The talk would introduce the audience to the art of doing causality research in experimental as well as real-world contexts. We would discuss the science behind real-world randomization examples, such as A/B tests, focusing on how such tests form the holy grail of causal deductions. We would also talk extensively about contexts where randomization is impossible or impractical, and the various statistical techniques (e.g. quasi-experimental methods) that can be used to make causal inferences in such situations.
This talk is meant for professionals working in the data science industry and research enthusiasts looking to address causality related questions on using observational data. Also ideal for social scientists (e.g. sociologists, political scientists, economists, etc.) who wish to leverage large real-world datasets to answer questions about human and social behaviour.
Note: No prerequisites. Although, a basic understanding of machine learning and statistical/probability theory would be desirable.
A 10-minute introductory session on what is randomization, and why randomization forms the holy grail of causality related research both inside and outside the laboratory. Some real world examples of successful web-based companies who extensively employ randomized tests would also be discussed.
A 30-minute session where we would discuss cases where randomization is not feasible, and the possible statistical methods at our disposal to investigate causality in such contexts. This session would focus mainly on 4 quasi-experimental techniques – (i) Natural (quasi-)experiments, (ii) Instrumental Variables, (iii) Propensity Score Matching, and (iv) Regression Discontinuity models. These techniques work particularly well (although without guarantees) when large amounts of contextual data is available, as is the case with current big data contexts.
A 20-minute session coupled with Q&A that walks through the key similarities and differences in the way machine learning and causality practitioners go about their job. What can machine learning teach causality researchers? Conversely, what concepts from this workshop can be adapted to machine learning applications?
Yes, unless the talk is a live-only event (stated in event description) we will be recording the talk and will post the video recording online afterwards. We'll send out an email to all registrants 1-2 business days after the talk with a link to the video.
"IST" stands for Indian Standard Time. We are located in India, which is Indian Standard Time (UTC+05:30).
You need to sign in and go to your dashboard. You can log in to the TE talk from the "Upcoming" Section under the TE Talk tab.
The web participants window lets you see all the users logged into the current session. Any user can raise their hand to get the attention of the presenter. You can also use the chat for the same.
No - we'll take care of that for you. As an attendee, you are automatically muted.
We value your feedbacks, if you have any recommendations for improving our services, notice any bugs or any suggestions, feel free to share it with us by sending an email to firstname.lastname@example.org