cv
Basics
| Name | Xuan Kan |
| Label | Research Scientist |
| kanxuan1996@gmail.com | |
| Url | https://kanxuan.live |
| Summary | Research Scientist at Meta Monetization GenAI working on reinforcement learning, language model evaluation and reward modeling, and multimodal LLMs. |
Work
-
2024.05 - Present Menlo Park, CA, USA
Research Scientist
Monetization GenAI, Meta
Ads Creative Agent
- Designed and implemented an end-to-end agentic RL framework for ads image generation, combining policy optimization with multimodal reward signals and extending the framework toward video ads generation agents.
- Led the design of a multimodal automated evaluation system from prompt design through SFT and RL to scalably assess LLM/VLM-generated ad creatives, with findings published in peer-reviewed venues.
- Built a user preference prediction model for automatic creative variant selection, achieving a 40%+ save-rate lift in A/B tests.
-
2023.06 - 2023.09 Seattle, WA, USA
PhD Software Engineer Intern
Ads Core ML, Meta
Designed and built new ads models for the Signal Loss problem.
-
2022.10 - 2023.06 Atlanta, GA, USA
Student Researcher
Federated Assistant Team, Google
Employed federated learning and parameter-efficient transfer learning to reduce ASR data requirements and communication costs between servers and clients.
-
2022.05 - 2022.08 Seattle, WA, USA
PhD Software Engineer Intern
Ads Core ML, Meta
Built Multi-Task Multi-Label Hypergraph Neural Network models for the Signal Loss problem.
-
2019.02 - 2019.07 Beijing, China
Research Intern
Smart City Group, SenseTime Research
Accelerated mobile neural network inference for stereo matching with neural architecture search.