Neural Hamiltonian

Neural Hamiltonian methods integrate principles of Hamiltonian dynamics into neural network architectures, primarily aiming to improve model robustness, interpretability, and efficiency in tasks like graph neural networks (GNNs) and generative modeling. Current research focuses on applying these methods to enhance GNN performance in adversarial settings and improve node embedding by learning underlying graph manifolds, as well as developing more interpretable and computationally efficient Hamiltonian flows for normalizing flows. These advancements hold promise for improving the reliability and explainability of machine learning models across various applications, particularly in areas requiring robust performance and interpretable results.

Papers

November 27, 2023

A Graph Neural Network-Based QUBO-Formulated Hamiltonian-Inspired Loss Function for Combinatorial Optimization using Reinforcement Learning
Redwan Ahmed Rizvee, Raheeb Hassan, Md. Mosaddek Khan
Reinforcement Learning Loss Function Reward Function Graph Neural Quadratic Unconstrained Binary Quantum Optimization Neural Hamiltonian

October 10, 2023

Adversarial Robustness in Graph Neural Networks: A Hamiltonian Approach
Kai Zhao, Qiyu Kang, Yang Song, Rui She, Sijie Wang, Wee Peng Tay
Graph Neural Network Adversarial Attack Adversarial Robustness Adversarial Perturbation Energy Based GFlowNets Hamiltonian Formulation Neural Hamiltonian

May 30, 2023

Node Embedding from Neural Hamiltonian Orbits in Graph Neural Networks
Qiyu Kang, Kai Zhao, Yang Song, Sijie Wang, Wee Peng Tay
Graph Neural Network Node Classification Graph Node Graph Datasets Neural Hamiltonian

February 3, 2023

Fixed-kinetic Neural Hamiltonian Flows for enhanced interpretability and reduced complexity
Vincent Souveton, Arnaud Guillin, Jens Jasche, Guilhem Lavaux, Manon Michel
Generative Model Normalizing Flow Low Complexity Better Interpretability Hamiltonian Flow Neural Hamiltonian

Neural Hamiltonian

Papers

A Graph Neural Network-Based QUBO-Formulated Hamiltonian-Inspired Loss Function for Combinatorial Optimization using Reinforcement Learning

Adversarial Robustness in Graph Neural Networks: A Hamiltonian Approach

Node Embedding from Neural Hamiltonian Orbits in Graph Neural Networks

Fixed-kinetic Neural Hamiltonian Flows for enhanced interpretability and reduced complexity