ICACT20210192 Question.1
Questioner: abdulfuadrahman@gmail.com 2021-02-09 오전 11:28:18 |
ICACT20210192 Answer.1
Answer by Auhor satyabrataaich@gmail.com 2021-02-09 오전 11:28:18
Chrome  Click!! |
| Thank you for the slides. In the conclusion slide, you have mentioned that the application wasn’t developed and tested, yet this solution is indeed promising, so the question is how can you prove that the solution is promising? |
Thank you for the question. So the point I mention in the conclusion is that the framework is developed,but it is not developed in real time and not implemented in the real time. This solution is promising because similar kind of solution is already implemented for other problems by other researchers and it was found that this kind of solution have enough potential for acheieving the desiredobjectives. |
ICACT20210192 Question.2
Questioner: st.lzg@163.com 2021-02-08 오후 1:55:04 |
ICACT20210192 Answer.2
Answer by Auhor satyabrataaich@gmail.com 2021-02-08 오후 1:55:04
Chrome Click!! |
| The training data across the different data centers is unbalanced, did authors have a valid method to FL under this case with your framework? |
Thank you so much for the question. It's a great question. Definitely we are aware the class imbalance problem, Currently we are overcoming the problem based on the state of art model peoposed by Sarkar et al "Fed-Focal Loss for imbalanced data classification in Federated Learning" and Duan et al.,"Self-Balancing Federated Learning With Global Imbalanced Data in Mobile Systems". Those state of art model's are doing well. In the mean time we are developing our own algorithm to overcome this problem. In the future we will update with our new algorithm. Thank you |
ICACT20210192 Question.3
Questioner: fatou.mbodji@esp.sn 2021-02-08 오후 2:04:55 |
ICACT20210192 Answer.3
Answer by Auhor satyabrataaich@gmail.com 2021-02-08 오후 2:04:55
Chrome Click!! |
| Dear authors, thank you for your slides.
Given the impact of the type of blockchain in the accesses. What type of blockchain do you recommend in your proposal? |
Thank you so much for the question. It's a great question.Since the requirement is decentralized relaible and secure FL system, we are recommeding consortium blockchain,which has excellent features to handle privacy.If the author would like to know more about the consoortium blockchain and it's application in other model then have a look at the article written by Qi et al., "Privacy-preserving blockchain-based federated learning for traffic flow prediction". Thank you |
ICACT20210192 Question.4
Questioner: James_Lan@htx.gov.sg 2021-02-08 오후 4:49:25 |
ICACT20210192 Answer.4
Answer by Auhor satyabrataaich@gmail.com 2021-02-08 오후 4:49:25
Chrome Click!! |
| Hi, thanks for the sharing.
Can share your views on how does privacy regulation affects Federated Learning? |
Thank you so much for the question. It's a great question.
Traditional machine learning involves a data pipeline that uses a central server that hosts the trained model in order to make predictions.The downside of this architecture is that all the data collected by local devices and sensors are sent back to the central server for processing, and subsequently returned back to the devices. This round-trip limits a model’s ability to learn in real-time.Federated learning (FL) in contrast, is an approach that downloads the current model and computes an updated model at the device itself (ala edge computing) using local data.In a more general sense, Federated Learning allows for machine learning algorithms to gain experience from a broad range of data sets located at different locations. The approach enables multiple organizations to collaborate on the development of models, but without needing to directly share secure data with each other. We can say that Federated learning helps to develop robust model without revealing private information of the user.
|
ICACT20210192 Question.5
Questioner: joomi@inje.ac.kr 2021-02-09 오후 12:59:02 |
ICACT20210192 Answer.5
Answer by Auhor satyabrataaich@gmail.com 2021-02-09 오후 12:59:02
Chrome Click!! |
| Thank you for your slides. I would like to know how FL is protecting the privacy of healthcare record compared to the traditional method. |
Thank you for the question.Traditional machine learning involves a data pipeline that uses a central server that hosts the trained model in order to make predictions.The downside of this architecture is that all the data collected by local devices and sensors are sent back to the central server for processing, and subsequently returned back to the devices. This round-trip limits a model’s ability to learn in real-time.Federated learning (FL) in contrast, is an approach that downloads the current model and computes an updated model at the device itself (ala edge computing) using local data.In a more general sense, Federated Learning allows for machine learning algorithms to gain experience from a broad range of data sets located at different locations. The approach enables multiple organizations to collaborate on the development of models, but without needing to directly share secure data with each other |
ICACT20210192 Question.6
Questioner: shahram@ieee.org 2021-02-09 오후 2:09:46 |
ICACT20210192 Answer.6
Answer by Auhor satyabrataaich@gmail.com 2021-02-09 오후 2:09:46
Chrome Click!! |
| Thanks for your presentation. Have you implemented or simulated a prototype of your proposed architecture? What are the challenges in building a prototype? |
Thank you for the question.We have been working on the implementation part of the architecture. We have facing challenges for the hadndling of imbalance data sets and few other technical challnges and we hope that we will overcome it soon |
ICACT20210192 Question.7
Questioner: shahram@ieee.org 2021-02-09 오후 2:14:43 |
ICACT20210192 Answer.7
Answer by Auhor satyabrataaich@gmail.com 2021-02-09 오후 2:14:43
Chrome Click!! |
| Can you comment on the scalability of your proposed framework? |
Thank you for the question.In the early stage of our developement we have faced some issues on the scalability and having that in mind we are doing some changes in architecture, we hope we would able to solve the problem. |
ICACT20210192 Question.8
Questioner: wsub@korea.ac.kr 2021-02-11 오후 5:55:44 |
ICACT20210192 Answer.8
Answer by Auhor satyabrataaich@gmail.com 2021-02-11 오후 5:55:44
Chrome Click!! |
| I saw the presentation well. I think it is a good way to protect medical data based on blockchain. But, what i want to ask is about the nature of data that cannot be deleted once stored on the blockchain. Forr example, how will user information stored on the blockchain be handled if the user is no longer using platform? |
Thank you for the question. Even though the user don't use the platform, atleast the transcation of the user's data would be done with the user conscience. So the user will know that how the data's are handling . So misuse of the data can be avoided. |
