|
Data Science Colloquium Series 2:00pm -3:00pm, NKN Hall BAYESIAN STRUCTURAL TIME SERIES FORECASTING Shankar Sivaramakrishnan Aditya Birla Group. 22-11-19 Abstract Time series forecasting is of use in several real-world applications be it demand forecasting, price forecasting, trading etc. In the first half of the talk, we will discuss the landscape of time series forecasting techniques, from the commonly used ARIMA to the latest with deep learning, including some popular time series forecasting packages and model selection techniques. We will also discuss specific industrial use cases where some forecasting methods find better usability over other methods. In the second half of the talk, we will focus on one particular technique – Bayesian Structural Time Series (BSTS) – which is good for forecasting when the number of predictors is larger than the data points, and when the time series has relatively small number of data points. BIO Shankar Sivaramakrishnan is a practice leader in Group Data & Analytics (GDNA) team at Aditya Birla Group. Shankar manages AI and ML-based analytics requirements for textiles, carbon and chemicals businesses of Aditya Birla Group. His team works on customer demand management, market share improvement, raw material price prediction, energy optimization, factory floor analytics, predictive maintenance, quality improvements etc. for ABG group companies. His current technical interests are in probabilistic machine learning, Bayesian analysis and image analytics. Before joining GDNA @ ABG, Shankar was involved in enhancing adoption of industrial analytics, IoT, edge computing, IT/OT security, ML, AI and cloud technologies in Indian heavy industries. Prior to that, Shankar specialized in process-microstructure-property relationships of advanced manufacturing technologies related to high temperature materials used in aircraft engines. Shankar is a contributor to 12 US patents and 10+ international publications. He has a PhD in materials engineering from UIUC and BTech in metallurgy from IIT-Madras.
|
