AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

Interpretable by Design: Learning Predictors by Composing Interpretable Queries

Aditya Chattopadhyay, Stewart Slocum, Benjamin D. Haeffele, René Vidal, Donald Geman

TPAMI 2022 · Learning theory and algorithms › Prediction and decision losses · approach: Provable post-hoc explanations · 27 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 IP generative posterior over query-answers instead of CART decision tree on same queriesInformation Pursuit · MNISTaccuracy: 90.23% -> 96.78%Improved
use IP instead of decision treeInformation Pursuit · HuffPostaccuracy: 63.00% -> 71.21%Improved
stop IP early (about 2% of queries) versus asking all queries (MAP using Q) (ablation)Information Pursuit · MNISTaccuracy: 99.05% -> 96.78%Hurt
restrict black-box model to interpretable query featuresBlack-box · Fashion-MNISTaccuracy: 96.70% -> 88.43%Hurt
use IP explanations instead of Integrated Gradients top patchesInformation Pursuit · MNISTMAP accuracy: 78.48% -> 96.78%Improved
use IP explanations instead of DeepSHAPInformation Pursuit · MNISTMAP accuracy: 87.98% -> 96.78%Improved
use IP explanations instead of DeepSHAPInformation Pursuit · Fashion-MNISTMAP accuracy: 88.36% -> 85.60%Hurt
use IP explanations instead of Integrated Gradients (absolute)Information Pursuit · KMNISTMAP accuracy: 84.72% -> 91.02%Improved
increase patch size from 1x1 to 3x3 (ablation)Information Pursuit · MNISTavg queries to 99% confidence: 21.1 -> 5.2Improved
increase patch size from 3x3 to 4x4 (ablation)Information Pursuit · MNISTavg pixels used: 54.7 pixels for 4x4 vs 21.1 for 1x1Hurt
add query set complexity (MNIST to FashionMNIST)Information Pursuit · 3x3 patch queriesavg queries: 5.2 -> 14.5Hurt

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