General Manager and Head of Analytics at Ericsson – Digital Services
While lot of advancement is happening in the field of Machine Learning, real life experiences of implementing AI/ML products in field and at Scale are still limited. The challenges of transferring huge amount of data from field to one central computing infrastructure is prohibitively costly specially when we look at Telecom Networks. Building such costly infrastructure consisting of fat data pipes negates the cost case and expected RoI. Also both in Telco and IoT context, often the data is sensitive and private which imposes additional restriction on data transfer. The first part focusses on such challenges from real-world large scale implementation experiences. The talk then moves to on the emerging solutions for such challenges by employing Federated learning whereby the volume of data transfer is drastically reduced and all of the limited data moving across the network is effectively secured as only computed weights are shared through a secure aggregation mechanism. The talk also shares benefits of such methodology with some indicative comparison of accuracy and benefits The talk is followed by a technical session in the afternoon , where Data Scientists will guide a technical audience to code a FL algorithm on simulated data and compare performances of Centralized vs Federated models
While lot of advancement is happening in the field of Machine Learning, real life experiences of implementing AI/ML products in field and at Scale are still limited. The challenges of transferring huge amount of data from field to one central computing infrastructure is prohibitively costly specially when we look at Telecom Networks. Building such costly infrastructure consisting of fat data pipes negates the cost case and expected RoI. Also both in Telco and IoT context, often the data is sensitive and private which imposes additional restriction on data transfer. The first part focusses on such challenges from real-world large scale implementation experiences. The talk then moves to on the emerging solutions for such challenges by employing Federated learning whereby the volume of data transfer is drastically reduced and all of the limited data moving across the network is effectively secured as only computed weights are shared through a secure aggregation mechanism. The talk also shares benefits of such methodology with some indicative comparison of accuracy and benefits The talk is followed by a technical session in the afternoon , where Data Scientists will guide a technical audience to code a FL algorithm on simulated data and compare performances of Centralized vs Federated models
Implementing Intelligence: Scalability with affordable and secure algorithms By Kaushik Dey big data in healthcare | |
4 Likes | 4 Dislikes |
68 views views | 20.1K followers |
Science & Technology | Upload TimePublished on 11 Oct 2019 |
Không có nhận xét nào:
Đăng nhận xét