2025

• 
  Amazon: Towards Efficient Multimodal Language Models with Dual Model-Token Compression and Enhanced Reasoning Capability
• 
  Meta: Reading Recognition in the Wild
• 
  Columbia University: Empower the Next Billion Devices with AI

2022

• 
  University of Michigan: Empowering the Next Billion Devices with Deep Learning
• 
  Meta: Building Efficient, Scalable, and Inclusive Federated Learning Systems
• 
  Hong Kong University of Science and Technology: From Neural Architecture Search to Data-Centric AutoML
• 
  Rutgers University: From Neural Architecture Search to Data-Centric AutoML

2021

• 
  Stanford University: Empowering the Next Billion Devices with Deep Learning
• 
  University of Southern California: Empowering the Next Billion Devices with Deep Learning
• 
  ICML'21 AutoML Workshop Keynote: Encoding is an Important Design Decision in Neural Architecture Search

2020

• 
  Facebook AI Systems Faculty Summit: Efficient On-Device Distributed Deep Learning via Importance Sampling
• 
  ECCV'20 Tutorial on Automated Deep Learning: Does Unsupervised Architecture Representation Learning Help Neural Architecture Search?