DistBelief
E97079
DistBelief was Google’s early large-scale distributed deep learning system that laid the groundwork for the development of TensorFlow.
Statements (47)
| Predicate | Object |
|---|---|
| instanceOf |
distributed deep learning system
ⓘ
machine learning framework ⓘ |
| architecture | parameter server architecture ⓘ |
| authorAssociated |
Andrew Ng
ⓘ
Geoffrey Hinton ⓘ Greg Corrado NERFINISHED ⓘ Jeff Dean ⓘ Quoc V. Le ⓘ |
| basedOn |
data parallelism
ⓘ
deep neural networks ⓘ distributed computing ⓘ model parallelism ⓘ |
| capability | training models with billions of parameters ⓘ |
| developer |
Google
ⓘ
Google Brain ⓘ
surface form:
Google Brain team
|
| feature |
asynchronous stochastic gradient descent
ⓘ
data sharding ⓘ distributed training across many machines ⓘ model sharding ⓘ support for very large neural networks ⓘ |
| hardwareSupport |
CPU clusters
ⓘ
GPU acceleration ⓘ |
| influenced |
design of distributed ML systems
ⓘ
large-scale deep learning practices ⓘ |
| inspired | TensorFlow NERFINISHED ⓘ |
| license | proprietary ⓘ |
| notableExperiment | unsupervised learning of high-level features from YouTube videos ⓘ |
| notablePublication |
Large-Scale Distributed Deep Networks
ⓘ
surface form:
Building High-level Features Using Large Scale Unsupervised Learning
Large-Scale Distributed Deep Networks ⓘ
surface form:
Large Scale Distributed Deep Networks
|
| organization | Google Brain ⓘ |
| organizationUnit | Google Research ⓘ |
| predecessorOf | TensorFlow NERFINISHED ⓘ |
| programmingLanguage |
C++
ⓘ
Python ⓘ |
| researchDomain |
artificial intelligence
ⓘ
deep learning ⓘ |
| scale | thousands of machines ⓘ |
| status | internal system ⓘ |
| successor | TensorFlow NERFINISHED ⓘ |
| timePeriod | early 2010s ⓘ |
| trainingAlgorithm | stochastic gradient descent ⓘ |
| usedBy | internal Google products ⓘ |
| usedFor |
computer vision research
ⓘ
large-scale deep learning ⓘ natural language processing research ⓘ speech recognition ⓘ training deep neural networks ⓘ |
Referenced by (1)
Full triples — surface form annotated when it differs from this entity's canonical label.