MS in Machine Learning @ Carnegie Mellon University. Former Intern @ Microsoft Research, Research Affiliate @ MIT, SDE intern @ Uber, MITACS-GRI @ Dalhousie University
Hey! I am Soundarya Krishnan, a master's student in the Machine Learning Department at CMU. I work in the intersection of Machine Learning and Software Development. I was previously working as a research intern at Microsoft Research, where I was advised by Dr. Amit Sharma. Prior to that, I was a research affiliate at the Fluid Interfaces lab at MIT Media Lab, where I was advised by Prof. Pattie Maes.
I did my undergraduation in Physics and Computer Science at BITS Pilani Goa. During the course of my undergraduation, I was a researcher at APPCAIR in collaboration with TCS Research, where I was advised by Prof. Ashwin Srinivasan . Our work was largely in collaboration with Dr. Lovekesh Vig from TCS Research and Innovation Labs , and our research focus was AI for Healthcare.
I’m broadly interested in leveraging the power of Machine Learning to develop deployable solutions for social impact. I'm especially interested in improving public healthcare systems, and building interpretable and explainable systems for the same. Additionally, I'm very passionate about bridging the gender gap in technology, and co-founded the BITS Goa Women in Tech organisation!
Working under the guidance of Dr. Amit Sharma at Microsoft Research on model explainability using counterfactuals. One of the major contributors for the Diverse Counterfactual Explanations (DiCE) for ML library. Added support for non-gradient based models and efficient generation of counterfactuals. New release of the library out soon!
Working under the guidance of Prof. Pattie Maes. at the Fluid Interfaces group at MIT Media Lab on analysing multimodal signals obtained from the sleep mask FASCIA .
Worked in the AdServer team. Used ideas in Reinforcement Learning (Bayesian Thompson Sampling for the Multi Arm Bandit problem) to solve the Explore-Exploit Problem in picking Advertising. Worked with Java, Python, Hive, Apache Spark, SQL, HDFS, Amazon AWS Lambda.
Working under the supervision of Prof. Ashwin Srinivasan and Dr. Lovekesh Vig on projects involving building interpetable and explainable Deep Learning models for healthcare.
Was advised by Prof. Neil Julien Ross at Dalhousie University for a project on optimisation of Quantum circuits. Worked on utilising ideas in Group Theory and Linear Algebra (primarily the Lempel Algorithm) to reduce the number of expensive gates in a quantum circuit.
Advised by Prof. Sundaresan Balasubramaniam at CEERI Chennai . Developed an automated leather defect detection and analysis system using tensorflow in python for object detection and classification. Used heavy Image Processing in OpenCV for feature extraction, and Visual C++ for GUI, including MYSQL 5.7 for the database. Also used a uEye camera compatible with OpenCV API functions for real time classification.
In this work, we built CNN-LSTM models (with explanations) for automatic sleep scoring, along with signal processing to aid sleep researchers in sleep staging. We further integrated the pipeline on a real-time open-source web-based interface using socket programming, Node.js, Express.ejs for the website, D3.js for the visualization, and MongoDB for the database. We also obtained feedback regarding the efficacy of the various explanations and tools that we provide from sleep researchers. Preprint out soon!
In this work, our focus is on constructing models to assist a clinician in the diagnosis of COVID-19 patients in situations where it is easier and cheaper to obtain X-ray data. We use a new COVIDr dataset with important radiological annotations from a practicing radiologist. We build a deep neuro-symbolic model to diagnose COVID-19 and provide visual and textual explanations, with no significant loss in predictive accuracy compared to an end-to-end model. We find that the radiologist prefers simple representations, both visual and textual to aid in diagnosis.
Built a segmentation model in order to isolate the lung region from the rest of the X-ray, and built an interpretable model (with symptoms checking) to detect COVID-19 in chest X-Rays. Our model has an ROC of 99.8. Currently working on generating LIME-style explanation for the COVID-19 predictor in terms of the symptoms, so as to have an explainable model.
Built an NLP & Social Networks-based Python tool that scrapes data off the Telegram chats, classifies users according to activity, and suggests experts as well as timings for various topics, using tools such as Gephi. Tested on Ubuntu IRC chat logs, and has been implemented for a university chat group.
Developed a portable, cost-effective and customisable Electrocardiograph(ECG) Signal analyser for real time sleep apnea detection. We analysed our proposed implementation through a complete run on the MIT-Physionet dataset. Due to the low computational complexity of our proposed method, we find that it is well suited for deployment on embedded devices such as the Raspberry Pi.
Quantum circuits model a sequence of operations to be performed by a quantum computer. CNOT Dihedral is an important class of such quantum circuits. The project demonstrated that the cost of a quantum circuit could be optimized by reducing the # of T gates and replacing them with less expensive gates such as (S, CS or CCZ). This was shown by noting that every CNOT Dihedral circuit can be modeled by a phase polynomial, which in turn can be represented as a tensor. We can then use the Lempel algorithm to factorize this matrix, and thus reduce the T-count (details are given in the report).
BITS Goa Women in Tech (BGWiT) is an organisation that I co-founded along with some of the most brilliant and hardworking women I've met. Here's a quick blog post about why we started! To find out more about us, check out our website!
I had the best experience as a MITACS exchange student pursuing research at Dalhousie University, Halifax under Prof. Neil Julien Ross (one of the best professors that I've had the opportunity of working under)! Check out this blog post to know more about the MITACS application procedure, and my experience there!