I put a neural network in the Linux kernel scheduler

An experimental implementation of a neural network-based Linux kernel scheduler that predicts task time slices to optimize system responsiveness. The scheduler uses a deadline-based approach where a PyTorch-trained model predicts future time slices based on task name and priority. Training data comes from scheduling traces during WebGL aquarium benchmarks with stress testing. The implementation uses Rust with the torch crate and demonstrates improved performance, achieving 60 fps compared to 30-40 fps with the default scheduler, despite the overhead of running neural network inference in the kernel.