Scholar
Liam Collins
Google Scholar ID: MRLe02cAAAAJ
Snap
Citations & Impact
Citations
1,572
H-index
10
i10-index
10
Publications
20
Co-authors
12
list available
Publications
Resume (English only)
Academic Achievements
- - Publications:
- - NeurIPS 2024 Spotlight Presentation: In-Context Learning with Transformers: Softmax Attention Adapts to Function Lipschitzness
- - ICML 2024 Oral Presentation: Provable Multi-Task Representation Learning by Two-Layer ReLU Neural Networks
- - FL@FM-NeurIPS’23 Best Paper: Profit: Benchmarking Personalization and Robustness Trade-off in Federated Prompt Tuning
- - COLT 2023: InfoNCE Provably Learns Cluster-Preserving Representations
- - NeurIPS 2022: FedAvg with Fine-Tuning: Local Updates Lead to Representation Learning
- - FL-NeurIPS’22: PerFedSI: A Framework for Personalized Federated Learning with Side Information
- - ICML 2022: MAML and ANIL Provably Learn Representations
- - CoLLAs 2022 Oral Presentation: How does the Task Landscape Affect MAML Performance?
- - Awards:
- - Selected as a top reviewer for NeurIPS 2024
- - Best Paper at FL@FM-NeurIPS’23
Research Experience
- - Current Position: Research Scientist at Snap Research
- - Previous Experiences: Interned at Google Research and Amazon, working on federated prompt tuning and personalized federated learning with side information
Education
- - Ph.D.: University of Texas at Austin, supervised by Aryan Mokhtari and Sanjay Shakkottai
- - B.S.: Princeton University, worked with Yuxin Chen
Background
- - Research Interests: User representation learning, recommendation tasks, sequential and multi-modal interaction data
- - Professional Fields: Machine learning, recommendation systems, federated learning
- - Brief Introduction: Currently working as a Research Scientist at Snap Research on the User Modeling and Personalization (UMaP) team, focusing on learning user representations from sequential and multi-modal interaction data for downstream recommendation tasks.
Miscellany
- Personal interests not mentioned
