Static Automatic Batching In TensorFlow
Authors: Ashish Agarwal
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Benchmarks demonstrate speedups of one to two orders of magnitude on a range of tasks, from Jacobian computation, to auto-batching Graph Neural Networks. |
| Researcher Affiliation | Industry | Google Inc.. Correspondence to: Ashish Agarwal <agarwal@google.com>. |
| Pseudocode | No | The paper describes algorithms and examples of code usage (e.g., Example 1 shows Python code) but does not include a formally labeled 'Pseudocode' or 'Algorithm' block for its core methods. |
| Open Source Code | No | The paper does not contain an explicit statement about the release of its source code or a link to a code repository for the methodology described. |
| Open Datasets | Yes | Conv Mnist setup uses a convolutional architecture as described in (tensorflow, 2016). VGG16 is as described in (Simonyan & Zisserman, 2014). |
| Dataset Splits | No | The paper does not explicitly provide specific training/validation/test dataset splits (e.g., percentages, sample counts, or explicit mention of standard splits for the datasets used) needed to reproduce the experiment. |
| Hardware Specification | Yes | Experiments were run on a 6 core Intel Xeon E5-1650 3.60GHz CPU with 64GB of RAM and a NVIDIA Maxwell Titan X GPU. |
| Software Dependencies | No | The paper is focused on extending TensorFlow but does not provide specific version numbers for TensorFlow or any other software dependencies used in the experiments. |
| Experiment Setup | Yes | Linear Projection is a simple setup which applies linear projection on input data. Inputs are randomly generated float vectors with shape [768]. Projection matrix is a constant [768, 768] matrix of floats. Conv Mnist setup uses a convolutional architecture as described in (tensorflow, 2016). It is a stack of two conv-relu-maxpool blocks followed by a linear-relu-dropout-linear block. Inputs are batches of [28, 28] images and output has shape [10]. LSTM state size is 256. |