Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HCFSLN: Adaptive Hyperbolic Few-Shot Learning for Multimodal Anxiety Detection

Aditya Sneh, Nilesh Kumar Sahu, Anushka Sanjay Shelke, Arya Adyasha, Haroon R. Lone

TL;DR

This work tackles anxiety detection under data scarcity by proposing HCFSLN, a multimodal few-shot learning framework that uses adaptive hyperbolic embeddings, cross-modal attention, and an adaptive gating network to fuse audio, video, and physiological signals. Embeddings are projected into a trainable hyperbolic space and classified via prototype-based distance in the Poincaré ball, guided by a combined Hyperbolic Prototypical Loss and Angular Margin Loss. The authors introduce the Multi-Modal Anxiety Dataset (M2AD) with 108 participants and benchmark against six baselines on SAD and M2AD, achieving up to 88% accuracy in 1-shot audio and outperforming baselines by about 14%. The results demonstrate that hyperbolic geometry can better capture the complex relationships in anxiety cues, enabling robust, real-time screening on accessible devices in low-resource settings.

Abstract

Anxiety disorders impact millions globally, yet traditional diagnosis relies on clinical interviews, while machine learning models struggle with overfitting due to limited data. Large-scale data collection remains costly and time-consuming, restricting accessibility. To address this, we introduce the Hyperbolic Curvature Few-Shot Learning Network (HCFSLN), a novel Few-Shot Learning (FSL) framework for multimodal anxiety detection, integrating speech, physiological signals, and video data. HCFSLN enhances feature separability through hyperbolic embeddings, cross-modal attention, and an adaptive gating network, enabling robust classification with minimal data. We collected a multimodal anxiety dataset from 108 participants and benchmarked HCFSLN against six FSL baselines, achieving 88% accuracy, outperforming the best baseline by 14%. These results highlight the effectiveness of hyperbolic space for modeling anxiety-related speech patterns and demonstrate FSL's potential for anxiety classification.

HCFSLN: Adaptive Hyperbolic Few-Shot Learning for Multimodal Anxiety Detection

TL;DR

This work tackles anxiety detection under data scarcity by proposing HCFSLN, a multimodal few-shot learning framework that uses adaptive hyperbolic embeddings, cross-modal attention, and an adaptive gating network to fuse audio, video, and physiological signals. Embeddings are projected into a trainable hyperbolic space and classified via prototype-based distance in the Poincaré ball, guided by a combined Hyperbolic Prototypical Loss and Angular Margin Loss. The authors introduce the Multi-Modal Anxiety Dataset (M2AD) with 108 participants and benchmark against six baselines on SAD and M2AD, achieving up to 88% accuracy in 1-shot audio and outperforming baselines by about 14%. The results demonstrate that hyperbolic geometry can better capture the complex relationships in anxiety cues, enabling robust, real-time screening on accessible devices in low-resource settings.

Abstract

Anxiety disorders impact millions globally, yet traditional diagnosis relies on clinical interviews, while machine learning models struggle with overfitting due to limited data. Large-scale data collection remains costly and time-consuming, restricting accessibility. To address this, we introduce the Hyperbolic Curvature Few-Shot Learning Network (HCFSLN), a novel Few-Shot Learning (FSL) framework for multimodal anxiety detection, integrating speech, physiological signals, and video data. HCFSLN enhances feature separability through hyperbolic embeddings, cross-modal attention, and an adaptive gating network, enabling robust classification with minimal data. We collected a multimodal anxiety dataset from 108 participants and benchmarked HCFSLN against six FSL baselines, achieving 88% accuracy, outperforming the best baseline by 14%. These results highlight the effectiveness of hyperbolic space for modeling anxiety-related speech patterns and demonstrate FSL's potential for anxiety classification.

Paper Structure

This paper contains 26 sections, 14 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Multimodal Feature Fusion
  5. Cross-Modal Attention
  6. Gating Network
  7. Adaptive Hyperbolic Projection
  8. Distance Computation in Hyperbolic Space
  9. Residual Hyperbolic Block
  10. Prototype Computation in Hyperbolic Space
  11. Loss Functions and Optimization
  12. Hyperbolic Prototypical Loss
  13. Angular Margin Loss
  14. Experiments
  15. Datasets
  16. ...and 11 more sections

Figures (4)

  • Figure 1: Experimental setup.
  • Figure 2: Architecture of the proposed HCFSLN framework.
  • Figure 3: t-SNE prototype visualizations of models.
  • Figure 4: Ablation results showing the effect of loss type, curvature, and angular weight on accuracy.