Landscape Learning for Neural Network Inversion
This addresses a key bottleneck in inverse problems for computer vision, robotics, and graphics, offering a significant speedup.
The paper tackles the problem of unstable and slow neural network inversion by learning a loss landscape that makes gradient descent efficient, resulting in massive improvement and acceleration across tasks like GAN inversion and 3D pose reconstruction.
Many machine learning methods operate by inverting a neural network at inference time, which has become a popular technique for solving inverse problems in computer vision, robotics, and graphics. However, these methods often involve gradient descent through a highly non-convex loss landscape, causing the optimization process to be unstable and slow. We introduce a method that learns a loss landscape where gradient descent is efficient, bringing massive improvement and acceleration to the inversion process. We demonstrate this advantage on a number of methods for both generative and discriminative tasks, including GAN inversion, adversarial defense, and 3D human pose reconstruction.