Once a concept, AI is now ushering in a key stage of application. What’s the solution to the data silos among businesses? Given the enhanced regulation on data at home and abroad, what’s the solution to data privacy and security concerns? What’s the status quo of Federated Machine Learning and how to establish an ecosystem for FML in the future?

WeBank, IBM and other organizations jointly held the 1st International Workshop on Federated Machine Learning for User Privacy and Data Confidentiality (FML’19) in conjunction with the 28th International Joint Conference on Artificial Intelligence (IJCAI-19) on Aug. 12, 2019, to further discussion on these issues.

President of IJCAI, Chair of FML Steering Committee, Chief AI Officer of WeBank Professor Qiang Yang delivered opening remarks at the workshop. Dr. Shahrokh Daijavad from IBM and Dr. Jakub Konečný from Google presented keynote addresses. In the panel discussion, top scholars from WeBank, Bar-Ilan University, IBM, Squirrel AI, Google, Huawei, Clustar, Sinovation Ventures and many other renowned enterprises and universities shared and discussed their findings and experience in FML as an emerging AI technology.

This workshop received 40 papers, of which 12 were presented during the workshop, 19 presented via poster. Awards include Best Theory Paper Award, Best Application Paper Award, Best Student Paper Award, Best Presentation Award. Selected high quality papers will be invited for publication in a special issue in the IEEE Intelligent Systems journal. All these attracted numerous scholars to engage in discussions and join efforts for building the FML ecosystem.

Advertisement

Experts from IBM and Google Share Groundbreaking Findings with a Focus on the Theory and Application of FML

Privacy and security are becoming a key concern in our digital age. On 25th May last year, the implementation of General Data Protection Regulation (GDPR) by the EU, the toughest Act on data privacy protection, stressed that user data collection must be open and transparent. A series of laws and regulations from China and overseas also pose new challenges to the traditional way of handling data and model for cooperation. Seeking ways for AI to adapt to this new reality became top priority, a demand that led to this workshop on FML.

A wealth of solutions and breakthroughs were shared by Dr. Shahrokh Daijavad from IBM and Dr. Jakub Konečný from Google in their speech on FML.

Besides how FML can help tackle challenges in the business world, Dr. Shahrokh Daijavad also shared the concept of Fusion AI, which means to train models on widely distributed data sets, but fuse them to produce one equivalent to what centralized training would yield. “Unlike traditional machine learning, in Fusion AI, model parameters are shared and data is not transferred, which makes Fusion AI model better than models that moving data centrally.” Given the widely distributed data, the development of Fusion AI and FML became ever important and imminent.

“FML enables machine learning engineers and data scientists to work productively with decentralized data with privacy by default,” said Dr. Jakub Konečný from Google. He also shared with us how FML works and its use cases at Google. In the case of Gboard, as on-device data is privacy sensitive or large or is more relevant than server-side proxy data, and labels can be inferred naturally via user interaction, the application of Federated RNN compared to prior n-gram model can increase the accuracy of next-word prediction by 24%, and the click rate of prediction strip by 10%.

Advertisement

Major Figure Panelists Discuss the Way Ahead for FML

The moderator of the panel discussion, AI Principal Scientist of WeBank Dr. Lixin Fan joined panelists including Professor Benny Pinkas from Bar-Ilan University, Dr. Shahrokh Daijavad of IBM Academy of Technology, Chief Architect of Squirrel AI Dr. Richard Tong, Research Scientist of Google Dr. Jakub Konečný, Dr. Baofeng Zhang from CTO Office of CBG Software in Huawei, Executive VP of Clustar Dr. Junxue Zhang, VP of AI Institute in Sinovation Ventures Dr. Ji Feng and other experts in a host of in-depth exchanges with attendees, to shed light on the way ahead for FML.

Experts shared thoughts in the panel discussion on questions including but not limited to: How to meet the security and compliance requirements? Is there a way to extend the value of data while observing user privacy and data security? Given the classic trade-off between data regulation and development of AI, how to achieve the long-term goal of establishing a stable and win-win business ecosystem?

List of Award-Winners

Best Theory Paper Award, Best Application Paper Award, Best Student Paper Award and Best Presentation Award selected by all attendees were announced at the closing of the workshop.

Advertisement

Best Theory Paper Award: 
Preserving User Privacy for Machine Learning: Local Differential Privacy or Federated Machine Learning? By Huadi Zheng, Haibo Hu & Ziyang Han;

Best Application Paper Award: 
FedHealth: A Federated Transfer Learning Framework for Wearable Healthcare. By Yiqiang Chen, Jindong Wang, Chaohui Yu, Wen Gao & Xin Qin;

Best Student Paper Award: 
Quantifying the Performance of Federated Transfer Learning. By Qinghe Jing, Weiyan Wang, Junxue Zhang, Han Tian & Kai Chen;

Best Presentation Award: 
Federated Generative Privacy. By Aleksei Triastcyn and Boi Faltings.

President of IJCAI, Chief AI Officer of WeBank Professor Qiang Yang, Chief Architect of Squirrel AI Dr. Richard Tong, andVP of AI Institute in Sinovation Ventures Dr. Ji Feng presented the awards.

Advertisement

“The mission of this International Federated Machine Learning Workshop is to facilitate further understanding in the academia, business community as well as legal and regulatory institutions by promoting the establishment of FML ecosystem in the hope that more businesses will join and build a platform for students aspired to work in FML to find research teams that suit them,” said Professor Qiang Yang.

Held Aug. 10-16, 2019 in Macao, China, IJCAI-19 is one of the leading International Academic Conference on AI, attracting over 3000 AI research personnel and experts. The 1st International Workshop on Federated Machine Learning (FML’19) was a highlight for experts joining this event. Visionaries in the academia and industrial sector expressed the willingness to be part of the effort for academic research, application of FML in the future, and the development and boom of AI ecosystem.