Synapse Open Synapse
Biology

What to randomize matters: a controlled study of volume, axis and representation in low-data molecular generation

arXiv q-bio · · 1 min read

arXiv:2610.38835v1 Announce Type: cross Abstract: Chemical language models can struggle to generate new molecules when trained on small datasets, where a model may reproduce training examples rather than learn a useful generative distribution.

Latest from arXiv q-bio