Friday October 29 5:00 PM – Friday October 29 5:30 PM in Talks II

Impact of Noisy Data on Support Vector Machine and Deep Learning Algorithms on Edge Computing Device

Dr. Shrirang Ambaji Kulkarni

Prior knowledge:
Previous knowledge expected
Raspberry Pi, Python, Machine Learning and Deep Learning


Edge Computing augments the capabilities of cloud computing. In the present talk Raspberry Pi is considered as a edge gateway device with capabilities to process data and build predictive models. Support Vector Machine(SVM) a machine learning algorithm with pipeline architecture will be analyzed with respect to deep learning algorithm in the presence of noise and for computational time.


Edge Computing can augment the computational capabilities of cloud computing. It would be possible to build predictive models on the edge for sensitive data. However the underlying dataset may be affected by noise during data collection from the sensors. Noise in data is some unwanted delta random addition to the original data which makes it corrupt or unusable. This has repercussions for machine learning algorithms where by, the presence of noise could lead to trouble in training on the data and may at times lead to overfitting. Also this may reduce the capacity to generalize on the new data. Support Vector Machine (SVM) is a popular machine learning algorithm that is effective in high-dimensional space. The advantages of SVM is its stellar performance and brilliant results. However the downside of it is the training time required and sensitivity to noise. Also the advent of Deep Learning Algorithms has made SVM to take a back-seat. This talk focuses on the performance of Pipelined SVM with Deep-Learning Multi-Layer Perceptron in the presence of white Gaussian noise on a Raspberry Pi. Also the visualization of metrics will provide an interesting insights on the impact of noise on both the algorithms along with computational time used.