Conference Video: Meghamala Sinha - Causal Inference from Experiments and Observations
This is the Meghamala Sinha’s Tech Talk from the WiDS Puget Sound Conference 2020. Enjoy!
ABSTRACT:
Causal Inference is an important paradigm for data analysis in the fields of medical science, economics, engineering, humanities etc due to its utility in action planning, diagnosis, predictive applications. To increase statistical power for learning a causal network, data are often pooled from multiple observational and interventional experiments. However, if the direct effects of interventions are uncertain, multi-experiment data pooling can result in false causal discoveries, losing the very purpose of its application. For example, in medical science, a false positive result giving an erroneous indication that a particular disease is present (when it isn’t) can result in unnecessary medical tests and panic. To resolve this issue, I will discuss a novel data integration method, “Learn and Vote” to combine information from multiple interventional experiments with observations to learn more accurate causal networks which reduces the detection of false positives.
Meghamala Sinha is a PhD candidate at Oregon State University. She is majoring in Computer Science and minoring in Biological Data Science. Her research interest is Causal Inference and its application to data-driven areas like Machine Learning, AI, Intelligent Systems and Computational Biology. Her work centers around using fundamentals of Causality to differentiate true cause-effect relationships from mere associations in data and building a more robust and reliable inference model.