AI Paper Atlas

Research map / Optimization / Nonconvex and smooth optimization

On lp-hyperparameter Learning via Bilevel Nonsmooth Optimization

Takayuki Okuno, Akiko Takeda, Akihiro Kawana, Motokazu Watanabe

JMLR 2021 · Optimization › Nonconvex and smooth optimization · approach: Penalty and implicit-gradient bilevel methods

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 Algorithm 1 instead of bayesoptbilevel l0.8 hyperparameter learning · Facebook, p=1time (sec): 17.399 vs 146.144Improved
use Algorithm 1 instead of bayesoptbilevel l0.8 hyperparameter learning · Student, p=0.8time (sec): 2.348 vs 409.382Improved
use Algorithm 1 instead of gridsearchbilevel l0.8 hyperparameter learning · BodyFat, p=0.8sparsity: 0.000 vs 0.429Hurt
use Algorithm 1 over bayesopt and gridsearch for validation errorbilevel lp hyperparameter learning · 15 problem instancestime (sec): best time in 10 out of 15Improved
decrease p from 0.8 or 1 to 0.5Algorithm 1 · Facebooksparsity: 0.925 vs 0.057 and 0.245Improved
increase sample size of Facebook from 0.5 to fullbayesopt · Facebook, l0.8time (sec)Hurt
increase feature size of Studentbayesopt and gridsearch · Student, l0.8time (sec)Hurt
stop Algorithm 1 early when smoothing parameter mu is about 0.01 (ablation)Algorithm 1 · Facebook and BodyFat, l0.8test errorImproved
use fmincon (Alg.1-B) instead of implicit function approach (Alg.1-A) for subproblem (ablation)Algorithm 1 · Insurance, multiple hyperparameterssparsity: 0.000 -> 0.988Improved
use fmincon (Alg.1-B) instead of Alg.1-A (ablation)Algorithm 1 · Facebook, multiple hyperparametersfeasibilityHurt
use Algorithm 1 instead of bayesopt with multiple hyperparametersbilevel l0.5 problem · Student, 273 hyperparametersErrval: 0.771 vs 18.324Improved
use Algorithm 1 instead of bayesopt with multiple hyperparametersbilevel l0.5 problem · Facebook, 54 hyperparameterstime (sec): 20.247 vs 600.000Improved

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