AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Path Independent Equilibrium Models Can Better Exploit Test-Time Computation

Cem Anil, Ashwini Pokle, Kaiqu Liang, Johannes Treutlein, Yuhuai Wu, Shaojie Bai, J Zico Kolter, Roger Baker Grosse

NeurIPS 2022 · Architectures and efficiency › RNNs and normalization · approach: Efficient network compression and training · 3 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 a path independent network instead of a non-PI network (adversarial initialization attack, maze)weight-tied network · Maze 13x13Attacked accuracy (%): 0 -> 100Improved
Use a path independent network instead of a non-PI network (adversarial attack, prefix sum)weight-tied network · Prefix sum 64 bitAttacked accuracy (%): 0 -> 100Improved
Use a PI network instead of a non-PI networkweight-tied network · Maze 13x13Accuracy (%): 87.12 -> 100Improved
Use a PI network instead of a non-PI networkweight-tied network · Prefix sum 64 bitAccuracy (%): 66.66 -> 100Improved
Train with mixed zero/Gaussian noise hidden state initializationunrolled prefix-sum networks · Prefix sum in- and OOD validationaccuracyImproved
Train with randomized forward solver depthunrolled prefix-sum networks · Prefix sum in- and OOD validationaccuracyImproved
Add fixed point alignment penalty lossunrolled prefix-sum networks · Prefix sumAA score and upward generalizationHurt
Train DEQ with phantom gradients instead of IFTDEQ · Maze 9x9 in-distResidual: 1.4 -> 11.83Hurt
Train DEQ with phantom gradients instead of IFTDEQ · Maze 9x9 in-distAA score: 0.99 -> 0.96No clear effect
Train DEQ with phantom gradients instead of IFTDEQ · Maze 25x25 OODAccuracy (%): 100 -> 99.88No clear effect
Use Broyden solver instead of fixed point iterations at test timeunrolled network · MazeAccuracy (%): 99.98 -> 99.97No 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