project Multimodal Inverse Reinforcement Learning for Stock Trading We use optimally traded financial transactions observed as both transformer-encoded time-series data and CNN-encoded images to train an adversarial inverse reinforcement learning (AIRL) model that outputs robust trading reward functions. Bayesian NeRF Training Designed uncertainty-minimizing data acquisition algorithm for 3D reconstruction by modeling images as Gaussian mixtures. Used posterior predictive to reduce number of training images while maintaining quality. Build a walking quadruped robot from scratch I design, manufacture, and control a four-legged robot to walk.
