Research map / Architectures and efficiency / Pruning and efficient tuning
Joint pruning and quantization
Combines pruning, quantization, and low-rank decomposition for model compression.
78 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Pruning and efficient tuning, keyword cluster “sparsity · compression · structured”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Few recent reports The tracked benchmarks have at most one report from 2025–26 that could beat the old record, too few to say whether they are still moving.
19 papers in 2023–24 and 22 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 6.8% to 4.0%.
No longer reported by papers here: CIFAR-10 (Top-1 accuracy, last in 2023); ImageNet (Top-1 accuracy, last in 2023).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Dynamic Network Surgery for Efficient DNNs (NeurIPS 2016 · 140 citations)
- 2017 · On Compressing Deep Models by Low Rank and Sparse Decomposition (CVPR 2017 · 428 citations)
- 2018 · CLIP-Q: Deep Network Compression Learning by In-Parallel Pruning-Quantization (CVPR 2018 · 229 citations)
- 2019 · Cascaded Projection: End-To-End Network Compression and Acceleration (CVPR 2019 · 36 citations)
- 2020 · Dynamic Model Pruning with Feedback (ICLR 2020 · 98 citations)
- 2021 · Towards Compact CNNs via Collaborative Compression (CVPR 2021 · 81 citations)
- 2022 · BATUDE: Budget-Aware Neural Network Compression Based on Tucker Decomposition (AAAI 2022 · 23 citations)
- 2023 · HALOC: Hardware-Aware Automatic Low-Rank Compression for Compact Neural Networks (AAAI 2023 · 15 citations)
- 2024 · Data Shunt: Collaboration of Small and Large Models for Lower Costs and Better Performance (AAAI 2024 · 15 citations)
- 2025 · Temporal Action Detection Model Compression by Progressive Block Drop (CVPR 2025 · 5 citations)
- 2026 · Advancing SVD-based LLM Compression via Layer-Wise Error Model Search (ICML 2026 · 0 citations)
Earliest papers here
- Dynamic Network Surgery for Efficient DNNs (NeurIPS 2016 · 140 citations)
- On Compressing Deep Models by Low Rank and Sparse Decomposition (CVPR 2017 · 428 citations)
- Compression-aware Training of Deep Networks (NeurIPS 2017 · 133 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
CIFAR-10 · Top-1 accuracy (higher is better)
6 papers, 2020–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 92.96 | Neural Epitome Search for Architecture-Agnostic Network Compression (ICLR 2020 · 6 citations) |
| 2021 | 95.18 | Towards Compact CNNs via Collaborative Compression (CVPR 2021 · 81 citations) |
ImageNet · Top-1 accuracy (higher is better)
6 papers, 2017–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 75.2 | Compression-aware Training of Deep Networks (NeurIPS 2017 · 133 citations) |
| 2020 | 77.5 | Neural Epitome Search for Architecture-Agnostic Network Compression (ICLR 2020 · 6 citations) |
| 2023 | 81.1 | OTOv2: Automatic, Generic, User-Friendly (ICLR 2023 · 7 citations) |
Cifar10 · accuracy (higher is better)
5 papers, 2019–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 94.14 | Cascaded Projection: End-To-End Network Compression and Acceleration (CVPR 2019 · 36 citations) |
| 2023 | 95.3 | Integral Neural Networks (CVPR 2023 · 10 citations) |
| 2025 | 95.7 | Scale Efficient Training for Large Datasets (CVPR 2025 · 2 citations) |
What has been tried
60 papers here report 635 measured changes (a component added, removed or swapped, compared with a baseline): helped 66%, no clear effect 10%, hurt 22%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Remove 2-Stage Confidence Distillation | CIFAR-100-LT · overall accuracy | 63.25% -> 62.11% | Hurt | Data Shunt: Collaboration of Small and Large Models for Lower Costs and Better Performance (AAAI 2024 · 15 citations) |
