Research map / Label-efficient and robust learning / Knowledge distillation
Entropy-Monitored Kernelized Token Distillation for Audio-Visual Compression
Hyoungseob Park, Lipeng Ke, Pritish Mohapatra, Huajun Ying, sankar venkataraman, Alex Wong
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add kernelized token distillation (KTD) to KD | ViT-Tiny student · VGGSound | accuracy (%): 56.1 -> 61.4 | Improved |
| add entropy-monitored modality regularization to KTD + KD (ablation) | ViT-Tiny student · VGGSound | accuracy (%): 61.4 -> 62.0 | Improved |
| add entropy monitor to Hinton KD (ablation) | ViT-Tiny student · VGGSound | accuracy (%): 56.1 -> 59.9 | Improved |
| use RBF kernel instead of linear kernel (ablation) | KTD · VGGSound | accuracy (%): 60.2 -> 61.4 | Improved |
| use RBF kernel (gamma 0.5) instead of order-2 polynomial (ablation) | KTD · VGGSound | accuracy (%): 60.5 -> 61.4 | Improved |
| use KTD with linear kernel instead of MTST (softmax + masking) | KD-based distillation · VGGSound | accuracy (%): 57.6 -> 60.2 | Improved |
| distill the last layer instead of the first layer (ablation) | KTD · VGGSound | accuracy (%): 56.4 -> 60.2 | Improved |
| use instance-wise KTD instead of batch-wise (ablation) | KTD · VGGSound | accuracy (%): 59.7 -> 61.2 | Improved |
| add cross-modal kernelized token loss (ablation) | KTD · VGGSound | accuracy (%): 61.2 -> 61.0 | Hurt |
| distill all modalities instead of vision tokens only (ablation) | KTD + KD · VGGSound | accuracy (%): 54.5 -> 61.2 | Improved |
| use 1-layer linear entropy monitor instead of 3-layer MLP (ablation) | EM-KTD + KD · VGGSound | accuracy (%): 61.7 -> 62.0 | No clear effect |
| use sliding window kernelization (ablation) | EM-KTD + KD · VGGSound | accuracy (%): 61.9 -> 62.0 | No clear effect |
Closest papers by meaning
- From Vision to Audio and Beyond: A Unified Model for Audio-Visual Representation and Generation (ICML 2024 · similarity 0.87)
- CAV-MAE Sync: Improving Contrastive Audio-Visual Mask Autoencoders via Fine-Grained Alignment (CVPR 2025 · similarity 0.87)
- AVOC: Enhancing Hour-Level Audio-Video Understanding in Omni-Modal LLMs via Retrieval-Inspired Token Compression (NeurIPS 2026 · similarity 0.86)
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.86)
- Separating the "Chirp" from the "Chat": Self-supervised Visual Grounding of Sound and Language (CVPR 2024 · similarity 0.86)
- X-AVDD: Cross-Attentive Audio-Visual Dataset Distillation (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
