Research map / Architectures and efficiency / Spiking networks
ELSE: Efficient Deep Neural Network Inference through Line-based Sparsity Exploration
Zeqi Zhu, Alberto Garcia-Ortiz, Luc Waeijen, Egor Bondarev, Arash Pourtaherian, Orlando Moreira
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 |
|---|---|---|---|
| add line-based delta-sigma suppression on top of ReLU (ablation) | ELSE · UA-DETRAC SSD-MNV1 | GMAC: 0.59 -> 0.46 | Improved |
| add activation quantization (ablation) | ELSE · UA-DETRAC SSD-MNV1 | GMAC: 0.46 -> 0.34 | Improved |
| add event-aware sparsity penalty (ablation) | ELSE · UA-DETRAC SSD-MNV1 | mAP@0.5 change: -0.62% | Hurt |
| add MAC-aware penalty weighting (ablation) | ELSE · UA-DETRAC SSD-MNV1 | GMAC: 0.31 -> 0.23 | Improved |
| add line-based penalty (ablation) | ELSE · UA-DETRAC SSD-MNV1 | GMAC: 0.23 -> 0.19 | Improved |
| mix ELSE with temporal CATS on memory-heavy layers | ELSE · UA-DETRAC SSD-MNV1 | state memory: 20.20 MB -> 9.22 MB vs CATS | Improved |
| mix ELSE with spatial STAR | ELSE · MPII RN50 | GMAC reduction ratio: 5.75x -> 8.83x | Improved |
| use ELSE alone vs STAR on ResNet-50 pose estimation | ELSE · MPII RN50 | GMAC reduction ratio: 6.70x -> 5.75x | Hurt |
| mix ELSE with temporal CATS vs full temporal CATS on MNV2-SSDLite computation | ELSE · UA-DETRAC MNV2-SSDLite | GMAC reduction ratio: 4.51x -> 3.99x with 31.19 MB -> 14.56 MB | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- CATS — CATs: Cost Aggregation Transformers for Visual Correspondence (NeurIPS 2021) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CoreMatching: A Co-adaptive Sparse Inference Framework with Token and Neuron Pruning for Comprehensive Acceleration of Vision-Language Models (ICML 2025 · similarity 0.84)
- VLM in a flash: I/O-Efficient Sparsification of Vision-Language Model via Neuron Chunking (NeurIPS 2025 · similarity 0.84)
- MEET: Towards Memory-Efficient Temporal Sparse Deep Neural Networks (CVPR 2025 · similarity 0.83)
- Polar Sparsity: High Throughput Batched LLM Inferencing with Scalable Contextual Sparsity (NeurIPS 2025 · similarity 0.83)
- Effective Interplay between Sparsity and Quantization: From Theory to Practice (ICLR 2025 · similarity 0.83)
- ReLU Strikes Back: Exploiting Activation Sparsity in Large Language Models (ICLR 2024 · similarity 0.83)
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.
