In-context learning lifts LLM music valence prediction from R² 0.11 to 0.43
affige_yang · x · 2026-10-29
A Taiwan-based team (Chi-En Dai, Chih-Pin Tan, Yi-Hsuan Yang) published an arXiv paper using in-context learning (ICL) to improve continuous valence and arousal prediction for music.
Method:
- Instead of direct prediction, a small task-specific base model produces initial scores, refined via contextual demonstrations from an auxiliary dataset
- Text-based ICL uses lyrics and auto-generated audio captions; a multimodal variant adds raw audio
Results (entire MER60 dataset held out for testing):
- Text ICL with 80 shots raises valence R² from 0.110 to 0.427
- The multimodal variant reaches 0.346 with only 10 shots
Conclusion: training-free ICL is an effective inference-time strategy for music emotion prediction. Code and weights released; submitted to ICASSP 2027.
More from Research
- Researcher: one steering vector can make an RLHF model shout slurs, no prompt can match it — voooooogel · 2026-10-29
- OpenAI's math argument still unverified: no expert has publicly assessed it — maier_ak · 2026-10-29
- OpenAI's Lean-formalized conjectures await expert review; firm calls it raw data — maier_ak · 2026-10-29
- OpenAI model's Lean-verified manuscript targets local smoothing and Kakeya conjectures — maier_ak · 2026-10-29
- OpenAI model claims 59-page Lean proof of local smoothing and Kakeya in every dimension — maier_ak · 2026-10-29
- TTME beats test-time training on math discovery with 84% fewer FLOPs — rohanpaul_ai · 2026-10-29