Prof. Gopala Krishna, UC Berkeley gave a talk on Human Centered Spoken Language AI on 5 July. Here is the summary of his talk: 

Large scale pre-training has dominated several fields of AI including text, vision and speech. This paradigm is enabled by unsupervised approaches like self-supervised learning on large datasets. Though highly successful, these models are not interpretable and are not grounded in basic domain sciences. In this context, he talked about grounding these large models in more grounded representations based on human physiology, in the case of speech models. Specifically, he talked about the articulatory basis for spoken language and probing this knowledge in large speech models, and creating explainable and robust speech technologies. In the later part of the talk he talked about applications translating some of this knowledge to developing high performance speech neuroprosthesis to enable communication in paralyzed populations, and other applications. 

Prof. Gopala Krishna is an Assistant Professor in the Department of Electrical Engineering and Computer Sciences at UC Berkeley. Prof. Gopala Krishna leads the Berkeley Speech Group, with research interests in spoken language, from understanding the neuroscience of language to engineering large speech models, and their applications in health. He also holds an adjunct position in Neurosurgery at UCSF. He has been recognized with numerous awards including as a Rose Hills Innovator, Noyce Innovator, Hellman Fellow, Kavli Fellow, Google Research Award, JP Morgan AI Research Award etc.

Gopala got his Ph.D degrees in 2013 from Carnegie Mellon University and Instituto Superior Tecnico, Lisbon. He obtained his B.Tech in 2006 and MS in 2008 from IIIT Hyderabad in Computer Science, with Honors in AI.

July 2024