[Sci-Tech NOW] Researchers Use Text-Annotated Data to Unpack AI’s Black Box

The UNIST analysis staff has proposed a strategy for decoding the AI black field by annotating coaching knowledge with pure language descriptions. From left: Professor Tae-Hwan Kim, Researcher Chaeree Kim, and Researcher Jaeyeon Bae of UNIST. Photo courtesy of UNIST.

■ Ulsan National Institute of Science and Technology (UNIST) introduced on the twenty eighth {that a} staff led by Professor Tae-Hwan Kim from the Graduate School of AI has proposed a studying methodology that explains the AI black field by translating coaching knowledge into human-understandable ‘natural language.’ Existing analysis in Explainable AI (XAI) has primarily targeted on the post-hoc evaluation of a totally educated mannequin’s inside operations or prediction outcomes. In distinction, the analysis staff targeted on the ‘data’ itself, the supply of AI studying, aiming to elucidate the mannequin’s decision-making course of by articulating and analyzing the info’s options by textual descriptions.