Research map / Architectures and efficiency / RNNs and normalization
Learning to Optimize Tensor Programs
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 |
|---|---|---|---|
| Use a statistical cost model instead of black-box search alone | AutoTVM · ResNet-18 convolution workloads | operator speed relative to random search: 2 × | Helped |
| Use rank-based rather than regression-based cost-model training (ablation) | AutoTVM · ResNet-18 convolution workloads | search performance | Helped |
| Add diversity-aware candidate selection (ablation) | AutoTVM · ResNet-18 convolution workloads, most evaluated workloads | search performance | No clear effect |
| Add diversity-aware candidate selection (ablation) | AutoTVM · C6 | search performance | Helped |
| Use uncertainty-aware acquisition functions (ablation) | AutoTVM · ResNet-18 convolution workloads | search performance | No clear effect |
| Transfer historical workload data using a transferable representation | AutoTVM · Target workloads C7, C8, C9 | optimization speedup: 2 × to 10 × | Helped |
| Use configuration-space features for cross-domain transfer rather than within-domain optimization (ablation) | AutoTVM transfer model · Different workload domains | transfer effectiveness | Hurt |
| Use flattened AST features to transfer across convolution workloads (ablation) | AutoTVM transfer model · Convolution workloads | transfer effectiveness | Helped |
| Use flattened AST features to transfer across operator types (ablation) | AutoTVM transfer model · Different operator types | transfer effectiveness | Hurt |
| Use context-relation features (ablation) | AutoTVM transfer model · Different operator types | transfer effectiveness | Helped |
| Use transferable representations for cross-device transfer | AutoTVM · ARM Mali GPU to ARM Cortex-A53 | transfer effectiveness | Helped |
| Use AutoTVM tensor-program optimization and operator fusion instead of existing framework/library baselines | End-to-end DL inference workloads · Real-world end-to-end DL inference workloads | performance improvement: 1.2× to 3.8× | Helped |
Closest papers by meaning
- Tensorion: A Tensor-Aware Generalization of the Muon Optimizer (NeurIPS 2026 · similarity 0.88)
- COGNATE: Acceleration of Sparse Tensor Programs on Emerging Hardware using Transfer Learning (ICML 2025 · similarity 0.87)
- Convolutions and More as Einsum: A Tensor Network Perspective with Advances for Second-Order Methods (NeurIPS 2024 · similarity 0.86)
- Einsum Benchmark: Enabling the Development of Next-Generation Tensor Execution Engines (NeurIPS 2024 · similarity 0.86)
- Cost-Efficient LLM Training with Lifetime-Aware Tensor Offloading via GPUDirect Storage (NeurIPS 2025 · similarity 0.86)
- SpeedLoader: An I/O efficient scheme for heterogeneous and distributed LLM operation (NeurIPS 2024 · similarity 0.86)
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.