Tag: #dynamical-systems7 results
This position paper argues that time series modeling should shift from transformer-based architectures to a dynamical systems perspective. It emphasizes that capturing underlying physical rules and chaotic structures is essential for true out-of-domain generalization.
BEHAVE is a new AI framework that models groups of humans as complex dynamical systems. It uses kinematic micro-signals to map interaction graphs and forecast collective behavior in real-time.
Researchers propose a bidirectional conditional latent diffusion model that predicts dynamical systems both forward and backward in time. By comparing the starting state with the result of a forward-then-backward rollout, practitioners can estimate deployment-time error without ground truth, ensembles, or governing equations.
This research models latent CoT reasoning as trajectories in representation space, applying dynamical systems analysis to understand how models like CODI and COCONUT evolve. The study identifies distinct stability behaviors, offering a new framework to improve the interpretability and performance of latent reasoning methods.
New ICLR2026 paper introduces semi-analytical algorithm to construct stable/unstable manifolds of fixed points and cycles in ReLU-based RNNs. Aids explainable AI by revealing dynamical structures like basins of attraction. Key for scientific ML and dynamical systems reconstruction.
The ASIR Courage Model formalizes truth-disclosure as a state transition from suppression (S0) to expression (S1) when facilitative forces exceed inhibitory thresholds, via lambda(1+gamma)+psi > theta+phi. It extends from human truth-telling under stakes to AI systems under policy constraints and alignment filters. Feedback loops model recursive recalibration, explaining path dependence without attributing intention.
PhD student in cybersecurity seeks advice on adversarial ML, focusing on training-time attacks and test-time evasion. Background in math, proposes using differential geometry and dynamical systems. Asks for open challenges, recent math-tool papers, and resource suggestions.