Hybrid Seminar: Lav Varshney, "Information Lattice Learning: From Knowledge Discovery to Hip Hop Music Co-creation"

In person: 260 Dreese Laboratories

Title: Information Lattice Learning: From Knowledge Discovery to Hip Hop Music Co-creation

Abstract: Can AI learn rules and abstractions from raw data? How little prior knowledge and how little data are needed to do so? How interpretable can the learned rules and the rule-learning process be? Whereas humans are extremely flexible in abstracting raw patterns and distilling rules, current AI systems are mostly good at either applying human-distilled rules or capturing patterns in a task-driven fashion, but not at learning patterns in a human-interpretable way like human-induced theory and knowledge. We develop Information Lattice Learning (ILL) to distill human-interpretable understanding of data in a human-like manner, based on the core idea of computational abstraction and statistical learning/inference on lattices of abstractions. The resulting framework generalizes Shannon's information lattice and further brings learning into the picture. We demonstrate ILL for knowledge discovery, in learning music theory from sheet music, chemical laws from molecule structures, and rules of neurogenesis from single-cell RNA-seq data. For instance, ILL explicitly reconstructs about 70% of a standard music theory curriculum from only 370 of Bach's chorales, while also discovering new rules that interest music researchers. We also demonstrate ILL's near-human performance on truly-few-shot learning, where we achieve state-of-the-art results in character classifications from few and only few training examples---no extra pretraining or validation data. For instance, we can reach 80%/90% MNIST test accuracy by using only the first/first four training images per class. We finally demonstrate ILL in human-AI music co-creation platforms, where it lowers the bar for creativity and engages communities to grow their cultural wealth.

Bio: Lav R. Varshney is an associate professor of electrical and computer engineering at the University of Illinois Urbana-Champaign, with further appointments in computer science, neuroscience, industrial engineering, digital agriculture, and personalized nutrition, as well as the Discovery Partners Institute in Chicago.  He is also a founder of Kocree, Inc., a startup company building a social music co-creation platform.  He was a principal research scientist at Salesforce Research in 2019-2020 focused on AI ethics and AI for social good, and a research staff member at IBM Research in 2010-2013 where he led the Chef Watson project for culinary creativity to worldwide acclaim. His research interests include information theory, artificial intelligence, and creativity.