Unraveling the 'Why' in AI: Pranita Patil's Insights in Causal Deep Learning

Large amounts of data contain both unknown and unseen swaths of information. With data growth increasing daily and the possibility of innovations in data science becoming limitless, it is essential to understand how data and the information it holds can be harnessed with advanced technologies. And this is what Senior Data Scientist Pranita Patil is going to speak about today.

Interviewer: Pranita, your journey in data science and deep learning is truly inspiring. Can you start by telling us how it all began?

Pranita Patil: Absolutely. My fascination with machine and deep learning started during my undergraduate studies in Electronics Engineering at VIIT Pune, India, in the late 2000s. I graduated in 2010 and then moved to the United States for further education. I received a Master of Science in Electrical and Computer Engineering in July 2013 from Oklahoma State University, another Master of Science in Data Analytics in 2017, and PhD in data science in 2022 from Harrisburg University of Science and Technology with a 4.0 GPA.

Interviewer: That's an impressive academic background. Can you tell us more about your research work in causal deep learning?

Pranita Patil: Thank you. My research in causal deep learning primarily focused on addressing the challenges of accurately learning and inferring dynamic causal interactions between genes in complex system like gene regulatory network (GRN). Traditional models often struggle with this, leading to unreliable and inaccurate predictions. 

I introduced a novel technique to integrate Granger causality in deep learning, particularly graph neural networks, to identify causal relationships that lead to more accurate and interpretable predictions in areas like biology, neuroscience, and finance. I also developed decorrelated deep learning methods, which not only helped to mitigate bias effectively but also helped to reduce redundancy in the gene expression data by focusing on meaningful relationships rather than correlated or spurious. These innovative approaches allowed us to create more robust and interpretable models.

Interviewer: How did your personal experiences influence your professional path, particularly your interest in causal deep learning?

Pranita Patil: My grandmother's passing from a brain hemorrhage had a profound impact on me. She had health issues, but no one detected that it was related to a brain hemorrhage. I wondered if analyzing her test data more closely with advanced technology and finding causation in the data could have led to early detection and added more value from a causal deep-learning perspective. This personal experience deepened my commitment to exploring how artificial neural networks can detect unknowns in data, automating the process with high accuracy, and researching how to integrate causality and deep learning. 

Interviewer: Your work has had significant impacts, particularly in the fields of deep learning and medicine. Can you elaborate on some of your contributions?

Pranita Patil: In 2012, I developed a toolbox used by the well-known startup medtech company I was working for to detect heart murmurs and coronary heart disease. I also introduced a methodology for learning complex non-linear Granger causality interactions in temporal data. My work on Parkinson's disease (PD) and to create fairer AI systems has been particularly impactful. I published three papers in a well-regarded scientific journal on deep learning models that utilize rs-fMRI data to understand better and diagnose PD and mitigate any bias problems, and also that combine with the Granger Causality framework for learning latent causal relationships in gene regulatory networks.

Interviewer: What are the potential applications and impacts of causal deep learning across various industries?

Pranita Patil: Causal deep learning has the potential to revolutionize many industries. In healthcare, it can lead to more accurate diagnostics and personalized treatments. In finance, it can improve risk assessment and fraud detection by understanding the underlying causes of financial anomalies. In environmental science, it can help in understanding and mitigating the impacts of climate change by identifying causal factors. The ability to understand and act on true cause-and-effect relationships makes AI systems much more robust and beneficial.

Interviewer: AI and deep learning have seen incredible advancements, but there are still limitations. Can you explain the current challenges these models face, especially regarding correlation and causation?

Pranita Patil: Absolutely. Traditional AI and deep learning models primarily rely on correlation, which means they can identify patterns and relationships in data. However, they often fail to understand the true cause-and-effect relationships, which can lead to biased or inaccurate outcomes. For instance, in medical diagnostics, a model might correlate certain symptoms with a disease without understanding the underlying cause, leading to potential misdiagnoses.

Interviewer: What are your future goals and aspirations in AI and data science, particularly in causal deep learning?

Pranita Patil: My future goal is to leverage my AI knowledge to help communities across various fields. I want to ensure society can utilize AI models to advance technology, healthcare, and environmental management significantly. Understanding the underlying logic of AI is crucial, and I aim to further research and address the drawbacks and advancements in this area. 

Interviewer: Thank you, Pranita, for sharing your insights and experiences. Your work in causal deep learning is making a significant difference.

Pranita Patil: Thank you for having me. It's been a pleasure sharing my journey and aspirations.

In assessing her past experiences and achievements in AI, Pranita Paitl sees herself as someone leading innovative technologies that positively impact the quality of life and reduce the environmental impact.