AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

Implicit Models: Expressive Power Scales with Test-Time Compute

Jialin Liu, Lisang Ding, Stanley Osher, Wotao Yin

ICLR 2026 · Large language models › Reasoning and chain of thought · approach: Test-time compute scaling

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Studies implicit models (weight-tied networks iterated to a fixed point), arguing their expressive power scales with test-time compute.

Problem

Implicit models are known empirically to match larger explicit models with constant training memory, but the source of their expressive power is not well understood.

Approach

Frames implicit models as infinite-depth weight-tied networks whose expressive power grows with the amount of test-time computation spent iterating to a fixed point.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
increase test-time fixed-point iterations of the implicit operator (ablation)Implicit FNO · 2D steady Navier-Stokesrelative error: 1.1 -> 0.0784 (t=1 to t=50)Improved
increase test-time iterations (ablation)Implicit FNO · 2D steady Navier-Stokesempirical Lipschitz of iterate map: 23.1 -> 367Improved
increase test-time iterations (ablation)Implicit GNN · Linear programmingrelative error: 0.575 -> 0.146Improved
increase test-time iterations (ablation)implicit image deblurring (PGD, HQS) · BSDS500 deblurringPSNR and Lipschitz: L_t from 0.140/0.436 to about 5.0, PSNR rises and stabilizesImproved
use implicit HQS model instead of explicit DRUNetimage deblurring · BSDS500 deblurring test setPSNR (dB): 26.97 -> 29.18Improved
use implicit FNO instead of explicit FNO at same sizeFNO · 2D steady Navier-Stokes test setrelative error: 0.179 -> 0.078Improved
use implicit GNN instead of explicit GNNLP GNN · Linear programming testrelative error: 0.273 -> 0.177 (emb 8)Improved
smaller implicit model versus larger explicit modelLP GNN · Linear programming trainrelative error: 0.233 -> 0.203 and 0.183 -> 0.162Improved
widen explicit GNN from 8 to 32 (ablation)explicit GNN · Linear programming testrelative error: 0.273 -> 0.318Hurt
widen implicit GNN from 16 to 32 (ablation)implicit GNN · Linear programming testrelative error: 0.152 -> 0.156No clear effect
increase looped-transformer recurrent iterations (ablation)pretrained looped transformer (Geiping et al.) · LLM reasoning, charge/voltage vs charge/pay promptsqualitative output separationImproved

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