Institution profile

Eindhoven University of Technology

Academic institutioneurope · nl
Official website
Research library819linked papers
Opportunities0open roles
Selected work

Representative Papers

Pruning Small Pre-Trained Weights Irreversibly and Monotonically Impairs"Difficult"Downstream Tasks in LLMs

Sep 29, 2023

This work challenges the prevailing assumption that small-magnitude weights in large language models (LLMs) are redundant, proposing instead the “Junk DNA Hypothesis”: low-magnitude weights encode essential knowledge for solving difficult downstream tasks. Method: We conduct systematic magnitude-based pruning—both structured and unstructured—alongside multi-granularity task difficulty quantification (e.g., reasoning depth, distribution shift, generalization gap), validated across model scales (7B–70B) and diverse benchmarks (MMLU, GSM8K, HumanEval). Contribution/Results: Pruning induces irreversible, monotonic performance degradation strictly correlated with task difficulty—degradation persists even after extensive fine-tuning—whereas quantization exhibits no such effect. This is the first study to empirically establish the functional necessity of small-magnitude weights from a task-difficulty perspective. We further propose novel, quantifiable cross-task difficulty metrics and demonstrate a strong negative correlation between optimal pruning ratio and task difficulty.

7 citations1 influentialRead paper

Task Adaptation from Skills: Information Geometry, Disentanglement, and New Objectives for Unsupervised Reinforcement Learning

Jun 12, 2025International Conference on Learning Representations

Unsupervised reinforcement learning (URL) aims to acquire transferable skills for unknown downstream tasks, yet existing mutual information-based skill learning (MISL) lacks theoretical characterization of skill transferability. We identify that skill diversity and separability are essential for efficient downstream policy initialization—properties not guaranteed by MISL. To address this, we propose two novel objectives—WSEP and PWSEP—grounded in Wasserstein geometry, along with a decoupling-aware metric, LSEPIN. Crucially, we establish the first theoretical link between Wasserstein distance and downstream adaptation cost, rigorously proving that our framework ensures complete discovery of optimal initial policies. Experiments demonstrate significant improvements in zero-shot transfer performance across multiple benchmarks, consistently outperforming MISL. Our method yields more disentangled skill representations and superior policy pretraining, enabling more effective downstream adaptation.

7 citationsRead paper

A Complete Axiomatization of Branching Bisimilarity for a Simple Process Language with Probabilistic Choice - (Extended Abstract)

Nov 04, 2019The Art of Modelling Computational Systems

This paper addresses the challenge of axiomatizing behavioral equivalence for process languages featuring both nondeterministic and probabilistic choice. We introduce the first sound and complete equational axiomatization for branching bisimilarity and its rooted variant—rooted branching probabilistic bisimilarity. Our method integrates structural operational semantics with probabilistic concurrency models to construct an equational proof system that supports mechanical verification of probabilistic behavioral equivalence. The key contribution is the first complete characterization of branching bisimulation for a basic, recursion-free process language with probabilistic choice—overcoming a longstanding limitation of classical process algebraic axiomatizations, which traditionally cannot accommodate probabilistic choice. This work bridges the theoretical gap between probabilistic semantics and standard axiomatic methods, thereby establishing a formal foundation for equivalence analysis in probabilistic concurrent systems.

5 citations1 influentialRead paper

Lower Bounds for Dominating Set in Ball Graphs and for Weighted Dominating Set in Unit-Ball Graphs

May 10, 2020Treewidth, Kernels, and Algorithms

This study investigates the computational complexity of the Dominating Set problem and its variants—such as Connected Dominating Set and Steiner Tree—in ball graphs and unit ball graphs. Under the Exponential Time Hypothesis (ETH), the authors establish fine-grained reductions to prove that Dominating Set in three-dimensional non-unit ball graphs and Weighted Dominating Set in unit ball graphs both admit no subexponential-time algorithms running in $2^{o(n)}$ time. This work presents the first $2^{o(n)}$-time lower bounds for several classical covering problems in these geometric graph models, highlighting a fundamental distinction in algorithmic tractability between unit and non-unit ball graphs and ruling out the existence of efficient subexponential algorithms for these problems.

4 citations2 influentialRead paper

How evaluation choices distort the outcome of generative drug discovery

Dec 24, 2024Journal of Cheminformatics

Generative molecular design lacks standardized evaluation protocols, leading to unreliable benchmarking and inaccurate prospective screening. Method: Through systematic analysis of ~1 billion molecules, we identify library size as a critical confounding factor—common metrics (e.g., uniqueness, distributional similarity) exhibit strong size-dependent bias, yielding misleading conclusions. To address this, we propose computationally efficient, scale-invariant evaluation metrics; establish a robust framework for model comparison; provide practical guidelines for prospective molecular screening; and formally characterize the fundamental divergence between deep generative modeling objectives and drug discovery requirements under diversity constraints. Results: Empirical validation demonstrates that our new metrics substantially improve evaluation stability and reproducibility, offering a reliable, standardized assessment paradigm for generative drug discovery.

3 citationsRead paper
Recent publications

Latest Papers