Picture for Paul Michel

Paul Michel

Emergent Communication: Generalization and Overfitting in Lewis Games

Sep 30, 2022
Mathieu Rita, Corentin Tallec, Paul Michel, Jean-Bastien Grill, Olivier Pietquin, Emmanuel Dupoux, Florian Strub

Figure 1 for Emergent Communication: Generalization and Overfitting in Lewis Games

Figure 2 for Emergent Communication: Generalization and Overfitting in Lewis Games

Figure 3 for Emergent Communication: Generalization and Overfitting in Lewis Games

Figure 4 for Emergent Communication: Generalization and Overfitting in Lewis Games

* 36th Conference on Neural Information Processing Systems (NeurIPS 2022)

Access Paper or Ask Questions

AANG: Automating Auxiliary Learning

May 27, 2022
Lucio M. Dery, Paul Michel, Mikhail Khodak, Graham Neubig, Ameet Talwalkar

Figure 1 for AANG: Automating Auxiliary Learning

Figure 2 for AANG: Automating Auxiliary Learning

Figure 3 for AANG: Automating Auxiliary Learning

Figure 4 for AANG: Automating Auxiliary Learning

* 18 pages, 9 tables and 4 figures

Access Paper or Ask Questions

Balancing Average and Worst-case Accuracy in Multitask Learning

Oct 12, 2021
Paul Michel, Sebastian Ruder, Dani Yogatama

Figure 1 for Balancing Average and Worst-case Accuracy in Multitask Learning

Figure 2 for Balancing Average and Worst-case Accuracy in Multitask Learning

Figure 3 for Balancing Average and Worst-case Accuracy in Multitask Learning

Figure 4 for Balancing Average and Worst-case Accuracy in Multitask Learning

* Under review

Access Paper or Ask Questions

Should We Be Pre-training? An Argument for End-task Aware Training as an Alternative

Sep 15, 2021
Lucio M. Dery, Paul Michel, Ameet Talwalkar, Graham Neubig

Figure 1 for Should We Be Pre-training? An Argument for End-task Aware Training as an Alternative

Figure 2 for Should We Be Pre-training? An Argument for End-task Aware Training as an Alternative

Figure 3 for Should We Be Pre-training? An Argument for End-task Aware Training as an Alternative

Figure 4 for Should We Be Pre-training? An Argument for End-task Aware Training as an Alternative

* 16 pages, 4 figures

Access Paper or Ask Questions

Learning Neural Models for Natural Language Processing in the Face of Distributional Shift

Sep 03, 2021
Paul Michel

Figure 1 for Learning Neural Models for Natural Language Processing in the Face of Distributional Shift

Figure 2 for Learning Neural Models for Natural Language Processing in the Face of Distributional Shift

Figure 3 for Learning Neural Models for Natural Language Processing in the Face of Distributional Shift

Figure 4 for Learning Neural Models for Natural Language Processing in the Face of Distributional Shift

* PhD thesis

Access Paper or Ask Questions

Examining and Combating Spurious Features under Distribution Shift

Jun 14, 2021
Chunting Zhou, Xuezhe Ma, Paul Michel, Graham Neubig

Figure 1 for Examining and Combating Spurious Features under Distribution Shift

Figure 2 for Examining and Combating Spurious Features under Distribution Shift

Figure 3 for Examining and Combating Spurious Features under Distribution Shift

Figure 4 for Examining and Combating Spurious Features under Distribution Shift

* Accepted by ICML2021

Access Paper or Ask Questions

Modeling the Second Player in Distributionally Robust Optimization

Mar 31, 2021
Paul Michel, Tatsunori Hashimoto, Graham Neubig

Figure 1 for Modeling the Second Player in Distributionally Robust Optimization

Figure 2 for Modeling the Second Player in Distributionally Robust Optimization

Figure 3 for Modeling the Second Player in Distributionally Robust Optimization

Figure 4 for Modeling the Second Player in Distributionally Robust Optimization

* Accepted at ICLR 2021

Access Paper or Ask Questions

Weight Poisoning Attacks on Pre-trained Models

Apr 14, 2020
Keita Kurita, Paul Michel, Graham Neubig

Figure 1 for Weight Poisoning Attacks on Pre-trained Models

Figure 2 for Weight Poisoning Attacks on Pre-trained Models

Figure 3 for Weight Poisoning Attacks on Pre-trained Models

Figure 4 for Weight Poisoning Attacks on Pre-trained Models

* Published as a long paper at ACL 2020

Access Paper or Ask Questions