2025-02-11

Agenda

• Federated learning demo, 30-40 minutes, @Moshe Shadmon

• Physical AI, 10 minutes, @yong

• TSC members adjustment. (New & Change)

Participants

@Wilson Wang @Chun Liu @Gary Qi @Yike Gong @Mengyu Hang @Ning Jiang @Jianwen Pi @Jiayi Song @Joseph Sun @Lance Lim @Qingkun Li @Mo Li @Noe Otero @Tina Tsou @Wenhui Zhang @Hao Xie @Xun Chen @Ye Xu @Ying Wang @Yizheng Jiao @Zixiong Zhang @Akram Sheriff @anfernee.guan @Ashwanth Gnanavelu @Yona @Yongli Chen @HaruhisaFukano @Jenny Yang @李博睿 @刘秉伟 @LU CHENG @Rachel Roumeliotis @Roy Shadmon @Ruoyu Ying @Sam Han @Shunli @Suhaas Teja Vijjagiri @Sumeet Solanki @Tina Tsou @TinaTsou @Tom Qin @Victor Lu @万里 @Wilson Wang @Yin Li @Yu Wang - AIG

Recording

https://bytedance.us.larkoffice.com/minutes/obusof6m4d65432ihk183489

Summary The meeting discussed various topics related to the workstream, the main contents included: - Federated Learning Demo: Presented an automated federated learning process using Edge Lake, showing data sharing and model aggregation across multiple nodes with real-world data and evaluations. - Physical AI by Yong: Yong was absent, so this item was not covered. - TSC Member Adjustment: Discussed nominating career for membership and changing training Chris to Kevin due to Chris leaving. It was decided to send an email for further discussion. - Health and Flu Cases: Mentioned flu cases in the US and Wilson's recovery experience. - Remote Work and Locations: Discussed the locations of participants, such as Wilson in Zhengzhou and Tom in Beijing with plans to go to the US in March. - Other Topics: Included discussions about hardware used in federated learning and potential applications like code review. Chapters 00:22 A Conversation about Locations, Health and Meeting Agenda This section begins with Wilson Wang and Tom Qin sharing their locations, with Wilson in Zhengzhou and Tom in Beijing. They discuss future travel plans. Then, the group chats about the weather in different places. After that, they talk about people's health issues. Next, Tina Tsou mentions the ski week at her kids' school and asks about the agenda. The agenda includes a federated learning demo, physical AI presentation, and TSC member's adjustment. Finally, the presenter is ready to start sharing the screen. 08:02 Edge Lake's Breakthrough in Federated Learning: Architecture, Demo and Use Cases This section is presented by Roy Shadmon. He first introduced Edge Lake and federated learning, highlighting its benefits like decentralization and data privacy. He then discussed challenges to its adoption, such as heterogeneous data and operational complexity. He proposed Edge Lake to address these gaps. Next, he detailed a federated learning architecture and outlined a temperature prediction demo's setup, stating the process is mostly automated, and invited questions before starting the demo. 25:12 Discussion about Hardware for AI and Federated Learning Platform This section involves Jeff asking questions. He first clarifies the SLM abbreviation. Then he inquires about considering the inexpensive hardware Deep Seek uses, like older A100 boards. The response is the platform is hardware-agnostic. Jeff shares a customer story where code review led to requests for setting up large language models on Deep Seek to cut costs. He hoped the team was exploring using cheaper hardware like A100 boards for small models. The team says they haven't tried but it could work theoretically. 31:08 Demo and Introduction of Edge Lake's Federated Learning Platform This section is a demo of a federated learning process using Edge Lake. The speaker shows how to publish a training application, start training, and nodes automatically train, share sub - models, and aggregate them. They also demonstrate model evaluation during training and direct inference on input data. The speaker then summarizes Edge Lake's advantages, like handling data heterogeneity, ensuring availability, protecting data ownership, and reducing operational complexity, making it a largely automated platform. 43:31 Meeting Discussions on Federated Learning, Agenda Items and Membership Changes This section first discussed the potential of a project fitting into other models and projects, especially regarding federated learning. It highlighted the advantage of local data storage and proposed federated code review. Then, features like custom aggregations were introduced. After that, it moved on to agenda items, including a no-show for "physical AI." Finally, Wilson Wang and Tom Qin discussed TSC member adjustments, deciding to send an email for further discussion.