Aaron Defazio
Scholar

Aaron Defazio

Google Scholar ID: KEzJsdkAAAAJ
Fundamental AI Research team, Meta NY
Machine LearningNumerical Optimization
Homepage↗Google Scholar↗
Citations & Impact
All-time
Citations
5,407
 
H-index
22
 
i10-index
26
 
Publications
20
 
Co-authors
0
 
Publications
6 items
ScheduleFree+: Scaling Learning-Rate-Free & Schedule-Free Learning to Large Language Models
2026
Cited
0
Smoothing DiLoCo with Primal Averaging for Faster Training of LLMs
2025
Cited
0
Purifying Shampoo: Investigating Shampoo's Heuristics by Decomposing its Preconditioner
2025
Cited
0
Why Gradients Rapidly Increase Near the End of Training
2025
Cited
0
PARQ: Piecewise-Affine Regularized Quantization
2025
Cited
0
Directional Smoothness and Gradient Methods: Convergence and Adaptivity
arXiv.org · 2024
Cited
5
Resume
Academic Achievements
  • Has published several papers (specific list not provided).
Research Experience
  • Serves as a Research Scientist at Meta, focusing on fundamental research in artificial intelligence.
Background
  • Research Scientist in Fundamental AI Research at Meta, working on artificial intelligence.