AD
Aaron Defazio
cs.LGstat.MLmath.OCcs.AIcs.CVeess.IVaccelerationchallengecompressed sensingcs.IR
On Valency
published · living versionsW_5352xgra·v1 · currentpublished
fastMRI: An Open Dataset and Benchmarks for Accelerated MRI
with Jure Zbontar, Florian Knoll, Anuroop Sriram, Tullie Murrell +22
1 version
Preprints & journals
40 papers in the corpus · 2012–2026FlexGrad: a root-free approach to adaptive step sizes2610.09203v1 · Pablo Barros, Aaron Defazio, Vincent Guigues2026 · 0 citationsarXiv
ScheduleFree+: Scaling Learning-Rate-Free & Schedule-Free Learning to Large Language Models2605.19095v1 · Aaron Defazio2026 · 0 citationsarXiv
Smoothing DiLoCo with Primal Averaging for Faster Training of LLMs2512.17131v3 · Aaron Defazio, Konstantin Mishchenko, Parameswaran Raman et al.2025 · 0 citationsarXiv
Purifying Shampoo: Investigating Shampoo's Heuristics by Decomposing its Preconditioner2506.03595v2 · Runa Eschenhagen, Aaron Defazio, Tsung-Hsien Lee et al.2025 · 1 citationarXiv
Why Gradients Rapidly Increase Near the End of Training2506.02285v2 · Aaron Defazio2025 · 0 citationsarXiv
PARQ: Piecewise-Affine Regularized Quantization2503.15748v1 · Lisa Jin, Jianhao Ma, Zechun Liu et al.2025 · 0 citationsarXiv
The Road Less Scheduled2405.15682v4 · Aaron Defazio, Xingyu Alice Yang, Harsh Mehta et al.2024 · 27 citationsarXiv
Directional Smoothness and Gradient Methods: Convergence and Adaptivity2403.04081v2 · Aaron Mishkin, Ahmed Khaled, Yuanhao Wang et al.2024 · 1 citationarXiv
Optimal Linear Decay Learning Rate Schedules and Further Refinements2310.07831v2 · Aaron Defazio, Ashok Cutkosky, Harsh Mehta et al.2023 · 5 citationsarXiv
Prodigy: An Expeditiously Adaptive Parameter-Free Learner2306.06101v4 · Konstantin Mishchenko, Aaron Defazio2023 · 3 citationsarXiv
Mechanic: A Learning Rate Tuner2306.00144v2 · Ashok Cutkosky, Aaron Defazio, Harsh Mehta2023 · 3 citationsarXiv
MoMo: Momentum Models for Adaptive Learning Rates2305.07583v3 · Fabian Schaipp, Ruben Ohana, Michael Eickenberg et al.2023 · 3 citationsarXiv
Learning-Rate-Free Learning by D-Adaptation2301.07733v5 · Aaron Defazio, Konstantin Mishchenko2023 · 8 citationsarXiv
Grad-GradaGrad? A Non-Monotone Adaptive Stochastic Gradient Method2206.06900v1 · Aaron Defazio, Baoyu Zhou, Lin Xiao2022 · 0 citationsarXiv
Stochastic Polyak Stepsize with a Moving Target2106.11851v2 · Robert M. Gower, Aaron Defazio, Michael Rabbat2021 · 0 citationsarXiv
Adaptivity without Compromise: A Momentumized, Adaptive, Dual Averaged Gradient Method for Stochastic Optimization2101.11075v3 · Aaron Defazio, Samy Jelassi2021 · 41 citationsarXiv
Dual Averaging is Surprisingly Effective for Deep Learning Optimization2010.10502v1 · Samy Jelassi, Aaron Defazio2020 · 2 citationsarXiv
Using Deep Learning to Accelerate Knee MRI at 3 T: Results of an Interchangeability Study.32755163 · Recht, Michael P, Zbontar, Jure, Sodickson, Daniel K et al.2021 · 177 citationsAJR. American journal of roentgenology. 2020;215(6):1421-1429
Momentum via Primal Averaging: Theoretical Insights and Learning Rate Schedules for Non-Convex Optimization2010.00406v4 · Aaron Defazio2020 · 6 citationsarXiv
Almost sure convergence rates for Stochastic Gradient Descent and Stochastic Heavy Ball2006.07867v2 · Othmane Sebbouh, Robert M. Gower, Aaron Defazio2020 · 30 citationsarXiv