| Replace Taylor importance with disturbed Taylor importance | SA-1B, 77% pruning · MIoU | 62.04% -> 62.31% | Helped | SlimSAM: 0.1% Data Makes Segment Anything Slim (NeurIPS 2024 · 11 citations) |
| Drop 4 blocks | THUMOS14 · mAP (tIoU=0.5) | 70.43 -> 69.65 (↓0.78) | Hurt | Temporal Action Detection Model Compression by Progressive Block Drop (CVPR 2025 · 5 citations) |
| Remove the (s, k)-mer encoder | SnSt · compression ratio (bits/base) | 1.860 -> 1.879 | Hurt | Genomics Data Lossless Compression with (S, K)-Mer Encoding and Deep Neural Networks (AAAI 2025 · 2 citations) |
| add certainty-based sample selection | CIFAR10-C · mean online error (%) | 18.1 -> 18.0 | Helped | SURGEON: Memory-Adaptive Fully Test-Time Adaptation via Dynamic Activation Sparsity (CVPR 2025 · 2 citations) |
| use easy-to-hard cluster ordering instead of hard-to-easy ordering | ToCa, NoCaps Val Overall · CIDEr | 69.6 -> 71.5 | Helped | Scale Efficient Training for Large Datasets (CVPR 2025 · 2 citations) |
| Add pruned-reduce to BFS without pruning | LOGISTICS · execution time (s) | 20.80 -> 2.61 | Helped | Block-Level Goal Recognition Design (AAAI 2024 · 1 citation) |
| Remove hierarchical group lasso | CIFAR-10 · accuracy | 94.19{ ±0.39} -> 94.04{ ±0.10} | No clear effect | Maestro: Uncovering Low-Rank Structures via Trainable Decomposition (ICML 2024 · 1 citation) |
| optimize the score map rather than freezing it | GLDv2-test, RParis, ROxford (average), φ20% self-test · mAP | 43.00 -> 45.63 | Helped | Learning Compatible Multi-Prize Subnetworks for Asymmetric Retrieval (CVPR 2025 · 0 citations) |
| Set subsampled-matrix aspect ratio to 1.0 | CIFAR100 · test accuracy | 79.13{ ±0.158} (Q=0.5) -> 79.53{ ±0.177} (Q=1.0) | Helped | Eigenspectrum Analysis of Neural Networks without Aspect Ratio Bias (ICML 2025 · 0 citations) |
| Increase the allowed performance loss Δ from 0.25 to 0.5 | Adult · compression ratio | 23.0 -> 31.8 | Helped | Compressing tree ensembles through Level-wise Optimization and Pruning (ICML 2025 · 0 citations) |
| Add group-wise pruning after quantization to DARE | 8 image classification tasks, 97% sparsity · average accuracy | 89.7 -> 89.7 | No clear effect | Breaking the Compression Ceiling: Data-Free Pipeline for Ultra-Efficient Delta Compression (NeurIPS 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 11 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-10 · 2 of 11 papers
- CIFAR100 · 2 of 11 papers
Baselines
- Magnitude pruning · 2 of 11 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 41 papers; from OpenAlex and the papers' affiliations).
- ThirdAI Corp · 3 papers
- Rice University · 3 papers
- Huawei · 2 papers
- Harbin Institute of Technology · 2 papers
- Tsinghua University · 2 papers
- Northwestern Polytechnical University · 2 papers
- Carnegie Mellon University · 2 papers
- MBZUAI · 2 papers
- KU Leuven · 2 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Structured network pruning (162 papers)
- Sparse training from scratch (95 papers)
- Layer-wise pruning of LLMs (91 papers)
Elsewhere in the topic
- Low-rank compression and merging (226 papers)
- Adapter-based parameter-efficient tuning (149 papers)
- LoRA variants and theory (121 papers)
- Parameter-efficient transfer analysis (97 papers)
- Efficient ViT adaptation (62 papers)
- Lottery ticket discovery (42 papers)
- Pruning at initialization (24 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).