Hey there, I'm Jaron, a PhD candidate part of the DTAI research group at KU Leuven, supervised by Luc De Raedt. I do research in the intersection of probabilistic reasoning and deep learning (neurosymbolic AI).
| This summer I'll be on a research internship at Basis, working on a probabilistic programming library called Weighted. | Jun 25 |
| Next semester I'll be at UCLA to visit Prof. Guy Van den Broeck. | Feb 25 |
| "Extracting Finite State Machines from Transformers" was accepted at the Workshop on Mechanistic Interpretability at ICML2024.paper | Jun 24 |
| Attended the AI winter school at Paderborn University. | Feb 24 |
| Attended the DeepLearn 2023 Winter school. | Jan 23 |
| I received a prize from IBM at the NLC2CMD competition at NeurIPS20, for my work on program synthesis with LLMs during a research internship at Bell Labs. | Dec 20 |
| Embeddings as Probabilistic Equivalence in Logic Programs Proposes a distribution semantics of logic programming using probabilistic equivalence instead of probabilistic facts, leading to an end-to-end differentiable prover without local minima. |
NeurIPS25 |
| KLay: Accelerating Arithmetic Circuits for Neurosymbolic AI Introduces knowledge layers (KLay), a new GPU-optimized library for evaluating sparse arithmetic circuits, achieving speedups of multiple orders of magnitude over existing methods. papervideocode |
ICLR25 |
| The Gradient of Algebraic Model Counting (Oral, top 4.6%) Extends algebraic model counting from inference to learning by generalizing gradients and backpropagation to different semirings, unifying various learning algorithms in neurosymbolic AI. papervideocode |
AAAI25 |
| On the Hardness of Probabilistic Neurosymbolic Learning Studies the computational complexity of differentiating probabilistic reasoning and introduces WeightME, an unbiased gradient estimator with provides probabilistic guarantees. papervideocode |
ICML24 |
| Soft-Unification in Deep Probabilistic Programming Introduces DeepSoftLog, a principled probabilistic framework for soft-unification that addresses limitations in previous neural theorem provers and enables end-to-end differentiable logic rule learning. papervideocode
|
NeurIPS23 |
| I gave a talk a the StarAI lab at UCLA. | Nov 24 |
| Presented two posters at the first NeSy conference. | Sep 24 |
| I gave an invited seminar at the TU Wien Institute of Logic and Computation. | Jul 24 |
| "Soft-Unification in Deep Probabilistic Logic", talk at the Generative NeSy Workshop. | May 24 |
| "Neurosymbolic Learning, a Probabilistic Journey", DTAI seminar.slides | |
| "AI and your Research", Invited talk (non-technical) at the LC&Y institute on the use of AI in research.video | Mar 24 |
| Presented a poster at the Flanders AI research day 2023. | Nov 23 |
| Sam McManagan. "A Neurosymbolic Approach to Solving Referring Expression Comprehension Tasks" | 2025 |
| Andrei-Bogdan Florea. "Think before you Learn: Image Segmentation with Weak Supervision" | 2025 |
| Andres Van Schel. "An Evaluation of Current Mechanistic Interpretability Techniques on an Entailment Prediction Task in Propositional Logic with BERT" | 2025 |
| Wout Seynaeve. "Learning from Logical Constraints: A Unifying approach for Weakly Supervised Semantic Segmentation" | 2025 |
| Rik Adriaensen. "Extracting Finite State Machines from Transformers" | 2024 |