Bishal Ranjan Swain

Coupling Liquid Time-Constant Encoders with Modern Hopfield Memory

Continuous-time neural networks provide adaptive dynamics, but rely on a single hidden state to encode both fast input fluctuations and longer-term context. This shared representation forces rapidly changing inputs to overwrite slower contextual signals, causing the model to lose past information as