Georgia Institute of Technology

May 4, 2019

Researchers in the Machine Learning Center at Georgia Tech (ML@GT) will present 12 papers at the seventh annual International Conference on Learning Representations (ICLR), taking in New Orleans, La. May 6-9. Assistant professor Dhruv Batra is an area chair and associate professor Le Song will give an invited talk at the Representation Learning on Graphs and Manifolds workshop.

ICLR is one of the fastest growing artificial intelligence conferences in the world and is globally respected as a premier conference for artificial intelligence researchers who focus on representation learning which is generally referred to as deep learning.

ML@GT research includes a paper on a self-monitoring navigation agent in collaboration with Salesforce and the University of Maryland, and work on transferring policy with strategy optimization.

Only 500 or about a third of submissions were accepted as poster presentations and 24 as oral presentations. All of Georgia Tech’s work is in the poster session.

“ICLR has continued to grow and is now one of the premier conferences in artificial intelligence and machine learning. For ML@GT to have 12 papers in our third year as a center is a sign of our prominence in these communities and the quality of work that the center publishes,” said Byron Boots, an assistant professor in the School of Interactive Computing and ML@GT.

Conference presentations will touch on topics such as hierarchical modeling and sparse coding, and speakers include Ian Goodfellow, Apple’s new director of machine learning, and Cynthia Dwork, a distinguished scientist at Microsoft Research.

ICLR has helped lead the charge in increasing inclusivity and diversity at conferences by building on the efforts of groups like Black in AI, Queer in AI, Women in Machine Learning, and LatinX in AI.

Georgia Tech’s research:

• Multi-class Classification Without Multi-Class Labels
• A Closer Look at Few-Shot Classification
• Combinatorial Attacks on Binarized Neural Networks
• Adversarial Imitation via Variational Inverse Reinforcement Learning
• Self-Monitoring Navigation Agent via Auxiliary Progress Estimation (For more information on this research, check out our summary here.)
• On Computation and Generalization of Generative Adversarial Networks under Spectrum Control
• Modeling the Long Term Future in Model-Based Reinforcement Learning
• Label Super Resolution Networks
• Policy Transfer with Strategy Optimization
• L-Shapley and C-Shapley: Efficient Model Interpretation for Structured Data
• Learning a Meta-Solver for Syntax-Guided Program Synthesis
• DyRep: Learning Representations over Dynamic Graphs