arXiv subjects

Marco Lents

Publications and source records attributed to Marco Lents.

3 recordsLinked to original sources

arXiv2026

Simple, Safe, and Overlooked: Reclaiming Sustainable Domain Generalization with Statistical Color Matching

Hardware shifts, color variations, and changing patient characteristics between development and deployment routinely break trained medical image classifiers. Existing remedies fall short: standard color jittering provides insufficient diversity, while deep generative style transfer algorithms hallucinate features, destroy clinically relevant structures, and waste massive compute resources. To address this, we revisit classical statistical color matching and repurpose it as Colorist, a highly efficient data augmentation strategy that applies global mean-standard deviation matching directly in the RGB color space. We demonstrate that this training-free, fully interpretable approach safely generates structurally intact domain variations, outperforming deep generative models in structural fidelity and color alignment. Across out-of-distribution histopathology, peripheral blood, dermatology, and retinal datasets, it improves balanced accuracy by up to +9% over state-of-the-art domain generalization regularizers and by +13% over an unaugmented baseline. Moreover, by avoiding neural networks in the augmentation loop, Colorist preserves anatomical structure, minimizes carbon footprint, and integrates seamlessly into standard dataloaders. Together, these findings establish statistical matching as a safe, interpretable, yet overlooked alternative to deep architectures for clinical robustness. Source code is available at https://github.com/sdoerrich97/colorist.

cs.CV↗

arXiv2026

Vertical Fusion: Condensing Internal Representations for Robust ViT Classification

Despite exposing rich intermediate representations, Vision Transformers (ViTs) are almost exclusively utilized as black-box feature extractors, where only the last layer is considered for downstream tasks. We challenge this convention by introducing the notion of recoverability: the capacity of intermediate representations to correct last-layer failures. By evaluating independent classification probes at every model depth across 16 datasets, we observe that intermediate probes correctly classify 18% to 76% of samples that the last-layer probe misclassifies. We show that these gains are not primarily driven by predictive diversity, but by a redundancy-correctness correspondence, where the internal hierarchy acts as a series of stable, redundant probes of a shared discriminative signal. While established horizontal ensemble strategies (i.e., across multiple models) can improve performance, they incur high computational cost and ignore this vertical signal within a single model. To bridge this gap, we propose VFusion, a principled vertical aggregation strategy employing a learnable mapping into a low-dimensional latent space that synthesizes features across the internal ViT hierarchy. VFusion substantially outperforms established aggregation baselines in both in-distribution and out-of-distribution settings, notably closing 45% of the accuracy gap between the best individual layer and a theoretical oracle performance. Our gains consistently generalize across model sizes and pre-training regimes, confirming that VFusion offers a robust and efficient alternative to horizontal ensemble methods. The code is available at https://github.com/francescodisalvo05/vit-vertical-fusion.

cs.CV↗

arXiv2025

Topological gravity for arbitrary Dyson index

We use the well established duality of topological gravity to a double scaled matrix model with the Airy spectral curve to define what we refer to as topological gravity with arbitrary Dyson index $\upbeta$ ($\upbeta$ topological gravity). On the matrix model side this is an interpolation in the Dyson index between the Wigner-Dyson universality classes, on the gravity side it can be thought of as interpolating between orientable and unorientable manifolds in the gravitational path integral, opening up the possibility to study moduli space volumes of manifolds ``in between''. Using the perturbative loop equations we study correlation functions of this theory and prove several structural properties, having clear implications for the generalised moduli space volumes. Additionally we give a geometric interpretation of these properties using the generalisation to arbitrary Dyson index of the recently found Mirzakhani-like recursion for unorientable surfaces. Using these properties, we investigate whether $\upbeta$-topological gravity is quantum chaotic in the sense of the Bohigas-Giannoni-Schmit conjecture. Along the way we answer this question for the symplectic Wigner-Dyson class, not studied in the literature yet, and establish strong evidence for quantum chaos for this version of the theory, and thus for all bosonic varieties of topological gravity. We further argue for quantum chaoticity in the general $\upbeta$ case, based on novel constraints we find to be obeyed by genuinely non-Wigner-Dyson parts of the moduli space volumes. As for the general $\upbeta$ case the universal behaviour expected from a chaotic system is not known fully analytically we give a novel way to approach it, starting with the result of $\upbeta$ topological gravity and compare the results to a numerical evaluation of the universal result.

hep-th↗