Get our free extension to see links to code for papers anywhere online!

Chrome logo Add to Chrome

Firefox logo Add to Firefox

Picture for Daniel M. Roy

Daniel M. Roy

University of Toronto

NeurIPS 2020 Competition: Predicting Generalization in Deep Learning


Dec 14, 2020
Yiding Jiang, Pierre Foret, Scott Yak, Daniel M. Roy, Hossein Mobahi, Gintare Karolina Dziugaite, Samy Bengio, Suriya Gunasekar, Isabelle Guyon, Behnam Neyshabur

* 20 pages, 2 figures. Accepted for NeurIPS 2020 Competitions Track. Lead organizer: Yiding Jiang 

  Access Paper or Ask Questions

On the Information Complexity of Proper Learners for VC Classes in the Realizable Case


Nov 05, 2020
Mahdi Haghifam, Gintare Karolina Dziugaite, Shay Moran, Daniel M. Roy

* 5 Pages 

  Access Paper or Ask Questions

Deep learning versus kernel learning: an empirical study of loss landscape geometry and the time evolution of the Neural Tangent Kernel


Oct 28, 2020
Stanislav Fort, Gintare Karolina Dziugaite, Mansheej Paul, Sepideh Kharaghani, Daniel M. Roy, Surya Ganguli

* 19 pages, 19 figures, In Advances in Neural Information Processing Systems 34 (NeurIPS 2020) 

  Access Paper or Ask Questions

Enforcing Interpretability and its Statistical Impacts: Trade-offs between Accuracy and Interpretability


Oct 28, 2020
Gintare Karolina Dziugaite, Shai Ben-David, Daniel M. Roy

* 12 pages; minor edits 

  Access Paper or Ask Questions

In Search of Robust Measures of Generalization


Oct 22, 2020
Gintare Karolina Dziugaite, Alexandre Drouin, Brady Neal, Nitarshan Rajkumar, Ethan Caballero, Linbo Wang, Ioannis Mitliagkas, Daniel M. Roy

* 27 pages, 11 figures, 34th Conference on Neural Information Processing Systems (NeurIPS 2020), Vancouver, Canada 

  Access Paper or Ask Questions

Pruning Neural Networks at Initialization: Why are We Missing the Mark?


Sep 18, 2020
Jonathan Frankle, Gintare Karolina Dziugaite, Daniel M. Roy, Michael Carbin


  Access Paper or Ask Questions

Relaxing the I.I.D. Assumption: Adaptive Minimax Optimal Sequential Prediction with Expert Advice


Jul 13, 2020
Blair Bilodeau, Jeffrey Negrea, Daniel M. Roy

* 60 pages. Blair Bilodeau and Jeffrey Negrea are equal-contribution authors; order was determined randomly 

  Access Paper or Ask Questions

Improved Bounds on Minimax Regret under Logarithmic Loss via Self-Concordance


Jul 02, 2020
Blair Bilodeau, Dylan J. Foster, Daniel M. Roy

* Proceedings of the 37th International Conference on Machine Learning, ICML 2020 

  Access Paper or Ask Questions

On the role of data in PAC-Bayes bounds


Jun 19, 2020
Gintare Karolina Dziugaite, Kyle Hsu, Waseem Gharbieh, Daniel M. Roy

* 23 pages, 7 figures 

  Access Paper or Ask Questions

Sharpened Generalization Bounds based on Conditional Mutual Information and an Application to Noisy, Iterative Algorithms


Apr 27, 2020
Mahdi Haghifam, Jeffrey Negrea, Ashish Khisti, Daniel M. Roy, Gintare Karolina Dziugaite

* 17 pages 

  Access Paper or Ask Questions

Linear Mode Connectivity and the Lottery Ticket Hypothesis


Dec 11, 2019
Jonathan Frankle, Gintare Karolina Dziugaite, Daniel M. Roy, Michael Carbin

* This submission subsumes 1903.01611 ("Stabilizing the Lottery Ticket Hypothesis" and "The Lottery Ticket Hypothesis at Scale") 

  Access Paper or Ask Questions

In Defense of Uniform Convergence: Generalization via derandomization with an application to interpolating predictors


Dec 09, 2019
Jeffrey Negrea, Gintare Karolina Dziugaite, Daniel M. Roy

* 12 pages 

  Access Paper or Ask Questions

Information-Theoretic Generalization Bounds for SGLD via Data-Dependent Estimates


Nov 08, 2019
Jeffrey Negrea, Mahdi Haghifam, Gintare Karolina Dziugaite, Ashish Khisti, Daniel M. Roy

* 23 pages, 1 figure. To appear in, Advances in Neural Information Processing Systems (33), 2019 

