PromptGate Client Adaptive Vision Language Gating for Open Set Federated Active Learning
This work provides a solution for resource-constrained medical institutions to deploy data-efficient AI while respecting patient privacy, specifically by improving the efficiency of active learning in open-set federated environments where out-of-distribution noise is prevalent. It is an incremental improvement to existing federated active learning pipelines.
This paper addresses the challenge of open-set federated active learning in medical AI by proposing PromptGate, a dynamic VLM-gated framework that purifies unlabeled data pools before active learning queries. PromptGate achieves this by using federated Class-Specific Context Optimization with lightweight, learnable prompt vectors that adapt a frozen BiomedCLIP backbone to local clinical domains. Experiments demonstrate that PromptGate maintains over 95% in-distribution purity with 98% out-of-distribution recall, significantly outperforming static VLM prompting which degrades to 50% ID purity.
Deploying medical AI across resource-constrained institutions demands data-efficient learning pipelines that respect patient privacy. Federated Learning (FL) enables collaborative medical AI without centralising data, yet real-world clinical pools are inherently open-set, containing out-of-distribution (OOD) noise such as imaging artifacts and wrong modalities. Standard Active Learning (AL) query strategies mistake this noise for informative samples, wasting scarce annotation budgets. We propose PromptGate, a dynamic VLM-gated framework for Open-Set Federated AL that purifies unlabeled pools before querying. PromptGate introduces a federated Class-Specific Context Optimization: lightweight, learnable prompt vectors that adapt a frozen BiomedCLIP backbone to local clinical domains and aggregate globally via FedAvg -- without sharing patient data. As new annotations arrive, prompts progressively sharpen the ID/OOD boundary, turning the VLM into a dynamic gatekeeper that is strategy-agnostic: a plug-and-play pre-selection module enhancing any downstream AL strategy. Experiments on distributed dermatology and breast imaging benchmarks show that while static VLM prompting degrades to 50% ID purity, PromptGate maintains $>$95% purity with 98% OOD recall.