Scholar
Reza Bayat
Google Scholar ID: 58et-JMAAAAJ
University of Montreal, Mila
Citations & Impact
Citations
81
H-index
4
i10-index
3
Publications
10
Co-authors
0
Publications
Resume (English only)
Academic Achievements
- Contributed to several cutting-edge AI research projects, including:
- - Proposed 'Relaxed Recursive Transformers', a novel LLM compression method using parameter sharing and layer-wise LoRAs, achieving performance comparable to larger models.
- - Contributed to 'Adaptive Inference-Time Compute', enabling LLMs to predict mid-generation whether restarting improves output without external models.
- - Involved in developing 'Titans', a new architecture with neural long-term memory that scales beyond 2M context windows.
- - Participated in 'Large Concept Model (LCM)' research, operating in semantic concept space (SONAR) for cross-lingual zero-shot generalization.
- - Contributed to 'One-Minute Video Generation with Test-Time Training', using TTT layers to handle long-range dependencies in video generation.
Co-authors