Ouns El Harzli — accepted AI-conference papers

2026

Decoupling Dynamical Richness from Representation Learning: Towards Practical Measurement

Yoonsoo Nam, Nayara Fonseca, Seok Hyeong Lee, Chris Mingard, Niclas Alexander Göring, Ouns El Harzli +3

ICLR 2026poster

Dynamic feature transformation (the rich regime) does not always align with predictive performance (better representation), yet accuracy is often used as a proxy for richness, limiting analysis of their relationship. We propose a computationally efficient, performance-independent metric of richness…

Cited by 0 Source ↗Scholar ↗

2026

From Neural Networks to Logical Theories: The Correspondence between Fibring Modal Logics and Fibring Neural Networks

Ouns El Harzli, Bernardo Cuenca Grau, Artur d'Avila Garcez, Ian Horrocks, Tarek R. Besold

ICLR 2026poster

Fibring of modal logics is a well-established formalism for combining countable families of modal logics into a single fibred language with common semantics, characterized by fibred models. Inspired by this formalism, fibring of neural networks was introduced as a neurosymbolic framework for combini…

Cited by 0 Source ↗Scholar ↗

2025

Bayesian Treatment of the Spectrum of the Empirical Kernel in (Sub)Linear-Width Neural Networks

Ouns El Harzli, Bernardo Cuenca Grau

ICLR 2025poster

We study Bayesian neural networks (BNNs) in the theoretical limits of infinitely increasing number of training examples, network width and input space dimension. Our findings establish new bridges between kernel-theoretic approaches and techniques derived from statistical mechanics through the corre…

Cited by 0 Source ↗PDF ↗Scholar ↗

2024

Double-Descent Curves in Neural Networks: A New Perspective Using Gaussian Processes

Ouns El Harzli, Bernardo Cuenca Grau, Guillermo Valle-Pérez, Ard A. Louis

AAAI 2024technical

Double-descent curves in neural networks describe the phenomenon that the generalisation error initially descends with increasing parameters, then grows after reaching an optimal number of parameters which is less than the number of data points, but then descends again in the overparameterized regim…

Cited by 10 Source ↗PDF ↗Scholar ↗

2023

Cardinality-Minimal Explanations for Monotonic Neural Networks

Ouns El Harzli, Bernardo Cuenca Grau, Ian Horrocks

IJCAI 2023poster

In recent years, there has been increasing interest in explanation methods for neural model predictions that offer precise formal guarantees. These include abductive (respectively, contrastive) methods, which aim to compute minimal subsets of input features that are sufficient for a given predictio…

Cited by 5 Source ↗PDF ↗Scholar ↗