Eleonora Bergamin | Searcharxiv

arXiv2025

Preliminary analysis of Sus scrofa movement using Hidden Markov Models and Networks

This study examines the complex movement patterns and behavioral characteristics of wild boars using GPS telemetry data collected over a two-month period. Our methodological approach centers on the application of a Hidden Markov Model (HMM) to discern distinct behavioral states embedded within the trajectories. Furthermore, the study aimed to construct behavioral networks, derived from these segmented trajectories. The resultant network structures showed that the hidden behavioral patterns are mostly independent of geographical locations. While most locations have many behaviors occuring in them, our findings also suggest that Finally, the research incorporates a spatial trajectory analysis, complemented by raster data validation, to potentially delineate areas acting as repellents within the ecological context of Hainich National Park in Germany.

physics.soc-ph↗

arXiv2025

Information-theoretic reduction of deep neural networks to linear models in the overparametrized proportional regime

We rigorously analyse fully-trained neural networks of arbitrary depth in the Bayesian optimal setting in the so-called proportional scaling regime where the number of training samples and width of the input and all inner layers diverge proportionally. We prove an information-theoretic equivalence between the Bayesian deep neural network model trained from data generated by a teacher with matching architecture, and a simpler model of optimal inference in a generalized linear model. This equivalence enables us to compute the optimal generalization error for deep neural networks in this regime. We thus prove the "deep Gaussian equivalence principle" conjectured in Cui et al. (2023) (arXiv:2302.00375). Our result highlights that in order to escape this "trivialisation" of deep neural networks (in the sense of reduction to a linear model) happening in the strongly overparametrized proportional regime, models trained from much more data have to be considered.

math.ST↗