Research map / Architectures and efficiency / Attention and transformers
Adaptive Visual Autoregressive Acceleration via Dual-Linkage Entropy Analysis
Yu Zhang, Jingyi Liu, Feng Liu, Duoqian Miao, Qi Zhang, Kexue Fu, Changwei Wang, Longbing Cao
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 entropy instead of attention to select tokens (ablation) | NOVA on Infinity-2B · GenEval | overall: 0.69 -> 0.72 | Improved |
| use entropy instead of MSE to select tokens (ablation) | NOVA on Infinity-2B · GenEval | overall: 0.71 -> 0.72 | Improved |
| adaptive activation scale instead of fixed 6th start scale (ablation) | NOVA on Infinity-2B · GenEval | overall: 0.62 -> 0.72 | Improved |
| adaptive activation scale instead of fixed 10th start scale (ablation) | NOVA on Infinity-2B · GenEval / speedup | speedup: 1.63x -> 2.89x | Improved |
| add scale-linkage to layer-linkage-only variant (ablation) | NOVA on Infinity-2B · GenEval / speedup | speedup: 2.34x -> 2.85x | Improved |
| add layer-linkage on top of scale-linkage (ablation) | NOVA on Infinity-2B · GenEval | overall: 0.71 -> 0.72 | Improved |
| add layer-linkage to no-linkage variant (dagger) (ablation) | NOVA on Infinity-2B · GenEval | overall: 0.69 -> 0.70 | Improved |
| dynamic residual cache reuse instead of static caching from 6-th scale (ablation) | NOVA on Infinity-2B · GenEval | overall: 0.67 -> 0.72 | Improved |
| dynamic residual cache reuse instead of static caching from 8-th scale (ablation) | NOVA on Infinity-2B · GenEval | overall: 0.71 -> 0.72 | Improved |
| apply NOVA to Infinity-2B instead of FastVAR | Infinity-2B · GenEval / speedup | speedup: 2.45x -> 2.89x | Improved |
| apply NOVA to Infinity-8B | Infinity-8B · ImageReward | score: 1.030 -> 1.035 | No clear effect |
| apply NOVA to VAR-d36 on ImageNet 512 | VAR-d36 · ImageNet class-conditional | latency: 2.89s -> 1.97s; FID 2.63 -> 2.79 | Mixed |
Closest papers by meaning
- FasterVAR: Plug-and-Play Acceleration for Visual Autoregressive Models (ICML 2026 · similarity 0.91)
- ToProVAR: Efficient Visual Autoregressive Modeling via Tri-Dimensional Entropy-Aware Semantic Analysis and Sparsity Optimization (ICLR 2026 · similarity 0.91)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.88)
- Memory-Efficient Visual Autoregressive Modeling with Scale-Aware KV Cache Compression (NeurIPS 2025 · similarity 0.88)
- TTS-VAR: A Test-Time Scaling Framework for Visual Auto-Regressive Generation (NeurIPS 2025 · similarity 0.87)
- FlexVAR: Flexible Visual Autoregressive Modeling without Residual Prediction (NeurIPS 2025 · similarity 0.87)
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.
