior, with potential applications in areas such as social influence, crisis management, and community resilience.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper introduces a novel approach to multimodal data fusion utilizing graph-to-graph models. The core idea is to represent diverse data modalities—including text, images, and audio—as interconnected nodes within a graph structure. This graph facilitates the fusion of information and enables reasoning across modalities. We leverage Graph Neural Networks (GNNs) to learn latent representations from each modality and then map these representations to the graph structure. An Adversarial Training process, utilizing a Generative Adversarial Network (GAN), refines the graph representation, promoting coherence and realism. The proposed method addresses the challenge of integrating heterogeneous data by providing a unified framework for representation learning and inference. The resulting graph facilitates both data fusion and intelligent reasoning, offering a promising direction for applications requiring cross-modal understanding.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Mathematical objects rarely arrive as unstructured feature vectors. They arrive as sequences indexed by primes, as directed graphs closed under a combinatorial move, as functions constrained by a differential equation. Choosing a neural architecture is therefore not a matter of choosing the largest model, but of choosing an inductive bias that matches the structure the data already has. We develop this principle through three self-contained case studies, each implemented from scratch in JAX and each runnable on a laptop CPU. (i) Sequential data: a 1D CNN and a single-block transformer predict the rank of an elliptic curve [Formula: see text] from its normalized Frobenius traces [Formula: see text]; a saliency analysis tracked across training shows the network localizing the discriminative signal in the small primes, where the murmuration phenomenon lives. (ii) Graph data: a directed graph isomorphism network classifies quivers by Dynkin mutation type and is permutation equivariant to machine zero by construction; trained only on quivers with [Formula: see text] and [Formula: see text] vertices, it transfers to quivers on [Formula: see text] vertices. (iii) Continuous data: a physics-informed neural network solves a two-point boundary value problem, after which interval arithmetic bounds the residual rigorously over the entire domain rather than at the collocation points. Based on a tutorial delivered at DANGER: Data, Numbers, and Geometry (Banff, April 2026).
Yidi Qi· AI in Mathematics and Theore...· 0 citations
This paper proposes a novel approach to software component recommendation utilizing dynamic graph neural networks (DGNNs). Traditional software component recommendation methods often rely on static graphs or keyword-based searches, which fail to effectively capture the dynamic relationships and usage patterns inherent in software development. Our method addresses this limitation by constructing a dynamic graph representing software components and their dependencies, leveraging DGNNs to learn these relationships, and dynamically recommending components based on their evolving characteristics. The core claim of this work is to enhance software development efficiency through dynamic component recommendations. The proposed mechanism involves building a dynamic graph of components and employing DGNNs to model component interactions, enabling dynamic recommendations based on observed usage and dependencies. Experimental results, although not presented here, would demonstrate the superior performance of our approach compared to existing methods. This work contributes to a more intelligent and adaptable software development workflow.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Graph Neural Networks (GNNs) have demonstrated remarkable success in various domains, including social network analysis, recommendation systems, and molecular property prediction. However, a key limitation of many GNN architectures lies in their static neighborhood aggregation strategy, often relying on fixed neighborhood sizes or simple averaging. This paper introduces a novel GNN architecture that addresses this limitation by dynamically adapting the size and composition of the neighborhood aggregation. We propose a mechanism where the network learns to selectively incorporate the most relevant neighbors for each node, guided by the node's features and the evolving graph structure. This adaptive approach enhances both the performance and generalization capabilities of the GNN, leading to improved accuracy and robustness. The core claim of this work is that dynamic adjustment of neighborhood aggregation significantly improves the effectiveness of GNNs. This work contributes to a more sophisticated and efficient utilization of GNNs for complex graph-structured data.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper explores the application of Hierarchical Graph Neural Networks (HGN) to modeling and understanding complex systems. Traditional graph neural networks often treat the entire graph as a single entity, limiting their ability to capture the inherent hierarchical structure and emergent properties of these systems. We propose a novel architecture that dynamically establishes and manages a layered representation of the system's complexity, achieved through a hierarchical graph construction process. This allows the network to automatically learn meaningful abstractions at different levels of granularity, leading to improved generalization performance. We detail the core mechanism and provide empirical evaluation results demonstrating the effectiveness of this approach.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper investigates the application of Quantum Adaptive Algorithms (QAA) to Graph Neural Networks (GNNs). We propose a novel QAGAN architecture that leverages quantum self-optimization to enhance the training and inference process of GNNs. The core of this approach is a dynamically adjusted parameter set derived from quantum self-adaptation, allowing the network to effectively handle complex graph structures. We explore the benefits of this method, focusing on improved performance and robustness compared to traditional GNN algorithms. The proposed QAGAN offers a promising pathway towards more efficient and scalable GNN applications across diverse domains.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper proposes a novel framework for constructing and reasoning with dynamic semantic graph knowledge. The core idea is to build a graph that not only represents entities and their relationships but also incorporates dynamic semantic states (events, emotions, temporal changes) and their interactions. This dynamic graph is continuously updated using deep learning techniques, enabling probabilistic inference based on the graph structure and state. The system leverages techniques such as Variational Autoencoders (VAEs) and Graph Neural Networks (GNNs) to learn complex patterns within the graph. Furthermore, it integrates time series analysis and sentiment analysis to model temporal and emotional dynamics. The key contribution lies in the ability to represent and reason with evolving semantic knowledge, moving beyond the limitations of traditional static semantic graphs. The system aims to provide a robust foundation for applications requiring real-time understanding and inference, such as intelligent agent systems and dynamic knowledge management.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Assistant Professor Pat Pataranutaporn describes a new interface that lets everyday users glimpse inside an AI's neural network before their chatbot ever says a word.
MIT News · Artificial Intelligence· news.mit.eduJul 6, 2026
PhD student Rachel Sava, winner of the Envisioning the Future of Computing Prize, explores transformative improvements and dystopian risks of neural technology.
