Research map / Architectures and efficiency / Neural architecture search
Auto-NBA: Efficient and Effective Search Over the Joint Space of Networks, Bitwidths, and Accelerators
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 |
|---|---|---|---|
| Heterogeneous sampling (soft GS for weight updates, hard GS for precision updates) vs. hard GS sampling for both (ablation) | Auto-NBA joint search of network and precision · CIFAR-100 FPGA, DSP limit 512 | accuracy vs FPS trade-off: +1.95% accuracy at 1.75x FPS vs sequential baseline (heterogeneous sampling ablation shown only in figure) | Helped |
| Joint network-accelerator-precision search vs sequential optimization (search network/precision first, then accelerator) (ablation) | Auto-NBA · CIFAR-100 FPGA | accuracy vs FPS: +1.95% accuracy, 1.75x FPS | Helped |
| Searching mixed precision (vs fixed 16-bit) in co-search (ablation) | Auto-NBA · CIFAR-10 FPGA vs HS-Co-Opt/BSW | FPS: up to 6.79x FPS | Helped |
| Auto-NBA co-search vs NASAIC/ASIC co-exploration baselines | ASIC accelerator design · CIFAR-10 ASIC | EDP: 371.56x to 756.88x EDP reduction | Helped |
| Auto-NBA vs NHAS under 4-bit fixed precision | network-accelerator co-search · ASIC, NHAS comparison | accuracy and latency: +0.96% accuracy, 20.9% latency reduction | Helped |
| Auto-NBA search vs one-shot NAS methods (search time) | joint search pipeline · ImageNet search time | GPU hours: 3.6x to 30x less search time | Helped |
Closest papers by meaning
- AutoShrink: A Topology-Aware NAS for Discovering Efficient Neural Architecture (AAAI 2020 · similarity 0.87)
- Searching for Fast Model Families on Datacenter Accelerators (CVPR 2021 · similarity 0.87)
- NetAdaptV2: Efficient Neural Architecture Search With Fast Super-Network Training and Architecture Optimization (CVPR 2021 · similarity 0.86)
- Efficient Architecture Search for Diverse Tasks (NeurIPS 2022 · similarity 0.86)
- Distribution Consistent Neural Architecture Search (CVPR 2022 · similarity 0.86)
- AutoGO: Automated Computation Graph Optimization for Neural Network Evolution (NeurIPS 2023 · 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.