A universal compression theory for lottery ticket hypothesis and neural scaling laws
Published in International Conference on Learning Representations, 2026
Proves that generic permutation-invariant functions can be compressed to polylogarithmic size, yielding a theoretical foundation for the dynamical lottery ticket hypothesis and faster neural scaling laws.
Recommended citation: Hong-Yi Wang, Di Luo, Tomaso Poggio, Isaac L. Chuang, and Liu Ziyin. "A universal compression theory for lottery ticket hypothesis and neural scaling laws." International Conference on Learning Representations, pp. 115846-115871 (2026).
Download Paper
