Local Non Dissipativity

Local non-dissipativity research focuses on developing and analyzing neural network models that maintain or leverage information flow, even across long distances or in the presence of noise and uncertainty. Current efforts concentrate on designing architectures (like graph neural networks and neural ODEs) and training methods that guarantee or promote desirable properties such as stability, energy conservation, and robustness to adversarial attacks, often by incorporating concepts from dissipative systems theory and optimal control. This work is significant because it addresses limitations in existing neural network models, improving their reliability, interpretability, and applicability to complex systems in diverse fields like robotics, fluid dynamics, and materials science.