Researchers at KAIST have introduced a brain-inspired 'warmup' training method that briefly exposes AI models to random noise and labels before real data, improving the alignment between prediction ...