Research talk: Bucket of me: Using few-shot learning to realize teachable AI systems

Опубликовано 8 февраля 2022, 16:39

Speaker: Daniela Massiceti, Senior Researcher, Microsoft Research

We’re entering a technological era that is all about “me”—from personalized shopping recommendations to avatars, and even bespoke healthcare treatments. Deeper inspection of artificial intelligence (AI) systems, however, reveals that “me” is not really me. The coarse-grained ways that AI systems classify people has a significant impact on millions whose complex identities do not easily fit into a predefined bucket. Join Microsoft Researcher Daniela Massiceti in exploring how few-shot learning can realize a vision of teachable AI, giving users the agency to curate their own “bucket of me.” We will discuss recent advances in few-shot learning that now make it feasible to personalize a model using just the data a user selects about themselves. We invite researchers across machine learning and human-computer interaction to join in this vision through using a newly released real-world dataset for teachable object recognition collected by people who are blind or have low vision. Together, we’ll explore the vision of teachable AI systems, how we can leverage advances in few-shot learning to realize them, and how a new teachable benchmark and dataset can be a call to action in this space.

Learn more about the 2021 Microsoft Research Summit: Aka.ms/researchsummit