Next month, Mila researchers will showcase their work in full force at the 38th International Conference on Machine Learning (ICML). This annual event brings together some of the brightest minds of the machine learning research community and will run virtually from July 18 to 24.
Of the 5,513 submissions this year, 1,184 papers were accepted (21.5%); 30 of those papers were co-authored by Mila researchers.
The work being presented by Mila researchers covers a wide range of topics and demonstrates the far-reaching potential applications of machine learning in areas such as health, chemical physics, and even the airline industry.
In “Structured Convolutional Kernel Networks for Airline Crew Scheduling,” Mila core academic member Simon Lacoste-Julien and colleagues introduce a novel deep structured predictor: structured convolutional kernel networks (Struct-CKN). The initial solutions obtained using Struct-CKN on a flight-connection dataset revealed that it can be further refined by an airline crew scheduling solver.
Core faculty member Jian Tang and team designed a unified framework, Local-instance and Global-semantic Learning (GraphLoG), for self-supervised whole-graph representation learning, proposing hierarchical prototypes to capture global semantic clusters. The verified experiments on chemical and biological benchmark data sets demonstrate the method’s effectiveness, with potential implications for many tasks such as molecule properties prediction in drug and material discovery.
On the other end of the spectrum, in “Continuous Coordination As a Realistic Scenario for Lifelong Learning,” Mila University professors Sarath Chandar and Aaron Courville, as well as research intern Hadi Nekoei and Master’s student Akilesh Badrinaaraayanan introduce Lifelong Hanabi, a continual multi-agent reinforcement learning testbed in which every task is coordinating with a partner that’s an expert player of Hanabi.
Below, accepted papers (spotlight and oral), workshops, and tutorials co-authored and co-organized by Mila members:
Continual learning with deep architectures
Vincenzo Lomonaco, Irina Rish
Random Matrix Theory and ML (RMT+ML)
Fabian Pedregosa, Jeffrey Pennington, Thomas Trogdon, Courtney Paquette
Out-of-Distribution Generalization via Risk Extrapolation (REx)
David Krueger, Ethan Caballero, Joern-Henrik Jacobsen, Amy Zhang, Jonathan Binas, Dinghuai Zhang, Rémi Le Priol, Aaron Courville
On Disentangled Representations Learned from Correlated Data
Frederik Träuble, Elliot Creager, Niki Kilbertus, Francesco Locatello, Andrea Dittadi, Anirudh Goyal, Bernhard Schölkopf, Stefan Bauer
Diffusion Earth Mover’s Distance and Distribution Embeddings
Alexander Tong, Guillaume Huguet, Amine Natik, Kincaid MacDonald, Manik Kuchroo, Ronald Coifman, Guy Wolf, Smita Krishnaswamy
An End-to-End Framework for Molecular Conformation Generation via Bilevel Programming
Minkai Xu, Wujie Wang, Shitong Luo, Chence Shi, Yoshua Bengio, Rafael Gomez-Bombarelli, Jian Tang
Catastrophic Fisher Explosion: Early Phase Fisher Matrix Impacts Generalization
Stanislaw Jastrzebski, Devansh Arpit, Oliver Astrand, Giancarlo Kerg, Huan Wang, Caiming Xiong, Richard Socher, Kyunghyun Cho, Krzysztof J Geras
Exploration in Approximate Hyper-State Space for Meta Reinforcement Learning
Luisa Zintgraf, Leo Feng, Cong Lu, Maximilian Igl, Kristian Hartikainen, Katja Hofmann, Shimon Whiteson
Non-Autoregressive Electron Redistribution Modeling for Reaction Prediction
Hangrui Bi, Hengyi Wang, Chence Shi, Connor Coley, Jian Tang, Hongyu Guo
Trajectory Diversity for Zero-Shot Coordination
Andrei Lupu, Brandon Cui, Hengyuan Hu, Jakob Foerster
Robust Representation Learning via Perceptual Similarity Metrics
Saeid A Taghanaki, Kristy Choi, Amir Hosein Khasahmadi, Anirudh Goyal
Efficient Deviation Types and Learning for Hindsight Rationality in Extensive-Form Games
Dustin Morrill, Ryan D’Orazio, Marc Lanctot, James Wright, Michael Bowling, Amy Greenwald
INNF+: Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models
Chin-Wei Huang, David Krueger, Rianne Van den Berg, George Papamakarios, Tian Qi Chen, Danilo J. Rezende
Workshop Home Page