Analogous to Evolutionary Algorithm: Designing a Unified Sequence Model
6. prosince 2021
Řečníci
Jiangning Zhang
Sprecher:in · 0 Follower:innen
Wenzhou Chen
Sprecher:in · 0 Follower:innen
Yabiao Wang
Sprecher:in · 0 Follower:innen
Chengjie Wang
Sprecher:in · 0 Follower:innen
Feiyue Huang
Sprecher:in · 0 Follower:innen
O prezentaci
Inspired by biological evolution, we explain the rationality of Vision Transformer by analogy with the proven practical Evolutionary Algorithm (EA) and derive that both of them have consistent mathematical representation. Analogous to the dynamic local population in EA, we improve the existing transformer structure and propose a more efficient EAT model, and design task-related heads to deal with different tasks more flexibly. Moreover, we introduce the spatial-filling curve into the current vision transformer to sequence image data into a uniform sequential format. Thus we can design a unified EAT framework to address multi-modal tasks, separating the network architecture from the data format adaptation. Our approach achieves state-of-the-art results on the ImageNet classification task compared with recent vision transformer works while having smaller parameters and greater throughput. We further conduct multi-model tasks to demonstrate the superiority of the unified EAT, e.g., Text-Based Image Retrieval, and our approach improves the rank-1 by +3.7 points over the baseline on the CSS dataset.Inspired by biological evolution, we explain the rationality of Vision Transformer by analogy with the proven practical Evolutionary Algorithm (EA) and derive that both of them have consistent mathematical representation. Analogous to the dynamic local population in EA, we improve the existing transformer structure and propose a more efficient EAT model, and design task-related heads to deal with different tasks more flexibly. Moreover, we introduce the spatial-filling curve into the current vis…
Organizátor
NeurIPS 2021
Konto · 1,9k Follower:innen
O organizátorovi (NeurIPS 2021)
Neural Information Processing Systems (NeurIPS) is a multi-track machine learning and computational neuroscience conference that includes invited talks, demonstrations, symposia and oral and poster presentations of refereed papers. Following the conference, there are workshops which provide a less formal setting.
Baví vás formát? Nechte SlidesLive zachytit svou akci!
Profesionální natáčení a streamování po celém světě.
Sdílení
Doporučená videa
Prezentace na podobné téma, kategorii nebo přednášejícího
Intrinsic Control of Variational Beliefs in Dynamic Partially-Observed Visual Environments
Später ansehen
Ewigspeicher-Fortschrittswert: 0 = 0.0%
Symmetry Discovery with Deep Learning
Später ansehen
Krish Desai, …
Ewigspeicher-Fortschrittswert: 0 = 0.0%
Fast Pure Exploration via Frank-Wolfe
Später ansehen
Po-An Wang, …
Ewigspeicher-Fortschrittswert: 0 = 0.0%
Towards Sample-Efficient Overparameterized Meta-learning
Später ansehen
Yue Sun, …
Ewigspeicher-Fortschrittswert: 0 = 0.0%
Counterfactual Invariance to Spurious Correlations
Später ansehen
Ewigspeicher-Fortschrittswert: 0 = 0.0%
Automated Dynamic Mechanism Design
Später ansehen
Hanrui Zhang, …
Ewigspeicher-Fortschrittswert: 0 = 0.0%