AI Paper Atlas

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

Computing Linear Restrictions of Neural Networks

Matthew Sotoudeh, Aditya V Thakur

NeurIPS 2019 · Learning theory and algorithms › Prediction and decision losses · approach: Theory of neural network sensitivity · 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 exact line restrictions to compute integrated gradients instead of relying on sampled estimatesIntegrated Gradients on CIFAR10 networks · CIFAR10mean relative IG error (%): 24.95; 24.05; 45.34Mixed
Use trapezoidal sampling instead of left-Riemann samplingIntegrated Gradients on CIFAR10 networks · CIFAR10, convsmallsamples to reach 5% relative IG error: 136.74 -> 91.67Improved
Use trapezoidal sampling instead of left-Riemann samplingIntegrated Gradients on CIFAR10 networks · CIFAR10, convmediumsamples to reach 5% relative IG error: 150.31 -> 91.57Improved
Use trapezoidal sampling instead of left-Riemann samplingIntegrated Gradients on CIFAR10 networks · CIFAR10, convbigsamples to reach 5% relative IG error: 269.43 -> 222.79Improved
Use trapezoidal sampling rather than left or right sumsIntegrated Gradients on CIFAR10 networks · CIFAR10samples to reach 5% relative IG errorMixed
Compare adversarial directions with random directions using exact line partitionsNeural networks trained on MNIST · MNISTlinear-partition density: 1.36; 0.98; 1.22Mixed
Compare adversarial directions with random directions using exact line partitionsNeural networks trained on CIFAR10 · CIFAR10linear-partition density: 1.78; 1.67; 1.84Improved
Train with DiffAI instead of normal trainingMNIST network · MNIST, FGSM directionlinear-partition density: 3.05Improved
Train with DiffAI instead of normal trainingMNIST network · MNIST, random directionlinear-partition density: 2.50Improved
Train with DiffAI instead of normal trainingCIFAR10 network · CIFAR10, FGSM directionlinear-partition density: 3.37Improved
Train with DiffAI instead of normal trainingCIFAR10 network · CIFAR10, random directionlinear-partition density: 3.43Improved
Train with PGD instead of normal trainingMNIST network · MNIST, FGSM and random directionslinear-partition density: 0.88; 0.80Mixed

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