ICACT20210211 Question.1
Questioner: n.brian20202@gmail.com 2021-02-10 오후 12:18:56 |
ICACT20210211 Answer.1
Answer by Auhor skdiakffn@ajou.ac.kr 2021-02-10 오후 12:18:56
Chrome  Click!! |
| About the identifier The thigh shows 100%, but other parts of the body do not achieve 100%. Do you know the reason for this?
Also, is there any noise in the collected data? |
We do not know why the identifier information thigh represents 100%, but other parts of the body do not achieve 100%. There is also noise in the collected data. |
ICACT20210211 Question.2
Questioner: 524078785@qq.com 2021-02-10 오후 12:13:04 |
ICACT20210211 Answer.2
Answer by Auhor skdiakffn@ajou.ac.kr 2021-02-10 오후 12:13:04
Chrome Click!! |
| Were the data about UWB in different parts of the body collected from the same person, or from different people? Do you consider individual contingency and how to avoid the impact of individual differences on data accuracy? |
The data were collected from the same person. Also, to avoid individual contingencies and individual differences, we can use transfer learning or fine-tuning. |
ICACT20210211 Question.3
Questioner: seunghee5@etri.re.kr 2021-02-10 오후 12:14:45 |
ICACT20210211 Answer.3
Answer by Auhor skdiakffn@ajou.ac.kr 2021-02-10 오후 12:14:45
Chrome Click!! |
| Doesn't your suggested method depend on the user's gender, height, or weight? If it depends on the user's physique, is there any way to compensate? |
Our proposed method can vary sufficiently depending on the gender, height, and weight of the user. If the user's physique is different, collecting a little bit of data and going through transfer learning or fine-tuning will quickly learn a model suitable for other users. |
ICACT20210211 Question.4
Questioner: seunghee5@etri.re.kr 2021-02-10 오후 12:21:25 |
ICACT20210211 Answer.4
Answer by Auhor skdiakffn@ajou.ac.kr 2021-02-10 오후 12:21:25
Chrome Click!! |
| The results of this paper show high accuracy. I wonder how many people have been tested with the test results provided in this paper? |
We tested the model learned from the same person with data collected from the same person again. That is, there is one person who has been tested.
As we predict that different people have different physical characteristics, collecting data from different people and collecting models will reduce accuracy.
However, if we collect some data from others and undergo transfer learning or fine-tuning, the model will again show high accuracy. |
ICACT20210211 Question.5
Questioner: ch.lee@etri.re.kr 2021-02-22 오후 4:54:47 |
ICACT20210211 Answer.5
Answer by Auhor skdiakffn@ajou.ac.kr 2021-02-22 오후 4:54:47
Chrome Click!! |
| Thank you for your presentation. Did you conduct experiments with different CNN architectures? I am wondering how you determine the CNN architecture.
|
We experimented with different architectures and performed hyperparameter tuning with the number of hidden layers and the number of nodes in the hidden layer as hyperparameters. |
ICACT20210211 Question.6
Questioner: rusly.eic04g@nctu.edu.tw 2021-02-10 오후 10:03:24 |
ICACT20210211 Answer.6
Answer by Auhor skdiakffn@ajou.ac.kr 2021-02-10 오후 10:03:24
Chrome Click!! |
| Thank you for interesting work. From presentation I didn't quite understand if experiments were conducted, or computer simulations was done? If it is a real experiment, may you share details of your setup? If it is a purely computer simulation, how did you model human body, what tools have been used? |
|
