AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Feature Integration Spaces: Joint Training Reveals Dual Encoding in Neural Network Representations

Omar Claflin

AAAI 2026 · Architectures and efficiency › RNNs and normalization · approach: Sparse dictionary representations · 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
add a neural factorization machine on SAE features to predict reconstruction residualsSAE (OpenLLaMA-3B layer 16) · WikiText-103 validationreconstruction error: 0.3672 -> 0.2811Improved
train SAE and NFM jointly instead of sequentially (ablation)SAE + NFM · WikiText-103reconstruction improvement over SAE: 23% -> 41.3%Improved
train SAE and NFM jointly instead of sequentially (ablation)SAE + NFM · WikiText-103pathological KL divergence reduction: 29.8% -> 51.6%Improved
add nonlinear NFM component on top of linear component (ablation)Joint SAE + NFM · WikiText-103KL divergence reduction: 50.9% -> 51.9%No clear effect
nonlinear component alone (ablation)Joint SAE + NFM · WikiText-103share of total improvement: 16.5% with 3% of parametersImproved
add NFM integration features then clamp secondary feature 4022Secondary SAE feature 4022 · 2x2 formality x emotion logit effectsANOVA interaction: F=5.06, p=0.027Improved
clamp linear integration weights (ablation)NFM linear component · 2x2 formality x emotion logit effectsANOVA interactionNo clear effect
clamp non-sensitive secondary feature 1113 (ablation)Secondary SAE feature 1113 · 2x2 formality x emotion logit effectsANOVA: F=0.022, p=0.996No clear effect
secondary SAE trained directly on original residuals instead of NFM embeddings (ablation)Secondary SAE · reconstruction lossreconstruction lossNo clear effect
joint training produces low squared norm features specialized for integrationJoint SAE + NFM · SAE feature weight normsfeature integration contribution by weight: <0.2: 82.8% vs >0.2: 71.3%Improved

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