This paper tackles dataset distillation by giving a clear, math-backed way to keep only the most useful bits of data, so models can learn well from far fewer images.

#dataset distillation#data condensation#Shapley value

Next-Embedding Prediction Makes Strong Vision Learners

Beginner

Sihan Xu, Ziqiao Ma et al.Dec 18arXiv

This paper introduces NEPA, a very simple way to teach vision models by having them predict the next patch’s embedding in an image sequence, just like language models predict the next word.

#self-supervised learning#vision transformer#autoregression

RecTok: Reconstruction Distillation along Rectified Flow

Intermediate

Qingyu Shi, Size Wu et al.Dec 15arXiv

RecTok is a new visual tokenizer that teaches the whole training path of a diffusion model (the forward flow) to be smart about image meaning, not just the starting latent features.

#Rectified Flow#Flow Matching#Visual Tokenizer