TaCo: Targeted Concept Removal in Output Embeddings for NLP via Information Theory and Explainability.
Fanny JourdanLouis BéthuneAgustin PicardLaurent RisserNicholas AsherPublished in: CoRR (2023)
Keyphrases
- information theory
- information theoretic
- statistical learning
- jensen shannon divergence
- statistical mechanics
- conditional entropy
- relative entropy
- natural language processing
- statistical physics
- kullback leibler divergence
- text mining
- shannon entropy
- question answering
- mdl principle
- video data
- dimensionality reduction
- rate distortion theory