Research

I describe my approach to mathematics as computational thinking: make a problem accessible through computational tools. Pure mathematics gives the application while applied mathematics gives the tool.

Publications

2026

C14

Fast Isotopy Computation for T-Curves

Z. Geiselmann, M. Joswig, L. Kastner, K. Mundinger, S. Pokutta, CS, M. Wack, M. Zimmer

C13

Neural Sum-of-Squares: Certifying the Nonnegativity of Polynomials with Transformers

N. Pelleriti, CS, S. Liu, D. Martinez-Rubio, M. Zimmer, S. Pokutta

P7

Limits of Combinatorial Patchworking

Z. Geiselmann, M. Joswig, L. Kastner, K. Mundinger, S. Pokutta, CS, M. Wack, M. Zimmer

2025

J13

Compression-aware Training of Neural Networks using Frank–Wolfe

M. Zimmer, CS, S. Pokutta

C12

Neural Discovery in Mathematics: Do Machines Dream of Colored Planes?

K. Mundinger, M. Zimmer, A. Kiem, CS, S. Pokutta

C11

An Unsure Note on an Un-Schur Problem

O. Parczyk, CS

P6

Infinitely Many Counterexamples to a Conjecture of Lovász

A. Abiad, F. Garbe, X. Povill, CS

2024

J11

Extending the Continuum of Six-Colorings

K. Mundinger, S. Pokutta, CS, M. Zimmer

C10

Categorification of Flag Algebras

A. Kiem, S. Pokutta, CS

C9

Extending the Continuum of Six-Colorings

K. Mundinger, S. Pokutta, CS, M. Zimmer

C8

The 4-color Ramsey Multiplicity of Triangles

A. Kiem, S. Pokutta, CS

C7

Sparse Model Soups: A Recipe for Improved Pruning Via Model Averaging

M. Zimmer, CS, S. Pokutta

P5

Forcing Graphs to be Forcing

A. Kiem, O. Parczyk, CS

2023

C6

How I Learned to Stop Worrying and Love Retraining

M. Zimmer, CS, S. Pokutta

P4

PERP: Rethinking the Prune-Retrain Paradigm in the Era of LLMs

M. Zimmer, M. Andoni, CS, S. Pokutta

2022

J10

Another note on intervals in the Hales–Jewett theorem

N. Kamčev, CS

C3

New Ramsey Multiplicity Bounds and Search Heuristics

O. Parczyk, S. Pokutta, CS, T. Szabó

2021

2020

J6

A step beyond Freĭman’s theorem for set addition modulo a prime

P. Candela, O. Serra, CS

J5

On a problem of Sárközy and Sós for multivariate linear forms

J. Rué, CS

P3

Projection-Free Adaptive Gradients for Large-Scale Optimization

C. Combettes, CS, S. Pokutta

P2

Deep Neural Network Training with Frank–Wolfe

S. Pokutta, CS, M. Zimmer

2019

J4

On the optimality of the uniform random strategy

C. Kusch, J. Rué, CS, T. Szabó

P1

Independent Chains in Acyclic Posets

N. Salia, CS, C. Tompkins, O. Zamora

2018

2017

J1

A note on sparse supersaturation and extremal results for linear homogeneous systems

CS

37 entries

Talks

2026

Coloring the Plane with Neural Networks

IBS ECOPRO Seminar Institute for Basic Science (IBS) Daejeon

2025

2024

2023

The four-color Ramsey multiplicity of triangles

UPC LIMDA Joint Seminar Universitat Politècnica de Catalunya Barcelona

2022

New Ramsey multiplicity bounds and search heuristics

UPC LIMDA Joint Seminar Universitat Politècnica de Catalunya Barcelona

2019

The probabilistic intuition in positional games

TU Ilmenau Research Seminar on Combinatorics Technische Universität Ilmenau

2018

Going beyond 2.4 in Freĭman’s 2.4k-theorem

GRAPHS@IMPA Seminar Instituto de Matemática Pura e Aplicada (IMPA) Rio de Janeiro

2017

Sparse supersaturation and extremal results for linear homogeneous systems

UPC LIMDA Joint Seminar Universitat Politècnica de Catalunya Barcelona

Random strategies are nearly optimal for generalized van der Waerden games

UPC LIMDA Joint Seminar Universitat Politècnica de Catalunya Barcelona

2016

Threshold functions for systems of equations in random sets

UPC LIMDA Joint Seminar Universitat Politècnica de Catalunya Barcelona

2015

54 entries

Projects

Jan 2024 – Dec 2025

MATH+ Excellence Cluster & MaRDI AA5-9

LEAN on Me

Formalizes results from combinatorics and optimization in Lean and develops proof tactics and course material for formal verification.

Jun 2023 – May 2027

BMFTR 16IS23025B

AI4Forest

Develops deep-learning methods for global forest monitoring: 10 m canopy-height mapping, biomass estimation, and disturbance detection.

Oct 2022 – Sep 2025

MATH+ Excellence Cluster EF1-21

Scaling up Flag Algebras in Combinatorics

Scales Razborov's flag algebras through symmetry reduction and extends them to new settings such as Rado multiplicity in vector spaces.

Jan 2021 – Dec 2022

MATH+ Excellence Cluster EF1-9

Adaptive Algorithms through Machine Learning

Uses machine learning to adapt branch-and-cut solvers at run time, learning cut selection, branching, and heuristic scheduling.