CoCoA-Mix: Confusion-and-Confidence-Aware Mixture Model for Context Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hong, Dasol, Lee, Wooju, Myung, Hyun
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908399407464448
author Hong, Dasol
Lee, Wooju
Myung, Hyun
author_facet Hong, Dasol
Lee, Wooju
Myung, Hyun
contents Prompt tuning, which adapts vision-language models by freezing model parameters and optimizing only the prompt, has proven effective for task-specific adaptations. The core challenge in prompt tuning is improving specialization for a specific task and generalization for unseen domains. However, frozen encoders often produce misaligned features, leading to confusion between classes and limiting specialization. To overcome this issue, we propose a confusion-aware loss (CoA-loss) that improves specialization by refining the decision boundaries between confusing classes. Additionally, we mathematically demonstrate that a mixture model can enhance generalization without compromising specialization. This is achieved using confidence-aware weights (CoA-weights), which adjust the weights of each prediction in the mixture model based on its confidence within the class domains. Extensive experiments show that CoCoA-Mix, a mixture model with CoA-loss and CoA-weights, outperforms state-of-the-art methods by enhancing specialization and generalization. Our code is publicly available at https://github.com/url-kaist/CoCoA-Mix.
format Preprint
id arxiv_https___arxiv_org_abs_2506_07484
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CoCoA-Mix: Confusion-and-Confidence-Aware Mixture Model for Context Optimization
Hong, Dasol
Lee, Wooju
Myung, Hyun
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
I.2.6; I.5.2
Prompt tuning, which adapts vision-language models by freezing model parameters and optimizing only the prompt, has proven effective for task-specific adaptations. The core challenge in prompt tuning is improving specialization for a specific task and generalization for unseen domains. However, frozen encoders often produce misaligned features, leading to confusion between classes and limiting specialization. To overcome this issue, we propose a confusion-aware loss (CoA-loss) that improves specialization by refining the decision boundaries between confusing classes. Additionally, we mathematically demonstrate that a mixture model can enhance generalization without compromising specialization. This is achieved using confidence-aware weights (CoA-weights), which adjust the weights of each prediction in the mixture model based on its confidence within the class domains. Extensive experiments show that CoCoA-Mix, a mixture model with CoA-loss and CoA-weights, outperforms state-of-the-art methods by enhancing specialization and generalization. Our code is publicly available at https://github.com/url-kaist/CoCoA-Mix.
title CoCoA-Mix: Confusion-and-Confidence-Aware Mixture Model for Context Optimization
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
I.2.6; I.5.2
url https://arxiv.org/abs/2506.07484