Research map / Architectures and efficiency
Quantization: research map
785 papers on Quantization in Architectures and efficiency, from ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR (2016–2026), grouped into 2 clusters and 8 approaches. The busiest year so far is 2026.
What it is
Quantization stores and computes a network with fewer bits per number (for example 8 or 4 instead of 16 or 32), cutting memory and speeding up inference. Post-training quantization of large language models without retraining is the most active area.
Terms you will see on this page:
- Quantization
- representing weights or activations with low-precision numbers.
- Post-training quantization (PTQ)
- quantizing an already-trained model with little or no retraining.
- Mixed precision
- using different bit widths for different layers or values.
Start reading with the most cited papers here:
- Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference (CVPR 2018 · 3,940 citations)
- Mixed Precision Training (ICLR 2018 · 999 citations)
Within Architectures and efficiency, its share grew from 6.0% in 2023–24 to 9.6% in 2025–26 (118 → 341 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore Quantization in the interactive map
Working on something in this topic? Describe your idea in scime atlas to see which approach it falls under, the closest papers by meaning and how crowded the spot has become.
What changed recently
Approaches whose share of the topic moved most between 2023–24 and 2025–26 (ICML, NeurIPS, ICLR, CVPR and AAAI):
- Post-training quantization of LLMs rising: 21% → 31% of the topic (25 → 106 papers)
- KV cache quantization rising: 9% → 18% of the topic (11 → 61 papers)
- Low-precision training formats falling: 12% → 4% of the topic (14 → 12 papers)
- Post-training weight quantization falling: 14% → 10% of the topic (16 → 34 papers)
Papers from 2025–2026 that are already cited most:
- Automatic Joint Structured Pruning and Quantization for Efficient Neural Network Training and Compression (CVPR 2025 · 21 citations)
- Quantization without Tears (CVPR 2025 · 6 citations)
- Analog Foundation Models (NeurIPS 2025 · 2 citations)
Approaches and key papers
Post-training quantization · 480 papers
Keywords: post quantization, ptq, compression
Approaches in this cluster:
- Post-training weight quantization (134 papers)
Compresses weights by quantization with sharpness control, loss-aware objectives, and quantization noise. - Post-training quantization of LLMs (144 papers)
Handles outliers and low-bit errors when quantizing large language models after training. - Post-training quantization of ViTs (128 papers)
Reduces quantization error in vision transformers using Hessian, Fisher, and outlier-aware reconstruction. - KV cache quantization (74 papers)
Compresses key-value caches with vector and mixed-precision quantization for long-context inference.
Most cited and most cited since 2024:
- QLoRA: Efficient Finetuning of Quantized LLMs (NeurIPS 2023 · 824 citations)
- Soft-to-Hard Vector Quantization for End-to-End Learning Compressible Representations (NeurIPS 2017 · 358 citations)
- OWQ: Outlier-Aware Weight Quantization for Efficient Fine-Tuning and Inference of Large Language Models (AAAI 2024 · 72 citations)
- KVQuant: Towards 10 Million Context Length LLM Inference with KV Cache Quantization (NeurIPS 2024 · 51 citations)
Low-precision formats and mixed precision · 305 papers
Keywords: mixed precision, floating, point
Approaches in this cluster:
- Mixed-precision quantization (134 papers)
Assigns bit-widths per layer for low-bit network training and inference with learnable quantizers. - Low-precision training formats (63 papers)
Designs 8-bit floating-point formats and accumulation schemes for low-precision neural network training. - Ultra-low-bit LLM quantization (66 papers)
Pushes language models to FP4, 1-bit, and multi-bit formats with scaling-law analysis. - Binary neural networks (42 papers)
Binarizes weights and activations with information-retention and ensemble techniques.
Most cited and most cited since 2024:
- Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference (CVPR 2018 · 3,940 citations)
- Mixed Precision Training (ICLR 2018 · 999 citations)
- JointSQ: Joint Sparsification-Quantization for Distributed Learning (CVPR 2024 · 18 citations)
- Towards Efficient Verification of Quantized Neural Networks (AAAI 2024 · 17 citations)
Related topics in Architectures and efficiency
- Pruning and efficient tuning (1,335)
- Attention and transformers (2,598)
- Neural architecture search (548)
- RNNs and normalization (3,071)
- Spiking networks (333)
- Mixture of experts (318)
- CNN design (1,051)
