AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Neural Sculpting: Uncovering hierarchically modular task structure in neural networks through pruning and network analysis

Shreyas Malakarjun Patil, Loizos Michael, Constantine Dovrolis

NeurIPS 2023 · Architectures and efficiency › RNNs and normalization · approach: Mechanistic circuit interpretability · 0 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
Train with plain dense training instead of pruningStandard NN training · Input separable function graphedge-weight product test (null hypothesis rejected)Hurt
Standard training on reused sub-functionStandard NN training · Reused sub-function graphunits covering 90% of path weightHurt
Add iterative unit and edge pruning to trainingNN training · Boolean function graphsmodule detection success rateImproved
Increase NN depth beyond the function graph depthNeural Sculpting · Boolean function graphsmodule detection success rateImproved
Increase number of output nodes reusing the gate sub-functionNeural Sculpting · Single reused sub-function graphsuccess rateImproved
Increase input overlap between two separable sub-functionsNeural Sculpting · Two input separable sub-functions graphsuccess rate of input module detectionHurt
Decrease number of times each sub-function is usedNeural Sculpting · Two input separable sub-functions graphsuccess rateHurt
Use NN depth below the number of hierarchical levels (5)Neural Sculpting · Hierarchical function graphsub-function recoveryHurt
Use NN depth at least the number of hierarchical levels (5)Neural Sculpting · Hierarchical function graphsub-function recoveryImproved
Use two-digit MNIST classification taskNeural Sculpting · Two-MNIST-digit classificationcompletely separable modules: 33 out of 36Improved
Increase NN depthNeural Sculpting · Hierarchical MNIST taskrecovery of two input separable modulesImproved
Increase NN depthNeural Sculpting · Hierarchical MNIST taskmiddle layer unit separabilityNo clear effect

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