IEEE CCNC 2025 and CES 2025 Participation Report(Muhammad Ayat Hidayat)

Academic Information

Summary of Presentation

(WiP) FedCSAC: Improving Accuracy and Privacy in Fully Decentralized Machine Learning with Clustered Sharding and Adaptive Differential Privacy Clipping.

Author

Muhammad Ayat Hidayat, Yugo Nakamura, Yutaka Arakawa

Content

Federated learning  (FL) is a decentralized approach to machine learning based on data distribution. In FL, the local client only sends gradient, not raw data to the central server for aggregation, this makes FL secure and private. However, Several studies show that FL is not safe enough. Differential Privacy (DP) which injects noise into the gradient to distort its original value, can be implemented to enhance privacy protection. However, DP can decrease model accuracy and can increase the risk of communication bottlenecks at the central server. This study addresses the above issue by removing the central server and bringing aggregation to the edge. Our solution also involves clustering and data sharding, ensuring each process has representative data portions. This improves the signal-to-noise ratio, addresses data imbalance, and enhances accuracy.

Question and Answer

1. What is the process if a new worker is joining the cluster?

There is the configuration file for each worker, in this configuration, there is the IP address for the chief and all workers in the cluster, for adding new workers this file configuration needs to be updated so that the existing worker can know about the newly joined worker for synchronization.

2. How do you measure latency ?

In latency, we measure the time needed for information to travel between devices including the worker and chief node during the training process.

3. How to know a set of groups in the cluster?

The worker can know the group in the cluster by accessing the configuration file on their storage; this configuration file contains a list of all workers and chief nodes in the same group. Experience

It was my first time visiting Las Vegas and the United States. After a long flight, I finally arrived in Las Vegas on January 9th, 2025, at 7 PM local time. The next day, I attended the Consumer Electronics Show (CES) at the Venetian Convention Center and the Las Vegas Convention Center. CES is one of the largest and most influential tech events in the world, showcasing the latest innovations in electronics and information technology. Major tech companies like LG and Panasonic were featured at the event. CES also honors groundbreaking innovations for their unique features and technological advancements (CES Best Innovation Award).

What interests me the most is the use of VR and AI in medical and healthcare applications. For example, fundamental surgery uses VR to simulate the surgical process with incredible detail and precision, while MoveFreeker employs inclusive AI to manage, measure, and make recommendations for essential movements, such as muscle strength and gait, which are vital for achieving a healthier lifestyle. I attended the final day of CES, which was only half a day, but it was still a magnificent and fascinating event. The next day, on January 11th, 2025, I attended the CCNC 2025 conference. The conference focused on communication and networking, with keynote speeches on both days from leading academics and companies. They discussed the implementation of terahertz frequency for near-field communication and space networks, as well as how to measure the quality of wireless communication for the next generations of 5G and 6G networks.

My presentation is on the second day, January 12th, 2025, after the lunch break. It was held during Session 27, which focused on the security and privacy of federated learning. After the presentation, there are three questions from the audience, which mainly focus on evaluation metrics, configuration regarding group, and procedure for newly joined workers in the system. The conference ended on 13th January 2025, and I was coming back to Japan on 14th of January 2025.  

At this conference, I not only gained new insights into the field of federated learning privacy and security, but I also had the opportunity to meet with experts, professors, and employees from major companies in Japan. We discussed our shared research interests and explored potential future collaborations. In addition to attending the conference, I also had the chance to explore Las Vegas, which has a unique atmosphere and delicious local food. This was my first time visiting, and I can't wait to go back in the future.