The connection between human and machine learning (IMAGE)
Caption
A new connection between human and machine learning has been discovered: While conceptual regions in human cognition for long have been modelled as convex regions, Tetkova et al. present new evidence that convexity playes a similar role in AI. So-called pretraining by self-supervision leads to convexity of conceptual regions and the more convex the regions are, the better the model wil learn a given specialist task in supervised fine-tuning.
Credit
DTU
Usage Restrictions
For other uses, please contact: press@dtu.dk
License
Original content