![[2602.19385] Adaptive Data Augmentation with Multi-armed Bandit: Sample-Efficient Embedding Calibration for Implicit Pattern Recognition](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.19385] Adaptive Data Augmentation with Multi-armed Bandit: Sample-Efficient Embedding Calibration for Implicit Pattern Recognition
Summary
The paper presents ADAMAB, a novel framework for efficient embedding calibration in few-shot pattern recognition, leveraging adaptive data augmentation through a Multi-Armed Bandit approach.
Why It Matters
This research addresses the challenges of recognizing implicit patterns in AI, particularly in scenarios with limited training data. By improving sample efficiency and reducing computational costs, ADAMAB has the potential to enhance the performance of AI models in real-world applications, making it relevant for both researchers and practitioners in the field of machine learning and computer vision.
Key Takeaways
- ADAMAB offers a sample-efficient method for embedding calibration in few-shot learning.
- The framework utilizes a Multi-Armed Bandit strategy for adaptive data augmentation.
- Significant accuracy improvements (up to 40%) can be achieved with minimal training data.
- The approach mitigates computational overhead while maintaining model performance.
- The research contributes to overcoming limitations of pre-trained models in pattern recognition tasks.
Computer Science > Computer Vision and Pattern Recognition arXiv:2602.19385 (cs) [Submitted on 22 Feb 2026] Title:Adaptive Data Augmentation with Multi-armed Bandit: Sample-Efficient Embedding Calibration for Implicit Pattern Recognition Authors:Minxue Tang, Yangyang Yu, Aolin Ding, Maziyar Baran Pouyan, Taha Belkhouja Yujia Bao View a PDF of the paper titled Adaptive Data Augmentation with Multi-armed Bandit: Sample-Efficient Embedding Calibration for Implicit Pattern Recognition, by Minxue Tang and 4 other authors View PDF HTML (experimental) Abstract:Recognizing implicit visual and textual patterns is essential in many real-world applications of modern AI. However, tackling long-tail pattern recognition tasks remains challenging for current pre-trained foundation models such as LLMs and VLMs. While finetuning pre-trained models can improve accuracy in recognizing implicit patterns, it is usually infeasible due to a lack of training data and high computational overhead. In this paper, we propose ADAMAB, an efficient embedding calibration framework for few-shot pattern recognition. To maximally reduce the computational costs, ADAMAB trains embedder-agnostic light-weight calibrators on top of fixed embedding models without accessing their parameters. To mitigate the need for large-scale training data, we introduce an adaptive data augmentation strategy based on the Multi-Armed Bandit (MAB) mechanism. With a modified upper confidence bound algorithm, ADAMAB diminishes the g...
