Adaptive gates in CLIP-style few-shot prompt learning often collapse due to gradient magnitude imbalance and gate degradation, failing to beat fixed prompts.
Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
When Adaptation Fails: A Gradient-Based Diagnosis of Collapsed Gating in Vision-Language Prompt Learning
Adaptive gates in CLIP-style few-shot prompt learning often collapse due to gradient magnitude imbalance and gate degradation, failing to beat fixed prompts.