Christopher J. Hillar

Mathematician  ·  Algebraic

About

I work at the intersection of pure mathematics, computer science, AI, and theoretical neuroscience.

I earned my Ph.D. in Mathematics from U.C. Berkeley with an NSF Graduate Fellowship under Bernd Sturmfels, following a B.S. in Mathematics and Computer Science from Yale. I had an NSF Postdoctoral Fellowship under Frank Sottile at Texas A&M. I was a project scientist at the Redwood Center for Theoretical Neuroscience (U.C. Berkeley) and worked with the Tecott Lab at U.C. San Francisco. I co-founded the company Awecom.

Research

Interests

A single thread runs through my work: finding the structure that makes hard representation problems tractable — from solving structured polynomial systems to exploring how the retina codes the world.

Selected work

Journal of the London Mathematical Society · 2007

Solvability of symmetric word equations in positive definite letters

with Scott Armstrong

Let S(X, B) be a symmetric (‘palindromic’) word in two letters X and B. A theorem due to Hillar and Johnson states that for each pair of positive definite matrices B and P, there is a positive definite solution X to the word equation S(X, B)=P. They also conjectured that these solutions are finite and unique. In this paper, we resolve a modified version of this conjecture by showing that the Brouwer degree of such an equation is equal to 1 (in the case of real matrices). It follows that, generically, the number of solutions is odd (and thus finite) in the real case. Our approach allows us to address the more subtle question of uniqueness by exhibiting equations with multiple real solutions, as well as providing a second proof of the result of Hillar and Johnson in the real case.

Proceedings of the American Mathematical Society · 2008

An elementary and constructive solution to Hilbert's 17th Problem for matrices

with Jiawang Nie

We give a short and elementary proof of a theorem of Procesi, Schacher and (independently) Gondard, Ribenboim that generalizes a famous result of Artin.

Bulletin of the London Mathematical Society · 2012

Equations solvable by radicals in a uniquely divisible group

with Lionel Levine amd Darren Rhea

We study equations in groups G with unique mth roots for each positive integer m. We obtain the first known infinite families of word equations not solvable by radicals, and conjecture a complete classification.

Advances in Mathematics · 2012

Finite Gröbner bases in infinite dimensional polynomial rings and applications

with Seth Sullivant

We prove the Independent Set Conjecture in algebraic statistics.

Journal of the Association for Computing Machinery · 2013

Most tensor problems are NP-hard

with Lek-Heng Lim

The natural generalizations of matrix problems to tensors are, almost without exception, computationally intractable.

Advanced Studies in Pure Mathematics · 2018

Equivariant Gröbner bases

with Robert Krone and Anton Leykin

Algorithmic computation in polynomial rings is a classical topic in mathematics. However, little attention has been given to the case of rings with an infinite number of variables until recently when theoretical efforts have made possible the development of effective routines. Ability to compute relies on finite generation up to symmetry for ideals invariant under a large group or monoid action, such as the permutations of the natural numbers. We summarize the current state of theory and applications for equivariant Gröbner bases, develop several algorithms to compute them, showcase our software implementation, and close with several open problems and computational challenges.

arXiv · 2018

Maximum entropy distributions on graphs

with Andre Wibisono

Inspired by applications to theories of coding and communication in networks of nervous tissue, we study maximum entropy distributions on weighted graphs with a given expected degree sequence.

Journal of Mathematical Neuroscience · 2018

Robust exponential memory in Hopfield networks

with Ngoc Tran

Shows how a recurrent network can store an exponential number of memories robustly, which also solves the hidden clique problem in computer science.

Nature Scientific Reports · 2018

Active state organization of spontaneous behavioral patterns

with Giorgio Onnis, Darren Rhea, and Laurence Tecott

Mouse genetics are 99% predicted by behavior alone.

IEEE Transactions Signal Processing · 2019

On the uniqueness and stability of dictionaries for sparse representation of noisy signals

with Charles Garfinkle

We provide very general conditions guaranteeing when dictionaries yielding the sparsest encodings are unique and stable with respect to measurement or modeling error.

ICLR · 2023

Bispectral neural networks

with Sophia Sanborn, Christian Shewmake, and Bruno Olshausen

We present a neural network architecture, Bispectral Neural Networks (BNNs), for learning representations that are invariant to the actions of compact commutative groups.

COLT · 2024

Harmonics of learning: Universal Fourier features emerge in invariant networks

with Giovanni Luca Marchetti, Danica Kragic, and Sophia Sanborn

We formally prove that, under certain conditions, if a neural network is invariant to a finite group then its weights recover the Fourier transform on that group.

Entropy · 2025

Detecting signatures of criticality using divergence rate

with Tenzin Chan and De Wen Soh

Guided by the classical theory of rate–distortion (RD) from information theory, we propose a measure for detecting and characterizing critical phenomena from data.

Preprint · 2026

Implicit bias and invariance: How Hopfield networks efficiently learn graph orbits

with Michael Murray, Tenzin Chan, and Kedar Karhadkar

Many learning problems involve symmetries, and while invariance can be built into neural architectures, it can also emerge implicitly when training on group-structured data. We study this phenomenon in classical Hopfield-Amari networks and illustrate how they can infer the isomorphism class of a graph from a small, random sample. Our results reveal that: (i) graph isomorphism classes can be represented within a three-dimensional invariant subspace, (ii) using gradient descent to minimize energy flow (MEF) has an implicit bias toward norm-efficient solutions, which underpins a polynomial sample complexity bound for learning isomorphism classes, and (iii) across multiple learning rules, parameters converge toward the invariant subspace as sample sizes grow.

Full list

For the complete and current publication record, see my Google Scholar profile.

Talks

Selected talks

Writing

Articles & expository

Research papers, notes, and expository pieces.

The complete, always-current list lives on Google Scholar.

Code

Software & data

Reference implementations and computational supplements accompanying the papers.

Problems

Problem of the Month

An open or instructive problem, refreshed periodically — a small standing invitation to think about something hard.

Problem: Characterize those positive integers m,n such that the following holds: For every group of size |G| = m, and any g in the group, we have that g has a unique n-th root.