Phil Long

machine learning researcher

[Yen-Huan Li @yenhuan_li]

==== My recommendations today ====
Large Stepsize Gradient Descent for Logistic Loss: Non-Monotonicity of the Loss Improves Optimization Efficiency https://t.co/CEgtjmLRaf

(1/2)

[Phil Long @philipmlong]

@sarielhp @sushnt I think I understand now. X must contain an
element of D, otherwise you could take E=D. It also must contain at least six elements outside D, otherwise some rotation of the outer six "petals" would contain an empty petal.

[Phil Long @philipmlong]

@sarielhp @sushnt I'm sorry; I don't understand.

[Phil Long @philipmlong]

@sarielhp @sushnt It looks like at least six are necessary: https://t.co/Hf6pDVFVB1

[Phil Long @philipmlong]

G. Valiant. Matrix Multiplication in Quadratic Time and Energy? Towards a Fine-Grained Energy-Centric Church-Turing Thesis. https://t.co/don8lyLsdD

[MLCommons @MLCommons]

Introducing the AlgoPerf: Training Algorithms Benchmark! Compete for a share of the $50,000 prize pool by submitting more effective and efficient neural network training algorithms. Learn more https://t.co/csfXkUCKbN

#Algorithms #MachineLearning #Competition

[Phil Long @philipmlong]

The code from the original version of "Sharpness-Aware Minimization and the Edge of Stability" had a bug. Patch: https://t.co/opwrhbbNZ8. New version of the paper: https://t.co/sRzzNZUhlt.

[Phil Long @philipmlong]

Code for the paper with Peter Bartlett called "Sharpness-Aware Minimization and the Edge of Stability": https://t.co/TonP6KeNFs.

[Phil Long @philipmlong]

New paper with Peter Bartlett called "Sharpness-Aware Minimization and the Edge of Stability": https://t.co/sRzzNZTJvV.

[Stat.ML Papers @StatMLPapers]

Sharpness Minimization Algorithms Do Not Only Minimize Sharpness To Achieve Better Generalization. (arXiv:2307.11007v1 [cs.LG]) https://t.co/gOMaDp0bNK

[Phil Long @philipmlong]

Dai, Y., Ahn, K., & Sra, S. (2023). The Crucial Role of Normalization in Sharpness-Aware Minimization. https://t.co/5342cousXn

[Timothy Gowers @wtgowers @wtgowers]

One of the best known open problems in combinatorics is the union-closed conjecture, which states that if you have a finite collection X of sets such that if A and B belong to X then so does the union of A and B, then at least one element of X belongs to at least half of them. 1/

[Phil Long @philipmlong]

@aryehazan I started using sympy over the past few months, and love it.

[Phil Long @philipmlong]

S. Kale, J. Lee, C. De Sa, A. Sekhari, and K. Sridharan. From Gradient Flow on Population Loss to Learning with Stochastic Gradient Descent. https://t.co/Fxsui7JlMz.

[Phil Long @philipmlong]

@SalehCU @obousquet Thanks!

[Phil Long @philipmlong]

New paper with Peter Bartlett and @obousquet called "The Dynamics of Sharpness-Aware Minimization: Bouncing Across Ravines and Drifting Towards Wide Minima": https://t.co/yM49FWZCNI.

[Phil Long @philipmlong]

@rvcraiu @niladrichat Here's an upstream paper for context: https://t.co/0eUWcgw7CB. Here a paper that is even further upstream: https://t.co/w29szKpDs1.

[Phil Long @philipmlong]

New paper with @niladrichat called "Deep linear networks can benignly overfit when shallow ones do": https://t.co/M3krVzr8fy.

[Phil Long @philipmlong]

@aryehazan @RichardHanania @EconTalker @sullydish It's not a podcast, but I like watching 3blue1brown videos: https://t.co/FVzyqVSzwj.