Detecting Deepfakes: From Classical Physics to Deep Learning

About the talk

As the world braces for the largest election year with over 64 nations heading to the polls, the influence of artificial intelligence looms large, poised to make a profound impact. From the proliferation of misinformation to the emergence of deepfakes and the exploration of positive use cases, the landscape of political discourse is undergoing a seismic shift.

In this session, Gautham Koorma and Romit Barua will discuss the following:

A classical physics based forensic technique for detecting manipulation in images
The approaches from their paper on detecting cloned voices and issues related to generalization of ML models for deepfake detection on “in the wild” media
A modern deep learning based technique for detecting fake videos

Have questions for the speakers? Leave a comment.
About the speakers

Gautham Koorma is a Machine Learning Engineer and Researcher working on deep learning models for audio signal processing and computer vision. He has recently graduated with a Master’s in Information Science from UC Berkeley, specializing in machine learning, digital forensics, policy, product management, and entrepreneurship.

Romit Barua is a Machine Learning Researcher and has previously worked as Head of Investment Control at Texas Exchange Bank and Risk Analyst at Q Investments. He has also recently graduated with a Master’s in Management Information Systems from UC Berkeley, specializing in Data Science. His research at UC Berkeley focused on audio deepfake detection.

Rumanu will be moderating the talk. She is a Machine Learning Researcher who has spent her weekends for the past five years at tech meetups and conferences, her interest peaking at the intersection of adtech and data privacy preservation. She toys around with humor(ous?) NLP usecases when not hitting comedy open mics to be a better tech speaker.

More at http://has.gy/5npf