Abstract
本文:Tensorflow
Github: https://github.com/tensorflow/tensorflow
Task: Detail on Tensorflow dataflow model
特点:
- operates at large scale and in heterogeneous environments
- supports a variety of applications
Method
- use dataflow graphs to represent computation, shared state, and operations that change that state
- maps the nodes of a dataflow graph(can across many machines/multiple devices) into a cluster
1. Intro
2. Background & motivation
2.1 Previous system: DistBelief
2.2 Design principles
2.3 Related Work
3. TensorFlow execution model
3.1 Dataflow graph elements
3.2 Partial and concurrent execution
3.3 Distributed execution
3.4 Dynamic control flow
4. Extensibility case studies
4.1 Differentiation and optimization
4.2 Training very large models
4.3 Fault tolerance
4.4 Synchronous replica coordination
5. Implementation
6. Evaluation
6.1 Single-machine benchmarks
6.2 Synchronous replica microbenchmark
6.3 Image classification
7. Conclusions
- Large-Scale TensorFlow CDeepFuzz learning Readinglarge-scale tensorflow cdeepfuzz learning combinatorial cdeepfuzz learning reading differentiation cdeepfuzz automatic reading pre-trained cdeepfuzz natural reading state-of-the-art cdeepfuzz the reading cdeepfuzz networks reading testing comprehensive cdeepfuzz compiler learning deepmutation cdeepfuzz mutation learning metamorphic cdeepfuzz compilers learning expedition weak-label learning reading