Generalist Vision Language Models for Fast Radio Burst detection: a zero-shot benchmark against a specialized detector

Jul 16, 2026

Upload by Quinn

Fast Radio Bursts (FRBs) are millisecond radio transients whose automated detection typically depends on highly specialized deep learning systems trained on large, task-specific datasets. This study tests small open-weight generalist Vision-Language Models (VLMs) in a zero-shot, prompt-only setting to detect FRBs in dynamic spectra without fine-tuning or labeled examples. Using 3000 simulated L-band spectra, benchmarked with structured RFI, noise, and FRBs, the models are compared sample-by-sample with SwinYNet and evaluated under rewritten prompting for multi-class classification.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages24

Reading Time1 h 0 min

Introduction

FRBs and automated detection pipelines

Visual and multimodal deep learning approaches

Motivation for zero-shot generalist VLMs

arXiv :2607 .07382v 1 [ cs .LG] 8 Jul 2026 Generalist Vision-Language Models for Fast Radio Burst detection: a zero-shot benchmark against a specialized detector Raiff H. Santos ,∗ Amilcar R. Queiroz ,† [K. E. L. de](K. E. L. de) Farias ,‡ and Rafael A. Batista § Universidade Federal de Campina Grande, R. Apr´ıgio Veloso 882, Bairro Universit´ario, 58429-900, Campina Grande, PB, Brazil Tharcisyo S. S. Duarte ¶ Instituto de Forma¸c˜ao de Educadores, Universidade Federal do Cariri, Rua Oleg´ario Em´ıdio de Ara´ujo, s/n, Aldeota, 63260-000, Brejo Santo, CE, Brazil and Universidade Federal de Campina Grande, R. Apr´ıgio Veloso 882, Bairro Universit´ario, 58429-900, Campina Grande, PB, Brazil (Dated: July 9, 2026) Fast Radio Bursts (FRBs) are millisecond-duration radio transients whose automated detection increasingly relies on highly specialized deep learning models. These detectors achieve exceptional performance, but they require large task-specific training datasets and cannot be redefined without retraining. In this work, we evaluate whether small, open-weight, locally run generalist VisionLanguage Models (VLMs) can detect FRBs in dynamic spectra under a zero-shot, prompt-only regime, with no fine-tuning and no labeled examples, returning structured decisions with a naturallanguage justification. From a controlled set of 3000 simulated L-band dynamic spectra containing FRBs, structured Radio Frequency Interference (RFI), and noise, we draw a balanced binary benchmark of 2000 samples and compare two such VLMs (Gemma 4 2B and 4B), sample by sample, against the state-of-the-art specialized detector SwinYNet. At the default threshold, Gemma 4 2Breaches an accuracy of 93.65%, with no statistically significant difference from SwinYNet (92.90%), while showing a significantly lower false-positive rate on structured RFI (6.4% vs. 25.0%) and no false positives on pure noise. SwinYNet retains a perfect probabilistic ranking on this benchmark (ROC-AUC of 1.0000 vs. 0.9482), a ceiling that the zero-shot VLM approaches from general-purpose pretraining alone. Rewriting the prompt alone reconfigures the same models for three-class FRB/RFI/noise classification on the full set of 3000 spectra, where they reach up to 86% accuracy without a single false FRB. I. INTRODUCTION Fast Radio Bursts (FRBs) are intense radio transients with durations of only a few milliseconds and predominantly of extragalactic origin. Since the discovery of the first FRB by [1], these events have become one of the most intriguing topics in modern astrophysics and cosmology [2–7] . FRBs are characterized by their large dispersion measures (DMs) [3, 8, 9], which generally exceed the expected contribution from the Milky Way, indicating propagation through the ionized intergalactic medium (IGM) . Over the past years, hundreds of FRBs have been detected by several radio observatories, including the Parkes telescope, CHIME [10], ASKAP [11], FAST [12], and MeerKAT [13], revealing a diverse population of repeating and apparently non-repeating sources. Because of their extragalactic characteristics, they have recently been used for cosmological probes in several scenarios [14–20] . ∗ [raiffhugos@gmail.com](raiffhugos@gmail.com)[ ](raiffhugos@gmail.com)† [amilcarq@df.ufcg.edu.br](amilcarq@df.ufcg.edu.br)[ ](amilcarq@df.ufcg.edu.br)‡ [klecio.lima@uaf.ufcg.edu.br](klecio.lima@uaf.ufcg.edu.br) § [rafael.batista@uaf.ufcg.edu.br](rafael.batista@uaf.ufcg.edu.br)[ ](rafael.batista@uaf.ufcg.edu.br)¶ [tharcisyo.duarte@ufca.edu.br](tharcisyo.duarte@ufca.edu.br) Processing these streams in real-time requires automated detection pipelines. Traditional pipelines rely on matched filtering and dedispersion, which are computationally expensive [6, 21–23] . Historically, this process was dominated by CPU-based software packages like PRESTO (PulsaR Exploration and Search TOolkit) [24, 25], which established the foundational algorithms for incoherent dedispersion [2

FAQs

What is the main goal of the document’s study?

Evaluate whether generalist, open-weight Vision-Language Models can detect fast radio bursts in dynamic spectra under a zero-shot, prompt-only regime, without fine-tuning or labeled examples.

How is the benchmark for FRB detection constructed?

From 3000 simulated L-band dynamic spectra containing FRBs, structured Radio Frequency Interference (RFI), and noise, a balanced binary benchmark of 2000 samples is created for comparison, and additional prompt rewriting enables three-class classification across the full set.

How do the generalist VLMs perform compared with the specialized detector SwinYNet?

At the default threshold, Gemma 4 2B reaches 93.65% accuracy versus SwinYNet’s 92.90%, with a lower false-positive rate on structured RFI and no false positives on pure noise; however, SwinYNet maintains perfect ranking quality (ROC-AUC 1.0000).

Jul 16, 2026

Upload by Quinn