Labs & Research4 min read
AI Research
A transformer that unifies density and score estimation could shrink a key bottleneck across AI and science
A new transformer model called DiScoFormer estimates both the density and score of any distribution from a finite sample in a single forward pass. It beats kernel density estimation by up to 37x in density error and 6.5x in score error at 100 dimensions, and it can adapt to new distributions without retraining.
2026-07-11