Vishal Doshi, Ying-Zong Huang, Vasudha Shivamoggi, (MIT Alums), Ritesh Madan, Balaji Rengarajan, (Celect, Inc.), Devavrat Shah, Professor of EECS
| Feb/01 | Wed | 09:00AM-05:00PM | 32-124 |
Enrollment: Limited: Advance sign-up required
Prereq: Interest in topic and readiness for firehose
The traditional approach to building prediction systems has been tailored
to a specific goal. For example, the recommendation system for a media portal
like Netflix or YouTube is built primarily to serve a singular goal of
finding media that an individual may find engaging. This approach works
great when one has time and / or resources to invest in building such a
system and prediction goal remains unchanged.
In many scenarios, one wishes to build a scalable prediction system with
the following desiderata: (a) system gets up and running without team of
data scientists (just like using a database), (b) system can incorporate
new data sources seamlessly, (c) supports generic prediction goals, (d)
scalable and robust, and (e) ability to consume unstructured data (a la
text, image).
At Celect, Inc. (founded out of MIT), such a system has been built.
Specifically, it is a software layer that changes a given scalable storage
infrastructure into scalable prediction infrastructure. In this one day
course, we will discuss: (a) system abstraction and interfaces, (b)
detailed demonstration, and (c) various case-studies with relevant
datasets.
Please feel free to bring your dataset and prediction problem to give it a
try!
Space is Limited, Please Register <https://goo.gl/mDXluo>
Sponsor(s): Electrical Engineering and Computer Science
Contact: Devavrat Shah, devavrat@mit.edu