Defa Zhu
Scholar

Defa Zhu

Google Scholar ID: v4ySl6MAAAAJ
ByteDance
AGI
Homepage↗Google Scholar↗
Citations & Impact
All-time
Citations
192
 
H-index
5
 
i10-index
4
 
Publications
9
 
Co-authors
0
 
Contact
Emailzhudefa_ai@163.comTwitterOpen ↗
Publications
9 items
Virtual Width Networks
2025
Cited
0
SeeDNorm: Self-Rescaled Dynamic Normalization
2025
Cited
0
UltraMemV2: Memory Networks Scaling to 120B Parameters with Superior Long-Context Learning
2025
Cited
0
Seed-Thinking-v1.5: Advancing Superb Reasoning Models with Reinforcement Learning
2025
Cited
1
Expert Race: A Flexible Routing Strategy for Scaling Diffusion Transformer with Mixture of Experts
2025
Cited
0
Frac-Connections: Fractional Extension of Hyper-Connections
2025
Cited
0
Over-Tokenized Transformer: Vocabulary is Generally Worth Scaling
2025
Cited
0
Ultra-Sparse Memory Network
arXiv.org · 2024
Cited
0
Resume
Academic Achievements
  • - 'Expert Race: A Flexible Routing Strategy for Scaling Diffusion Transformer with Mixture of Experts', ICML 2025
  • - 'Frac-Connections: Fractional Extension of Hyper-Connections', Tech Report
  • - 'Over-tokenized transformer: Vocabulary is generally worth scaling', ICML 2025
  • - 'Ultra-Sparse Memory Network', ICLR, 2025
Research Experience
  • - Researcher at ByteDance, focusing on Large Language Models (LLMs)
Education
  • - Master's Degree, Chinese Academy of Sciences, 2020, focused on Generative AI research
  • - Bachelor's Degree, Northeastern University, 2017
Background
  • Research Interests: Developing stronger architecture of Large Language Models (LLMs). Professional Field: AI Research.