Research map / Learning theory and algorithms / Combinatorial and search optimization
Static Automatic Batching In TensorFlow
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| Replace sequential loop for Jacobian computation with auto-vectorized pfor loop | Jacobian computation (VGG16, LSTM) · GPU jacobian benchmark | jacobian rows per second: up to 60x at high output sizes | Helped |
| Use vectorized_map (pfor) for per-example gradients instead of sequential loop | Per-example gradients (LSTM, ConvMNIST) · GPU, batch size 256 LSTM | example gradients per second: outperforms tf.while_loop by a factor of 38 | Helped |
| Auto-vectorize pfor instead of tf.while_loop for forward pass of linear projection | Linear projection · GPU and CPU forward pass | GFLOPS vs batch size: iterative approaches 1 to 1.5 orders of magnitude slower | Helped |
| Auto-vectorize pfor for LSTM forward pass with variable-length sequences | LSTM · GPU and CPU tokens/sec | tokens per second | Mixed |
| Auto-vectorize pfor for ConvMNIST forward pass | ConvMNIST · GPU and CPU images/sec | images per second | Helped |
| Auto-vectorize pfor for GatedGraphNN | GatedGraphNN · GPU edges/sec | edges processed per second | Helped |
Closest papers by meaning
- On-the-fly Operation Batching in Dynamic Computation Graphs (NeurIPS 2017 · similarity 0.89)
- Synergistic Tensor and Pipeline Parallelism (NeurIPS 2025 · similarity 0.82)
- Einsum Benchmark: Enabling the Development of Next-Generation Tensor Execution Engines (NeurIPS 2024 · similarity 0.82)
- Geometry-aware training of factorized layers in tensor Tucker format (NeurIPS 2024 · similarity 0.82)
- Pfeife: Automatic Pipeline Parallelism for PyTorch (ICML 2025 · similarity 0.81)
- Accelerating neural network training: An analysis of the AlgoPerf competition (ICLR 2025 · similarity 0.81)
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.