Research map / Architectures and efficiency / RNNs and normalization
Unrolled Memory Inner-Products: An Abstract GPU Operator for Efficient Vision-Related Computations
Yu-Sheng Lin, Wei-Chao Chen, Shao-Yi Chien
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 UMI instead of OpenCV separable filter, short kernel | UMI GPU kernel · Separable filter k=3 | speedup ratio: 0.35 | Hurt |
| use UMI instead of OpenCV separable filter, long kernel | UMI GPU kernel · Separable filter k=30 | speedup ratio: 1.42 | Improved |
| use UMI instead of Parboil motion estimation | UMI GPU kernel · Motion estimation | speedup ratio: 6.51 | Improved |
| use UMI instead of Caffe for forward propagation, 3 + 1s | UMI GPU kernel · Forward propagation | speedup ratio: 19.9 | Improved |
| use UMI instead of Caffe for forward propagation, 9 + 1s | UMI GPU kernel · Forward propagation | speedup ratio: 26.4 | Improved |
| use UMI instead of Caffe for forward propagation, 3 + 2s | UMI GPU kernel · Forward propagation | speedup ratio: 1.80 | Improved |
| use cuDNN instead of UMI for forward propagation, 3 + 1s | cuDNN · Forward propagation | speedup ratio: 19.9 -> 100 | Improved |
| express motion estimation with UMI instead of naive CPU code | UMI · Motion estimation | operator tokens: 45 -> 11 | Improved |
| express forward propagation with UMI instead of naive CPU code | UMI · Forward propagation | identifier tokens: 110 -> 53 | Improved |
| express bilateral filter with UMI instead of naive CPU code | UMI · Bilateral filter | literal tokens: 7 -> 61 | Hurt |
| express GEMM with UMI instead of naive CPU code | UMI · GEMM | identifier tokens: 34 -> 34 | No clear effect |
Closest papers by meaning
- FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision (NeurIPS 2024 · similarity 0.84)
- Efficient Unrolled Networks for Large-Scale 3D Inverse Problems (CVPR 2026 · similarity 0.83)
- Run, Don't Walk: Chasing Higher FLOPS for Faster Neural Networks (CVPR 2023 · similarity 0.83)
- DEPrune: Depth-wise Separable Convolution Pruning for Maximizing GPU Parallelism (NeurIPS 2024 · similarity 0.83)
- FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning (ICLR 2024 · similarity 0.82)
- HiRED: Attention-Guided Token Dropping for Efficient Inference of High-Resolution Vision-Language Models (AAAI 2025 · similarity 0.82)
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.