Advancing machine learning for MR image reconstruction with an open competition: Overview of the 2019 fastMRI challenge.32506658 · Knoll, Florian, Murrell, Tullie, Sriram, Anuroop et al.2021 · 264 citationsMagnetic resonance in medicine. 2020;84(6):3054-3070
The Power of Factorial Powers: New Parameter settings for (Stochastic) Optimization2006.01244v3 · Aaron Defazio, Robert M. Gower2020 · 0 citationsarXiv
End-to-End Variational Networks for Accelerated MRI Reconstruction2004.06688v2 · Anuroop Sriram, Jure Zbontar, Tullie Murrell et al.2020 · 343 citationsarXiv
fastMRI: A Publicly Available Raw k-Space and DICOM Dataset of Knee Images for Accelerated MR Image Reconstruction Using Machine Learning.32076662 · Knoll, Florian, Zbontar, Jure, Sriram, Anuroop et al.2021 · 416 citationsRadiology. Artificial intelligence. 2020;2(1):e190007
MRI Banding Removal via Adversarial Training2001.08699v3 · Aaron Defazio, Tullie Murrell, Michael P. Recht2020 · 10 citationsarXiv
Offset Sampling Improves Deep Learning based Accelerated MRI Reconstructions by Exploiting Symmetry1912.01101v2 · Aaron Defazio2019 · 5 citationsarXiv
GrappaNet: Combining Parallel Imaging with Deep Learning for Multi-Coil MRI Reconstruction1910.12325v4 · Anuroop Sriram, Jure Zbontar, Tullie Murrell et al.2019 · 119 citationsarXiv
Beyond Folklore: A Scaling Calculus for the Design and Initialization of ReLU Networks1906.04267v3 · Aaron Defazio, L'eon Bottou2019 · 2 citationsNeural Comput & Applic (2022)
Methods of interpreting error estimates for grayscale image reconstructions1902.00608v1 · Aaron Defazio, Mark Tygert2019 · 0 citationsarXiv
On the Ineffectiveness of Variance Reduced Optimization for Deep Learning1812.04529v2 · Aaron Defazio, L'eon Bottou2018 · 32 citationsarXiv
Controlling Covariate Shift using Balanced Normalization of Weights1812.04549v2 · Aaron Defazio, L'eon Bottou2018 · 0 citationsarXiv
On the Curved Geometry of Accelerated Optimization1812.04634v2 · Aaron Defazio2018 · 3 citationsarXiv
fastMRI: An Open Dataset and Benchmarks for Accelerated MRI1811.08839v2 · Jure Zbontar, Florian Knoll, Anuroop Sriram et al.2018 · 587 citationsarXivon Valency
A Simple Practical Accelerated Method for Finite Sums1602.02442v2 · Aaron Defazio2016 · 54 citationsarXiv
New Optimisation Methods for Machine Learning1510.02533v2 · Aaron Defazio2015 · 4 citationsarXiv
A Comparison of learning algorithms on the Arcade Learning Environment1410.8620v1 · Aaron Defazio, Thore Graepel2014 · 20 citationsarXiv
A Convex Formulation for Learning Scale-Free Networks via Submodular Relaxation1407.2697v1 · Aaron J. Defazio, Tiberio S. Caetano2014 · 23 citationsAdvances in Neural Information Processing Systems 25 (NIPS 2012)
Finito: A Faster, Permutable Incremental Gradient Method for Big Data Problems1407.2710v1 · Aaron J. Defazio, Tib'erio S. Caetano, Justin Domke2014 · 117 citationsInternational Conference on Machine Learning 2014
SAGA: A Fast Incremental Gradient Method With Support for Non-Strongly Convex Composite Objectives1407.0202v3 · Aaron Defazio, Francis Bach, Simon Lacoste-Julien2014 · 1,207 citationsarXiv
A Graphical Model Formulation of Collaborative Filtering Neighbourhood Methods with Fast Maximum Entropy Training1206.4622v1 · Aaron Defazio (ANU), Tiberio Caetano2012 · 12 citationsarXiv
Career total: 58 works. 40 are in this corpus.
Profile built from the corpus for this byline.
Author records are still filling in while the Hub is in alpha. If this is your page, you'll be able to claim it soon. Spot a mistake? Tell us.