Frequency-guided flow generation? No, not just frequency, but energy (Part II)
A physicist’s view of K-Flow, from continuity equations and Hamiltonian transport to renormalization-group-inspired scale evolution.
@ The Chinese University of Hong Kong
Inspired by wave-like information propagation in nature, we view intelligence as a dynamic process evolving across space and time. Our work connects AI for Science, using AI to model and accelerate scientific discovery, with Science for AI, where insights from physics and dynamical systems guide the design of intelligent models.
Latest articles from the blog
A physicist’s view of K-Flow, from continuity equations and Hamiltonian transport to renormalization-group-inspired scale evolution.
Why is energy the essence of generation? We introduce K-Flow, a generative framework that aligns the flow path with the frequency and energy of the data by modeling in the K-amplitude space.
Bringing Transformers into 3D molecule generation via inertial frame canonicalization, geometry-aware positional encoding, and discrete-continuous joint autoregressive generation
