AI Paper Atlas

Research map / Optimization / Implicit bias and generalization

Optimizing Attention with Mirror Descent: Generalized Max-Margin Token Selection

Addison Kristanto Julistiono, Davoud Ataee Tarzanagh, Navid Azizan

JMLR 2026 · Optimization › Implicit bias and generalization · approach: Implicit bias in linear networks · 6 citations (OpenAlex)

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
use l1.1 potential instead of l2 (GD) in mirror descentMD for attention · IMDb, model size 3test accuracy: 81.66 -> 83.47Improved
use l1.1 potential instead of l2 (GD) in mirror descentMD for attention · IMDb, model size 4test accuracy: 81.05 -> 83.36Improved
use l1.1 potential instead of l2 (GD) in mirror descentMD for attention · IMDb, model size 6test accuracy: 82.22 -> 83.65Improved
use l1.1 potential instead of l3MD for attention · IMDb, model size 6test accuracy: 81.97 -> 83.65Improved
use l3 potential instead of l2 (GD)MD for attention · IMDb, model sizes 3, 4, 6test accuracy: 81.66 -> 82.57 (size 3); 82.22 -> 81.97 (size 6)Mixed
train with l1.1-MD instead of GDtransformer classifier · IMDb attention weightsnumber of weights >= 0.06 in W_K: 27,224 -> 18,206Improved
train with l1.1-MD instead of GDtransformer classifier · IMDb attention weightsnumber of weights >= 0.06 in W_V: 10,127 -> 7,643Improved
train with l1.1-MD instead of GDtransformer classifier · IMDb token selection on GPT-4o reviewsoptimal token selection: 3 out of 5 sample sentencesImproved
train with l1.1-MD instead of AdamViT · CIFAR-10test accuracy: approximately 82%No clear effect
train with lp-MD instead of Adam/SGDViT · CIFAR-100test accuracy: around 55%No clear effect
train with l1.1-MD instead of AdamViT · CIFAR-10 and CIFAR-100 attention weightssparsity of attention weightsImproved

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