
Niao He
Associate Professor of Computer Science · Optimization & Decision Intelligence Group · ETH Zurich
At Prof Rod’s reading desk
Questions to explore
- What do you gain and lose by estimating updates without backpropagation?
- Which privacy assumptions and evaluation results support a fine-tuning method?
Research interests
Niao He studies optimization, reinforcement learning and private fine-tuning.
- Optimization
- Reinforcement learning
- Private fine-tuning
Explore the work
Related opportunities
- Explore a PhD in optimization and learning at ODI
PhD opening · Standing invitation
- Explore postdoctoral research in optimization and reinforcement learning
Postdoc opening · Standing invitation
Prof Rod’s reading notes
Start with a paper
ICML · 2024
DPZero: Private Fine-Tuning of Language Models without Backpropagation
Prof Rod’s reading note: Compare private fine-tuning through forward passes with a gradient-based baseline. Check both memory savings and the stated privacy guarantee.
Bring a question of your own
Turn reading into research
Before reading the results, predict how a forward-pass-only update could change memory use and accuracy. Define a matched baseline and the privacy guarantee you would need to compare.
Keep exploring
More at ETH Zurich
- Afonso Bandeira
Optimization
- Alexandre Refregier
- Ana Klimovic
Machine learning · Computer systems
- Anastasiia Koloskova
Improve this page
Suggest a correction, share a paper to start with, or point readers to a group page or opening.