project High-Risk AI system requirements evaluations We analyze the EU AI Act to derive software-lifecycle-inspired requirements for high-risk AI applications and evaluate ML libraries' contribution to fulfilling them. Multimodal Imitation Learning for Stock Trading We use optimally traded financial transactions in transformer-encoded time-series data and images to train an adversarial inverse reinforcement learning model that outputs robust trading reward functions. Bayesian NeRF Training Modeling scene-centered images as mixture of Gaussian distributions, we limit the number of training observations required for neural radiance field training by selecting to minimize uncertainty. Quantum-enhanced video generative model We encode visual properties of source videos in entangled quantum state whose readout properties are used to cut and combine the videos into one.