AI Paper Atlas

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

NanoFlow: Scalable Normalizing Flows with Sublinear Parameter Complexity

Sang-gil Lee, Sungwon Kim, Sungroh Yoon

NeurIPS 2020 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Normalizing flow training and design · 7 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 plain shared estimator (NanoFlow-naive) instead of separate per-flow WaveFlow (ablation)WaveFlow · LJ Speechlog-likelihood: 5.2059 -> 5.1247Hurt
add flow indication embedding and decomposition (NanoFlow vs NanoFlow-naive) (ablation)NanoFlow · LJ Speechlog-likelihood: 5.1247 -> 5.1586Improved
NanoFlow vs WaveFlow 128 ch with about 1/8 of the parameters (ablation)NanoFlow · LJ Speechlog-likelihood: 5.2059 -> 5.1586Hurt
double the number of flows from 8 to 16 (ablation)NanoFlow · LJ Speechlog-likelihood: 5.1586 -> 5.1873Improved
double the number of flows from 8 to 16 (ablation)NanoFlow · LJ Speechsynthesis speed (kHz): 362 -> 186Hurt
add flow indication embedding to NanoFlow-naive (ablation)NanoFlow-naive · LJ SpeechMOS: 3.23 -> 3.63Improved
increase number of groups G from 4 to 64 (ablation)NanoFlow · LJ SpeechLL ratio: 0.0272 -> 0.0476Mixed
increase groups G from 4 to 64 (more autoregressive bias) (ablation)WaveFlow · LJ Speechlog-likelihood: 4.9785 -> 5.1586Improved
CIFAR10: NanoFlow-naive vs decomposition only (ablation)NanoFlow · CIFAR10bits per dim: 3.40 -> 3.32Improved
add flow indication embedding on top of decomposition (ablation)NanoFlow · CIFAR10bits per dim: 3.32 -> 3.27Improved
NanoFlow vs Glow-large with 28x more parameters (ablation)NanoFlow · CIFAR10bits per dim: 3.30 -> 3.27Improved
scale NanoFlow to 48 flows per scale (ablation)NanoFlow · CIFAR10bits per dim: 3.27 -> 3.25Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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