Enrich Data Analysis with Deep Learning

Speaker Bio

Professor Michael Chen is working on deep learning modeling and training from an optimizer's perspective. He is particularly interested in solving industrial problems with an interdisciplinary approach of numerical optimization, artificial intelligence, deep learning, and statistical data analysis. Prof. Chen is also an active developer. His Python package Dromedary Studio provides a new modeling language and integrates solvers from these fields. Prof. Chen has been leading multiple research projects funded by NSERC Discovery, Engage, MITACS, OCE and NRC. As an interdisciplinary researcher, Prof. Chen enjoys collaboration with colleagues and students on a wide range of topics in deep learning models and optimization algorithms.

Abstract

Drinking water quality has been a global concern for long. With the advancement of sensor hardware, monitoring devices which are small enough for home use are emerging, and the classification of drinking water containment based on the sensor data is the intelligent kernel. For example, we expect such a device to tell whether the faucet water is polluted by one of CaCO3, Sodium Chloride, Lead, K12 E.Coli, and Fungal. In our study, we analyzed spectrum data and applied several common classification models, including Logistic Regression, SVM and Decision Tree, without satisfactory accuracy. The problem exhibits very high noise-to-signal ratio due to the limitation of the sensor, various environmental factors, and many known and unknown other containments in the drinking water. The data is also highly nonlinear due to the spectrum technology itself. We adapt the Convoluted Neural Network (CNN) deep learning model into our data analysis, and immediately see a significant improvement of the accuracy to 93%. This success, as well as other success as reported by multiple literatures, convinced the speaker that deep learning models should be in the toolbox of every data scientist.

Date

Wednesday, March 18th, 2020 

Time

3:00 p.m.

Location

P327 (Peters Building)