AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Differentiable Adaptive Computation Time for Visual Reasoning

Cristobal Eyzaguirre, Alvaro Soto

CVPR 2020 · Architectures and efficiency › RNNs and normalization · approach: Differentiable optimization layers · 13 citations (OpenAlex)

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 DACT-MAC with a 15-step maximum instead of the reported non-adaptive MACMAC · CLEVR overallAccuracy: 98.6% -> 98.72%Improved
Add adaptive computation with DACTMAC · GQA test-devAccuracy: 77.52 -> 77.35Hurt
Use DACT rather than ACTMAC · GQA test-devAccuracy: 77.21 -> 77.35Improved
Increase DACT ponder cost from 1e-3 to 5e-2 (ablation)DACT-MAC · GQA test-devSteps: 3.69 -> 1.63Improved
Increase DACT ponder cost from 1e-3 to 5e-2 (ablation)DACT-MAC · GQA test-devAccuracy: 77.31 -> 77.23Mixed
Use a 15-step maximum rather than a 12-step maximum (ablation)DACT-MAC · CLEVR overallAccuracy: 98.72%Improved
Use a 15-step gated MAC rather than its 12-step counterpart (ablation)MAC+Gate · CLEVRAccuracyHurt
Add adaptivity to MACMAC · GQA test-devAccuracyHurt

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