Optimizing the Noise in Self-Supervised Learning: from Importance Sampling to Noise-Contrastive Estimation. (arXiv:2301.09696v1 [stat.ML]) ift.tt/cCYSUxq

Annealing/tempering are common ways to speed up sampling/optimization, but I was always a bit mystified how they help with *statistical* issues (i.e. sample complexity). New paper with talented (masters!!) student Yilong Qin: arxiv.org/abs/2306.09332

🚀 Deadline extended to the 3rd of October for the Workshop on Optimal Transport and Machine Learning @Neurips 2023! Amazing lineup of speakers ! PS: recent submissions to neurips/iclr/aistats are welcome Site: otmlworkshop.github.io #Monge David Alvarez Melis Charlotte Bunne marco cuturi

Happening this morning at poster #1223!

A Practical Diffusion Path for Sampling ift.tt/ICUBFo3