Chih-Chun Wang

Mathematical physics researcher at LMU Munich

I am a researcher in mathematical physics at LMU Munich, and my own research is in the mathematics of equivariant machine learning. I am applying for PhD positions starting in autumn 2027.

Most of what I do starts from one question: when the data has a symmetry, what does that symmetry force a model to look like? In a recent preprint, I show that no single polynomially parameterized equivariant architecture can represent all equivariant functions representable by an unconstrained multi-head self-attention layer. In other words, for attention the usual constructions are not universal, and the argument for that is algebraic.

On the physics side, I work on the foundations of density functional theory, which asks what a functional can represent once its constraints are fixed. The two subjects sound unrelated, and both of them ask which objects a parameterized family can represent.