Thoughts (AI & ML)
- Predict-then-Diffuse accepted at IJCNN Our new IJCNN paper shows how to make diffusion LLMs cheaper and more predictable by estimating response length before generation.
- A NGRC-based framework for control system projects Designing a controller for a real-world dynamical system is a challenge, particularly when the system is unknown, complex, chaotic, or nonlinear. We propose and demonstrate a method to completely virtualize a control system project from the real world, requiring minimum knowledge of the underlying dynamical system and displaying performance equivalent to real-world simulation-based methods.