AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

User Embeddings are Superpositions of Interpretable Behavioral Modes

Jonathan Kozaczuk, Zheng Dong, Noyan Evirgen, Keivan Majidi, Eddie K Ng, Chris Porter, Qi Zhao, Jia Geng, Changmao Li, Brandon Feng

NeurIPS 2026 · Architectures and efficiency › RNNs and normalization · approach: Learned embeddings for retrieval · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes that user embeddings are sparse superpositions of interpretable behavioral modes, making dense user representations interpretable.

Problem

User embeddings used in recommendation and personalization are dense and opaque, limiting interpretability of the behavioral information they encode.

Approach

Treats user embeddings as sparse superpositions of interpretable behavioral modes, a structure the authors state is directly predicted by the superposition hypothesis from interpretability research.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Merge SAE features with decoder cosine similarity above 0.45 (ablation)SAE · Synthetic mean-pooled embeddings (m=128)median per-feature Pearson r: about 0.35 -> about 0.70Improved
Replace K-Means encoding with SAEBehavioral-mode decomposition · Open e-commerce category tasksLift: 1.85x -> 1.93xImproved
Replace K-Means encoding with SAEBehavioral-mode decomposition · MovieLens (matched size and L0)Lift and AUROCNo clear effect
Replace K-Means encoding with SAEBehavioral-mode decomposition · MovieLens SASReccoverage: 0.811 -> 0.977Improved
Use SAE instead of a text-embedding baselineNL audience retrieval · Open e-commerce life-event tasksLift: 2.13x -> 5.30xImproved
Use SAE instead of PCABehavioral-mode recovery · Synthetic SASRec embeddingsmodes recovered: 0.00 -> 0.97Improved
Add capacity to K-Means (512 components instead of 75) (ablation)K-Means encoding · Synthetic mean-pooled embeddingscode-GT alignmentNo clear effect
Freeze the decoder (soft-frozen decoder SAE) instead of training it (ablation)SAE · Synthetic datafeature recoveryMixed
Use FastICA instead of SAEMode decomposition · Synthetic dataper-feature monosemanticity: 0.53 -> 0.75Improved
Replace LLM used for interpretation with open-weight Llama 3.3 70B (ablation)Auto-interpretation · Open e-commerce (one seed)SAE retrievalNo clear effect

Experimental setup

Builds on: superposition hypothesis

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