Model Merging Improves Zero-Shot Generalization in Bioacoustic Foundation Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Marincione, Davide, Crisostomi, Donato, Dessi, Roberto, Rodolà, Emanuele, Rossi, Emanuele
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908664667832320
author Marincione, Davide
Crisostomi, Donato
Dessi, Roberto
Rodolà, Emanuele
Rossi, Emanuele
author_facet Marincione, Davide
Crisostomi, Donato
Dessi, Roberto
Rodolà, Emanuele
Rossi, Emanuele
contents Foundation models capable of generalizing across species and tasks represent a promising new frontier in bioacoustics, with NatureLM being one of the most prominent examples. While its domain-specific fine-tuning yields strong performance on bioacoustic benchmarks, we observe that it also introduces trade-offs in instruction-following flexibility. For instance, NatureLM achieves high accuracy when prompted for either the common or scientific name individually, but its accuracy drops significantly when both are requested in a single prompt. We address this by applying a simple model merging strategy that interpolates NatureLM with its base language model, recovering instruction-following capabilities with minimal loss of domain expertise. Finally, we show that the merged model exhibits markedly stronger zero-shot generalization, achieving over a 200% relative improvement and setting a new state-of-the-art in closed-set zero-shot classification of unseen species.
format Preprint
id arxiv_https___arxiv_org_abs_2511_05171
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Model Merging Improves Zero-Shot Generalization in Bioacoustic Foundation Models
Marincione, Davide
Crisostomi, Donato
Dessi, Roberto
Rodolà, Emanuele
Rossi, Emanuele
Machine Learning
Artificial Intelligence
Sound
Foundation models capable of generalizing across species and tasks represent a promising new frontier in bioacoustics, with NatureLM being one of the most prominent examples. While its domain-specific fine-tuning yields strong performance on bioacoustic benchmarks, we observe that it also introduces trade-offs in instruction-following flexibility. For instance, NatureLM achieves high accuracy when prompted for either the common or scientific name individually, but its accuracy drops significantly when both are requested in a single prompt. We address this by applying a simple model merging strategy that interpolates NatureLM with its base language model, recovering instruction-following capabilities with minimal loss of domain expertise. Finally, we show that the merged model exhibits markedly stronger zero-shot generalization, achieving over a 200% relative improvement and setting a new state-of-the-art in closed-set zero-shot classification of unseen species.
title Model Merging Improves Zero-Shot Generalization in Bioacoustic Foundation Models
topic Machine Learning
Artificial Intelligence
Sound
url https://arxiv.org/abs/2511.05171