AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

DataMUX: Data Multiplexing for Neural Networks

Vishvak Murahari, Carlos E Jimenez, Runzhe Yang, Karthik R Narasimhan

NeurIPS 2022 · Architectures and efficiency › Attention and transformers · approach: Transformer expressivity and circuits · 9 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 Hadamard/Ortho multiplexing instead of unstructured mixing for larger N (ablation)T-MUX · MNLI/NERaccuracyNo clear effect
reduce self-attention heads from 12 to 2 (ablation)T-MUX · MNLI/NERaccuracyNo clear effect
use binary masking multiplexing instead of Gaussian Hadamard (ablation)T-MUX · retrieval/MNLI/NERaccuracyHurt
unfreeze (learn) Gaussian multiplexing vectors (ablation)T-MUX · retrieval/MNLI/NERaccuracyNo clear effect
use MLP demultiplexing instead of index embeddings (ablation)T-MUX · NLP tasksaccuracyHurt
add LowRank multiplexing instead of Ortho (ablation)MLP multiplexing · MNISTaccuracy: +5% at N=8Improved
use nonlinear conv multiplexing instead of Ortho for CNN (ablation)CNN multiplexing · MNISTaccuracy: 56 -> above 80 (N<=4 vs Ortho at N=8 not same N)Improved
increase number of multiplexed instances N from small to 40T-MUX · MNLI/NERaccuracy: 10%-15% drop at N=40Hurt
multiplex 40 instances for throughputT-MUX · MNLIthroughput: 18x at N=40, 11x at N=20Improved

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