AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Learning Dynamics of Attention: Human Prior for Interpretable Machine Reasoning

Wonjae Kim, Yoonho Lee

NeurIPS 2019 · Large language models › Language, safety and interpretability · approach: Reasoning mechanisms and failure modes · 4 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
Replace discrete MAC attention with continuous ODE-based attentionMAC · CLEVRAvg. accuracy: 98.9 -> 98.9No clear effect
Use 4 reasoning steps instead of 12MAC with DAFT · CLEVRAccuracy: 98.9% at step size 12 -> equal performance with step size 4No clear effect
Replace Dormand-Prince with Euler solver (ablation)DAFT MAC · CLEVRAccuracy: $98.6 \pm 0.2$ -> $98.7 \pm 0.2$No clear effect
Replace Dormand-Prince with Euler solver (ablation)DAFT MAC · CLEVRTime (ms): $365.5 \pm 12.5$ (2.37x) -> $167.9 \pm 1.7$ (1.09x)Improved
Replace Dormand-Prince with Runge-Kutta 4th order (ablation)DAFT MAC · CLEVRAccuracy: $98.9 \pm 0.2$ -> $98.9 \pm 0.2$No clear effect
Replace Dormand-Prince with Runge-Kutta 4th order (ablation)DAFT MAC · CLEVRTime (ms): $365.5 \pm 12.5$ (2.37x) -> $189.7 \pm 1.9$ (1.23x)Improved
Use a single-step Euler residual connection to restrict attention transitions (ablation)MAC · CLEVRaccuracyHurt
Add TLT to the objective with a Lagrange multiplier (ablation)MAC · CLEVRperformanceHurt
Reduce reasoning steps from four to two (ablation)DAFT MAC · CLEVR, Query Attributeaccuracy: 99.6 -> 99.3No clear effect
Reduce reasoning steps from four to two (ablation)MAC · CLEVR, Query Attributeaccuracy: 99.6 -> 97.5Hurt

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