Search papers, labs, and topics across Lattice.
This study utilizes Neural Ordinary Differential Equations (Neural-ODEs) with strategically chosen equilibrium points to enhance classification tasks by creating planted attractors that represent target classes. The method leverages the universal approximation properties of Neural-ODEs to shape the velocity field, effectively guiding inputs through defined basins of attraction towards their respective classes. The results demonstrate improved classification performance by effectively directing inputs to their corresponding targets, showcasing the potential of dynamic systems in machine learning applications.
Planted attractors in Neural-ODEs can transform classification tasks by directing inputs to their target classes through dynamically shaped velocity fields.
In this work, Neural ODEs equipped with a curated collection of equilibrium points have been successfully employed for classification tasks.The planted attractors serve as indicators for the target classes, while the velocity field leveraging the universal approximation capabilities of the architecture shapes the dynamical landscape.This process defines the basins of attraction of the trained model, effectively directing each input provided as an initial condition toward its corresponding destination target.
