AI Paper Atlas

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

Proppo: a Message Passing Framework for Customizable and Composable Learning Algorithms

Paavo Parmas, Takuma Seno

NeurIPS 2022 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Sequential Monte Carlo for inference · 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 total propagation (step-wise inverse-variance weighting of LR and RP) instead of choosing the best of RP or LRProppo gradient estimators · Chaotic RNN, beta around 2.5gradient accuracy ratio: up to around 100 timesImproved
use reparameterization gradients instead of likelihood ratio at low inverse temperatureMC gradient estimators · Chaotic RNN, low betagradient varianceImproved
increase inverse temperature beta into chaotic regime for reparameterization gradientsRP gradient estimator · Chaotic RNN, beta around 2.5gradient varianceHurt
increase horizon H from 10 to 100 for RP gradientsRP gradient estimator · Chaotic RNNgradient varianceHurt
replace step-wise TP with naive combination of separately computed LR and RP gradients (ablation)Total propagation · Chaotic RNNgradient accuracy ratioHurt
use TP in this RNN instead of prior cart-pole settingTotal propagation vs LR · Prior work comparisonimprovement over LR: 3 times -> multiple orders of magnitudeImproved

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