Research map / Label-efficient and robust learning / Recognition and long-tail
f-Similarity Preservation Loss for Soft Labels: A Demonstration on Cross-Corpus Speech Emotion Recognition
Biqiao Zhang, Yuqing Kong, Georg Essl, Emily Mower Provost
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 f-SPL (KL) multi-task loss to cross-entropy | Conv-Pool SER · MSP->IEMOCAP Valence | UAR (%): 64.45 -> 66.01 | Improved |
| add f-SPL (KL) loss | Conv-Pool SER · MSP->IEMOCAP Activation | UAR (%): 80.79 -> 81.67 | Improved |
| add f-SPL (RKL) loss | Conv-Pool SER · MSP->IEMOCAP Activation | UAR (%): 80.79 -> 81.16 | No clear effect |
| add f-SPL (RKL) loss | Conv-Pool SER · IEMOCAP->MSP Activation | UAR (%): 72.74 -> 74.30 | Improved |
| add f-SPL loss (any f) | Conv-Pool SER · IEMOCAP->MSP Valence | UAR (%): 61.31 -> 60.81 (KL); 60.44 (JS) | Hurt |
| choice of f divergence for f-SPL (ablation) | f-SPL · all four experiments | UAR (%) | No clear effect |
| add f-SPL to cross-entropy training | Conv-Pool SER · MSP->IEMOCAP Valence and IEMOCAP->MSP Activation | stability over epochs | Improved |
Closest papers by meaning
- Embedding Transfer With Label Relaxation for Improved Metric Learning (CVPR 2021 · similarity 0.84)
- Tailor Versatile Multi-Modal Learning for Multi-Label Emotion Recognition (AAAI 2022 · similarity 0.84)
- Class2Simi: A Noise Reduction Perspective on Learning with Noisy Labels (ICML 2021 · similarity 0.84)
- Learning Cross-Modal Retrieval With Noisy Labels (CVPR 2021 · similarity 0.83)
- $\epsilon$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise (NeurIPS 2024 · similarity 0.83)
- CARAT: Contrastive Feature Reconstruction and Aggregation for Multi-Modal Multi-Label Emotion Recognition (AAAI 2024 · similarity 0.83)
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.
