AI Paper Atlas

Research map / Probabilistic and scientific modeling / Variational and generative sampling

Reliable Measures of Spread in High Dimensional Latent Spaces

Anna Marbut, Katy McKinney-Bock, Travis J Wheeler

ICML 2023 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Tractable probabilistic models · 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
Use EEE instead of ACS or I(V) to compare spreadSynthetic latent-space distributions · Seven synthetic distributions across dimensionsagreement with expected relative spread orderingImproved
Use VRM instead of ACS or I(V) to compare spreadSynthetic latent-space distributions · Seven synthetic distributions across dimensionsagreement with expected relative spread orderingImproved
Use EEE rather than Eigenvalue Ratio (ER)Principal-component spread measurement · 100D cluster distributionsrelative spread discrimination: 0.5394 vs 0.0001; 0.6746 vs 0.0007Improved
Use average cosine similarity to assess spreadLatent-space distributions · Symmetric Clusters vs Sphererelative spread distinction: 0.0007 vs 0.0007Hurt
Use I(V) to distinguish Cone from other spherical distributionsLatent-space distributions · 100D Cone vs SphereI(V): 0.9944 vs 0.9988Hurt
Add random uniform noise to Word2Vec embeddingsPre-trained Word2Vec embeddings · Word2Vec Google NewsEEE: 0.4058 -> 0.0367Improved
Add random uniform noise to Word2Vec embeddingsPre-trained Word2Vec embeddings · Word2Vec Google NewsVRM: 0.1525 -> 0.0010Improved
Add random uniform noise to Word2Vec embeddingsPre-trained Word2Vec embeddings · Word2Vec Google NewsACS: 0.1317 -> 0.0022Improved
Add random uniform noise to Word2Vec embeddingsPre-trained Word2Vec embeddings · Word2Vec Google NewsI(V): 0.9251 -> 0.9996Improved
Use nearest-neighbor entropy measures based wholly on Euclidean distancesNNR and NNKL · Sphere, Cone, and Nested Shell, especially at high dimensionsrelative spread reliability: 100D NNR: 1.0023 vs 1.0022; 100D NNKL: 0.2558 vs 0.2853Hurt
Use KNN Overlap as a relative spread measureSynthetic latent-space distributions · High-dimensional synthetic distributionsrelative spread discrimination: 100D: 0.0078 vs 0.8491Hurt
Use KL-divergence-based measures for comparisons across dimensionalitiesDKLM and NNKL · Synthetic distributions across dimensionscross-dimensional comparability: 100D NNKL cluster scores are almost 300x the 2-dimensional counterpartsHurt

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