Federated machine learning for self-driving car and minimizing data heterogeneity effect

Name: Pokharel, Prastav

Organization : Tribhuvan University. IOE Pulchowk Campus. Department of Electronics and Computer Engineering

Name: Dawadi, Babu R.

Organization : Tribhuvan University. IOE Pulchowk Campus. Department of Electronics and Computer Engineering

Email : baburd@ioe.edu.np

keyword: Federated learning

LCSH: Automated vehicles

; Data heterogeneity

LCSH: Machine learning

Abstract: This study has implemented federated learning concept to train the car to make it autonomous. Data is collected with the help of simulated car developed by udacity for two different tracks. It records images from center, left, and right cameras with associated steering angle, speed, throttle, and brake. Then using Convolution Neural Network, it is trained to form the modal. After training, the modal is submitted to the server where the models from two different sources that are combined together to generate new modal which is further sent to client for further training. Multiple training have been carried out to analyze the performance of car in autonomous mode. We found that accuracy is not always dependent upon number of iteration. Also, the combined model has always less accuracy than individual model for that specific track from where it is generated. The server initializes the model and global control variate (c) and pass it to the entire client i.e. car for our case. After receiving initial model and control variate, the car will update model and its local control variate (ci). With the help of correction term (c – ci), the server will converge the model in right direction minimizing the effect of data heterogeneity or client drift. Along with implementing federated machine learning, we focus on minimizing the effect of data heterogeneity that arises while training.

King Mongkut's University of Technology North Bangkok. Central Library

Address: BANGKOK

Email: library@kmutnb.ac.th

Created: 2023

Modified: 2023-10-09

Issued: 2023-10-09

บทความ/Article

application/pdf

BibliograpyCitation : In King Mongkut's University of Technology North Bangkok. Faculty of Information Technology and Digital Innovation. The 19th International Conference on Computing and Information Technology (IC2IT 2023) (pp.41-52). Bangkok : King Mongkut's University of Technology North Bangkok

eng

©copyrights King Mongkut's University of Technology North Bangkok

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	IC2IT 2023pp.41-52.pdf	579.71 KB

ใช้เวลา

0.030007 วินาที