An LLM-based evolutionary algorithm automatically designs training-free VLM adaptation code, improving few-shot classification accuracy over manually-designed baselines by up to 1.91 points.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
EvoVLMA: Evolutionary Vision-Language Model Adaptation
An LLM-based evolutionary algorithm automatically designs training-free VLM adaptation code, improving few-shot classification accuracy over manually-designed baselines by up to 1.91 points.