Guest Speaker 1: Arun Tripathi, Ph.D.
Time: 10/7 2:30 – 3:30
Company: IBM Watson Solutions Services
Title: Managing Consultant & Lead Data Scientist
Presentation Title: Machine Learning & AI: The Evolution and Current Landscape
Presentation Overview:
What is Machine Learning
A Brief history
Current Landscape
AI & Deep Learning
Some business applications
Future trends
Personal thoughts on navigating through the universe of machine learning
Materials:
Additional questions students sent in after the presentation:
1. generally speaking, are there that many companies have enough data to train deep learning models? I think of this question because I read a LinkedIn article by a data scientist recently, who said that deep learning has not been commonly used in companies due to the limited data volume. He argued that statistical analysis is still the mainstream.
Answer: I have a different point of view here. I think large volumes of data has been available for many many years already. And more and more, most companies have more data than what they know to do with. A lot of insights are contained in unstructured data, which many companies are just beginning to leverage. I think a big reason for pickup in AI and deep learning has been theoretical advances, and also increase in computational efficiency through the use of GPUs etc. And finally, I think humans are in general resistant to change – so it takes a while for a new technique/methodology to gain acceptance in the mainstream.
And finally, I think the argument of Deep Learning vs Traditional Statistics maybe misplaced. I don’t think either/or. There will always be situations where traditional statistics is more appropriate and Deep Learning maybe an overkill. And similarly, there will be situations where deep learning is the way to go – e.g. image recognition, language translation, self-driving cars etc. So it is best to have a good idea of both approaches, and learn understand which technique is appropriate for a given situation.
2. As Statistics has so many branches, I was wondering what techniques or theories (Stochastic Process, Time-Series Analysis, Non-parametric Statistics, Bayesian…) are most beneficial to machine learning/deep learning/ai?
Answer: I would say learn as much as you can, and depending on the industry one goes into, certain classes of techniques are applied more than others. But I would suggest, at least, to learn a range of techniques such as multi-variate regression, logistic regression, generalized linear models, decision trees, neural nets, time series models, and Bayesian statistics. And then over time pick up more as you grow in career. Learning, I think, is almost a life long process.