Knob: A Physics-Inspired Gating Interface for Interpretable and Controllable Neural Dynamics

Researchers propose Knob, a new framework that applies control theory principles to neural networks by mapping gating dynamics to mechanical systems. The approach enables real-time human adjustment of AI model behavior through intuitive physical parameters like damping and frequency, offering both static and continuous processing modes.

• →Knob framework connects deep learning with classical control theory through physics-inspired gating mechanisms.
• →The system uses damping ratio and natural frequency parameters to create an intuitive interface for human operators.
• →It employs logit-level convex fusion that reduces model confidence when neural network branches produce conflicting predictions.
• →The framework supports dual-mode inference: standard processing for static tasks and state-preserving processing for continuous data streams.
• →Experiments on CIFAR-10-C validate the calibration mechanism and demonstrate consistent second-order control signatures.