Adaptive Test-Time Scaling for Zero-Shot Respiratory Audio Classification

2026-04-14

AI Summary

This work addresses the challenges of low computational efficiency, scarce labeled data, and high expert annotation costs in zero-shot respiratory sound classification by proposing TRIAGE, a novel framework that introduces test-time adaptive computation for the first time. TRIAGE dynamically routes input audio to one of three inference pathways—lightweight label cosine scoring, structured clinical description matching, or retrieval-augmented large language model reasoning—based on confidence-based routing, enabling on-demand resource allocation. Requiring no task-specific training, TRIAGE achieves an average AUROC of 0.744 across nine classification tasks, with nearly half of the samples classified at the lowest computational tier. It improves uncertainty-aware sample identification accuracy by up to 19% and matches or surpasses supervised baselines in overall performance.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Automated respiratory audio analysis promises scalable, non-invasive disease screening, yet progress is limited by scarce labeled data and costly expert annotation. Zero-shot inference eliminates task-specific supervision, but existing methods apply uniform computation to every input regardless of difficulty. We introduce TRIAGE, a tiered zero-shot framework that adaptively scales test-time compute by routing each audio sample through progressively richer reasoning stages: fast label-cosine scoring in a joint audio-text embedding space (Tier-L), structured matching with clinician-style descriptors (Tier-M), and retrieval-augmented large language model reasoning (Tier-H). A confidence-based router finalizes easy predictions early while allocating additional computation to ambiguous inputs, enabling nearly half of all samples to exit at the cheapest tier. Across nine respiratory classification tasks without task-specific training, TRIAGE achieves a mean AUROC of 0.744, outperforming prior zero-shot methods and matching or exceeding supervised baselines on multiple tasks. Our analysis show that test-time scaling concentrates gains where they matter: uncertain cases see up to 19% relative improvement while confident predictions remain unchanged at minimal cost.

Problem

• zero-shot classification
• respiratory audio analysis
• test-time scaling
• adaptive computation
• audio classification

Innovation

• adaptive test-time scaling
• zero-shot learning
• tiered inference
• audio-text embedding
• retrieval-augmented reasoning