Research map / Vision-language and multimodal / Human and embodied modeling
Analogous to Evolutionary Algorithm: Designing a Unified Sequence Model
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the EAT local path alongside global attention | DeiT-style image classification models · ImageNet-1k | Top-1 accuracy: 79.8 -> 80.4 | Helped |
| Replace the image and language retrieval networks with EAT | TIRG · CSS | rank-1: 70.1 -> 73.8 | Helped |
| Replace the language network with EAT | PTA · R2R unseen | SPL: 0.28 -> 0.33 | Helped |
| Replace the vision network with EAT | PTA · R2R seen | SPL: 0.61 -> 0.61 | No clear effect |
| Use 0 task-related head layers instead of the default 1 (ablation) | EAT-B · ImageNet-1k | Top-1 accuracy: 80.264 -> 79.692 | Hurt |
| Set task-related head layers to 2, 3, or 4 rather than the default 1 (ablation) | EAT-B · ImageNet-1k | Top-1 accuracy: 80.264 -> 80.422; 80.435; 80.454 | Helped |
| Set task-related head layers to 5 rather than the default 1 (ablation) | EAT-B · ImageNet-1k | Top-1 accuracy: 80.264 -> 80.446 | Helped |
| Set local ratio to 0.25 instead of 0.50 (ablation) | EAT-B · ImageNet-1k | Top-1 accuracy: 80.264 -> 80.280 | No clear effect |
| Set local ratio to 0.75 instead of 0.50 (ablation) | EAT-B · ImageNet-1k | Top-1 accuracy: 80.264 -> 79.518 | Hurt |
| Reduce FFN ratio from 4 to 2 (ablation) | EAT-B · ImageNet-1k | Top-1 accuracy: 80.264 -> 77.422 | Hurt |
| Reduce FFN ratio from 4 to 3 (ablation) | EAT-B · ImageNet-1k | Top-1 accuracy: 80.264 -> 79.176 | Hurt |
| Set convolution kernel size to 1 instead of 3 (ablation) | EAT-B · ImageNet-1k | Top-1 accuracy: 80.264 -> 79.128 | Hurt |
Closest papers by meaning
- Large-Scale Evolution of Image Classifiers (ICML 2017 · similarity 0.86)
- Adaptive-Learngene: Continual Expansion and Task-Aware Selection of Learngenes for Dynamic Environments (AAAI 2026 · similarity 0.86)
- Convolution Goes Higher-Order: A Biologically Inspired Mechanism Empowers Image Classification (NeurIPS 2025 · similarity 0.84)
- EvolveDirector: Approaching Advanced Text-to-Image Generation with Large Vision-Language Models (NeurIPS 2024 · similarity 0.84)
- Auto-Prox: Training-Free Vision Transformer Architecture Search via Automatic Proxy Discovery (AAAI 2024 · similarity 0.84)
- Elastic ViTs from Pretrained Models without Retraining (NeurIPS 2025 · 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.