CLUB: A Contrastive Log-ratio Upper Bound of Mutual Information

2020

Citations in the corpus

• Meta-Learning with Less Forgetting on Large-Scale Non-Stationary Task Distributions (2022, neutral): “However, the computation of MI values is intractable [44,18], since the form of joint probability distribution p(a,b) in Eq.”
• Disentangled Speaker and Language Representations Using Mutual Information Minimization and Domain Adaptation for Cross-Lingual TTS (2021, methodology): “Denot-ing speaker embedding and language embedding as z ( s ) and z ( l ) respectively, CLUB relies on approximating the conditional distribution q θ ( z ( l ) | z ( s ) ) to compute the upper bound.”