atlas

Research map / Optimization / Optimizers

Energetic Natural Gradient Descent

ICML 2016 · Optimization › Optimizers · approach: Natural gradient and Fisher approximations · 7 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's title only (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
Replace Fisher information metric with energy information matrix (energetic natural gradient), using d = |q(s,a1)-q(s,a2)|policy search on mountain car · Mountain car one-step updateperformance after one update: around 30% higher than Fisher at random policyHelped
Use ordinary gradient (G = I) instead of energetic natural gradientpolicy search on mountain car · Mountain car one-step updateperformance after one updateHurt
Gap between energetic and Fisher natural gradient shrinks as initial policy performance increasespolicy search on mountain car · Mountain car one-step updateadvantage of energetic over FisherMixed
Choose d so that EIM equals FIM (d = 1(w1≠w2)(p(w1)+p(w2))/2)energetic natural gradient · theoryequivalence to FisherNo clear effect
Set d so EIM equals Hessian (d(ω1,ω2)=0.5) on toy problem, giving Newton directionenergetic natural gradient on illustrative 3-outcome example · illustrative f∘pdirection of updateNo 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