Nikita Tiwari, AI Enabling Engineer for OEM PC Experiences in the Client Computing Group at Intel, presents the “Identifying and Mitigating Bias in AI” tutorial at the May 2024 Embedded Vision Summit.
From autonomous driving to immersive shopping, and from enhanced video collaboration to graphic design, AI is placing a wealth of possibilities at our fingertips. However, AI comes with vulnerabilities, which can result in costly mishaps.
In this talk, Tiwari explores risks related to bias in AI models. She examines the different types of biases that can arise in defining, training, evaluating and deploying AI models, and illustrates them with examples. She then introduces practical techniques and tools for detecting and mitigating bias, outlining their capabilities and limitations. She also touches on fairness metrics that can be useful when developing models.
See here for a PDF of the slides.
