Antonin Sulc, Ph.D.

Senior Researcher in neuro-symbolic AI and scientific machine learning

Download CV Email me
Portrait of Antonin Sulc on a Seattle ferry at sunset

Recent highlights

About

I am Antonin Sulc (Antonín Šulc), a Senior Researcher at Innovatrics, and I remain affiliated with Lawrence Berkeley National Laboratory (LBNL) as a Researcher.

Previously I was a Senior Data Scientist at the Helmholtz Association (DESY/XFEL), where I worked on applying language models to scientific data. I hold a Ph.D. in Computer Vision from the University of Konstanz, and my background is in signal processing and multi-modal data.

My current work centres on Modal Logical Neural Networks, a family of models that make modal logic (necessity, possibility, knowledge, belief, obligation) differentiable, with applications to multi-agent systems, quantum circuits and finance. The open-source library torchmodal implements them for PyTorch.

Read a short interview about my work and background: 3Q4 with Antonin Sulc (ATAP, LBNL).

Software & Open Source

metalsq

Few-shot meta-learning with a closed-form inner loop: each task is one regularised least-squares solve over a frozen random Fourier basis, and only the bandwidth, ridge and derivative weight are meta-learned. Derivative, operator and integral observations enter the same solve with closed-form derivatives and no autograd graph.

enn-public

Code for deep evidential networks with differentiable belief-function layers: every mass in the pipeline is a valid Dempster–Shafer mass function, combined with Yager's rule, so the mass on the ignorance set serves as an "I do not know" signal.

torchbelief

Differentiable belief functions and imprecise probabilities for PyTorch. It fuses sources that disagree (Dempster, Yager, TBM, Dubois–Prade, PCR, cautious rules), assigns blame to the faulty one, and decides or abstains when the evidence doesn't single out an answer. It includes evidential layers and losses that can sit inside a network.

event2vector

A scikit-learn style Python library for learning composable, interpretable representations of discrete event sequences. Published in PMLR Vol. 282 (NeurReps 2025).

torchmodal

A PyTorch framework for Differentiable Modal Logic, used to train Modal Logical Neural Networks (MLNNs).

FastLSQ

A lightweight framework for solving PDEs in one shot via Fourier features with exact analytical derivatives. It bypasses automatic differentiation and iterative training loops for linear and nonlinear problems.

PACuna

Automated data collection and fine-tuning pipelines for adapting large language models to particle-accelerator literature.

Publications

Showing 23 of 23 publications

Meta-LSQ: A Closed-Form Sobolev Inner Solver for Few-Shot Meta-Learning

A. Sulc

To appear in Proceedings of the 8th Northern Lights Deep Learning Conference (NLDL 2027), PMLR Vol. 352

arXiv:2611.24718 (2026), NLDL 2027 proceedings to follow

Few-shot meta-learning in which the inner loop is one ridge solve over a random Fourier basis. Only three scalars are meta-learned, and derivative or operator observations enter the same solve in closed form.

Neural Networks with Differentiable Belief-Function Layers

A. Sulc

Published in Belief Functions: Theory and Applications, Springer Nature Switzerland

Springer, Cham, pp. 61–71, 2027

Deep evidential networks built entirely from Dempster–Shafer mass functions combined with Yager's rule, whose mass on ignorance flags contested inputs and ranks human disagreement without ever seeing annotator distributions.

Towards Unlocking Insights from Logbooks Using AI

A. Sulc, G. Hartmann, J. Maldonado, V. Kain et al.

15th International Particle Accelerator Conference (IPAC'24)

A multi-laboratory study of retrieval-augmented generation for making accelerator logbooks searchable and useful.

Depth from Spectral Defocus Blur

S. Ishihara, A. Sulc, I. Sato

IEEE International Conference on Image Processing (ICIP), 2019

Estimates depth from the way defocus blur differs across spectral bands.

Talks & Media

Talks

  • Modal Logic Neural Networks

    Oral talk, 20th International Conference on Neurosymbolic Learning and Reasoning (NeSy 2026), Lisbon

  • Differentiable Modal Logic for Multi-Agent Diagnosis, Orchestration and Communication

    Oral talk, 3rd International Conference on Neuro-Symbolic Systems (NeuS 2026), University of Southern California

  • FastLSQ: Solving PDEs in One Shot

    ICLR 2026 Workshop on AI & PDE

  • Modal Logical Neural Networks for Financial AI

    ICLR 2026 Workshop on Financial AI (FinAI)

  • Towards Monocular Shape from Refraction

    Oral talk, British Machine Vision Conference (BMVC 2021)

Press & interviews

Contact

For collaborations, invited talks or press enquiries, email me or connect on LinkedIn.

Email

Antonin Sulc, interactive 3D portrait

Move the pointer, drag or use the arrow keys to look around; scroll or pinch to step closer. A 3D Gaussian-splat scene predicted from one photo with Apple's SHARP.