  Access Paper or Ask Questions

Fast-rate PAC-Bayes Generalization Bounds via Shifted Rademacher Processes


Aug 20, 2019
Jun Yang, Shengyang Sun, Daniel M. Roy

* 18 pages 

  Access Paper or Ask Questions

Black-box constructions for exchangeable sequences of random multisets


Aug 17, 2019
Creighton Heaukulani, Daniel M. Roy

* 12 pages 

  Access Paper or Ask Questions

NUQSGD: Improved Communication Efficiency for Data-parallel SGD via Nonuniform Quantization


Aug 16, 2019
Ali Ramezani-Kebrya, Fartash Faghri, Daniel M. Roy

* 21 pages, 6 figures 

  Access Paper or Ask Questions

The Lottery Ticket Hypothesis at Scale


Mar 05, 2019
Jonathan Frankle, Gintare Karolina Dziugaite, Daniel M. Roy, Michael Carbin


  Access Paper or Ask Questions

Entropy-SGD optimizes the prior of a PAC-Bayes bound: Generalization properties of Entropy-SGD and data-dependent priors


Mar 03, 2018
Gintare Karolina Dziugaite, Daniel M. Roy

* 18 pages, 6 figures; differentially private PAC-Bayes theorem moved to arXiv:1802.09583; new empirical results 

  Access Paper or Ask Questions

Data-dependent PAC-Bayes priors via differential privacy


Feb 26, 2018
Gintare Karolina Dziugaite, Daniel M. Roy

* 17 pages, 2 figures; subsumes and extends some results first reported in arXiv:1712.09376 

  Access Paper or Ask Questions

Exchangeable modelling of relational data: checking sparsity, train-test splitting, and sparse exchangeable Poisson matrix factorization


Dec 06, 2017
Victor Veitch, Ekansh Sharma, Zacharie Naulet, Daniel M. Roy

* 9 pages, 4 figures 

  Access Paper or Ask Questions

Computing Nonvacuous Generalization Bounds for Deep (Stochastic) Neural Networks with Many More Parameters than Training Data


Oct 19, 2017
Gintare Karolina Dziugaite, Daniel M. Roy

* Proceedings of the Thirty-Third Conference on Uncertainty in Artificial Intelligence, UAI 2016, August 11--15, 2017, Sydney, NSW, Australia 
* 14 pages, 1 table, 2 figures. Corresponds with UAI camera ready and supplement. Includes additional references and related experiments 

  Access Paper or Ask Questions

A study of the effect of JPG compression on adversarial images


Aug 02, 2016
Gintare Karolina Dziugaite, Zoubin Ghahramani, Daniel M. Roy

* 8 pages, 4 figures 

  Access Paper or Ask Questions

A characterization of product-form exchangeable feature probability functions


Jul 07, 2016
Marco Battiston, Stefano Favaro, Daniel M. Roy, Yee Whye Teh

* 21 pages 

  Access Paper or Ask Questions

The combinatorial structure of beta negative binomial processes


Jun 23, 2016
Creighton Heaukulani, Daniel M. Roy

* Bernoulli 2016, Vol. 22, No. 4, 2301-2324 
* Published at http://dx.doi.org/10.3150/15-BEJ729 in the Bernoulli (http://isi.cbs.nl/bernoulli/) by the International Statistical Institute/Bernoulli Society (http://isi.cbs.nl/BS/bshome.htm

  Access Paper or Ask Questions

The Mondrian Kernel


Jun 16, 2016
Matej Balog, Balaji Lakshminarayanan, Zoubin Ghahramani, Daniel M. Roy, Yee Whye Teh

* Accepted for presentation at the 32nd Conference on Uncertainty in Artificial Intelligence (UAI 2016) 

  Access Paper or Ask Questions

Measuring the reliability of MCMC inference with bidirectional Monte Carlo


Jun 07, 2016
Roger B. Grosse, Siddharth Ancha, Daniel M. Roy


  Access Paper or Ask Questions

Mondrian Forests for Large-Scale Regression when Uncertainty Matters


May 27, 2016
Balaji Lakshminarayanan, Daniel M. Roy, Yee Whye Teh

* Proceedings of the 19th International Conference on Artificial Intelligence and Statistics (AISTATS) 2016, Cadiz, Spain. JMLR: W&CP volume 51 

  Access Paper or Ask Questions

Neural Network Matrix Factorization


Dec 15, 2015
Gintare Karolina Dziugaite, Daniel M. Roy

* Minor modifications to notation. Added additional experiments and discussion. 7 pages, 2 tables 

  Access Paper or Ask Questions