AI Paper Atlas

Research map / Optimization / Implicit bias and generalization

Elastic Gradient Descent, an Iterative Optimization Method Approximating the Solution Paths of the Elastic Net

Oskar Allerbo, Johan Jonasson, Rebecka Jörnsten

JMLR 2023 · Optimization › Implicit bias and generalization · approach: Implicit regularization of SGD

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add momentum (gamma=0.5) to elastic gradient descent (ablation)EGD · Twitter Popularity (291624x77)execution time in seconds: 48 -> 6.0Improved
add momentum (gamma=0.5) to elastic gradient descent (ablation)EGD · Twitter Popularity (291624x77)model size: 60 -> 27Improved
add momentum (gamma=0.5) to elastic gradient descent (ablation)EGD · Superconductors (21263x81)test R2: 0.70 -> 0.66Hurt
add momentum (gamma=0.5) to elastic gradient descent (ablation)EGD · English Readability (2834x768)model size: 105 -> 115Hurt
use elastic gradient descent instead of the elastic net pathelastic net · Twitter Popularity (291624x77)execution time in seconds: 5100 -> 48Improved
use elastic gradient descent instead of the elastic net pathelastic net · Superconductors (21263x81)execution time in seconds: 340 -> 2.6Improved
use elastic gradient descent instead of coordinate descentcoordinate descent · Protein Structure (45730x9)test R2: 0.15 -> 0.24Improved
use elastic gradient descent instead of coordinate descentcoordinate descent · English Readability (2834x768)test R2: 0.58 -> 0.69Improved
use elastic gradient descent instead of the elastic netelastic net · Diabetes (alpha=0.3)number of models along path: 10 -> 3Improved
use elastic gradient descent instead of the elastic netelastic net · Diabetes (alpha=0.7)number of models along path: 11 -> 7Improved
add momentum to elastic gradient descent for p>n (ablation)EGD · synthetic two-block data (p>n=100)specificityImproved
keep number of non-zero parameters constant at 40 as p grows (ablation)EGD with momentum · synthetic two-block dataspecificityNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy