Ardam Vik, Assistant Professor at the University of Southern California, is pioneering research on robots that can learn human preferences with minimal feedback. His recent presentation, "Preference Learning from Minimal Human Feedback for Interactive Robotics," reveals methods for robots to intuitively adapt. Traditional machine learning often requires vast datasets, which poses challenges for robotics involving human interaction. Vik's team explores innovative feedback types, notably comparisons and comparative language, to enhance robots' learning experiences, thus addressing the difficulties posed by conventional Learning from Demonstrations (LfD) approaches, which struggle with human imperfection.
Machine learning has transformed several industries, but its success often depends on access to enormous datasets. In the case of GPT-4 or ImageNet, scale is everything.
Vik's group at USC tackles this challenge by harnessing different forms of human signals - demonstrations, language, gestures, and gaze. His recent work focuses specifically on two types of feedback: comparisons and comparative language.
Collection
[
|
...
]