AI Paper Atlas

Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes

Learning of Gaussian Processes in Distributed and Communication Limited Systems

Mostafa Tavassolipour, Seyed Abolfazl Motahari, Mohammad Taghi Manzuri

TPAMI 2019 · Probabilistic and scientific modeling › Uncertainty and Gaussian processes · approach: PAC-Bayes and decoupled GP learning

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 proposed dimension reduction instead of PCA when machines hold different covariancesDimension reduction for inner-product estimation · Gaussian data and MNIST digits 6 and 7 split by classinner-product distortionImproved
use proposed dimension reduction instead of PCA when machines hold identically distributed dataDimension reduction for inner-product estimation · Gaussian data and MNIST split uniformlyinner-product distortionNo clear effect
use per-symbol quantization instead of dimension reductionInner-product coding schemes · Synthetic 20-dim Gaussianbits per dimension until distortion near zero: 5 vs 12.5 bits per dimensionImproved
quantize GP training data at 1 bit per sample instead of higher rates (ablation)GP trained on quantized data · Synthetic 1-D GP, N=200posterior fidelity to true GPHurt
raise quantization rate to R>=6 bits (ablation)GP trained on quantized data · Synthetic 1-D GP, N=200posterior fidelity to true GP: R >= 6Improved
transmit heavily quantized data (low rate) versus transmitting nothing (ablation)Broadcast GP · SARCOS, KIN40K, ABALONEregression error (SMSE)Hurt
use broadcast quantized GP instead of rBCM at about 16 bits per sampleDistributed GP · SARCOS (linear kernel)SMSE: crosses rBCM at R = 16Improved
use broadcast quantized GP instead of BCM and rBCM at about 20 bits per sampleDistributed GP · KIN40K (SE kernel)SMSE: R around 20Improved
quantize local inducing variables instead of raw samplesSingle-center GP · KIN40KSMSEImproved

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