Fundamental and Applied Research in Machine Learning.

• Applying data science techniques to digital marketing to improve customer engagement and retention.
• Led a project to personalize email marketing campaigns for a $300 billion net worth company.
• Developed and shipped the algorithm to generate Personalization Insights Reports in Adobe Target.
• Led research projects in the fields of Reinforcement Learning and Explainable AI having applications in digital marketing.
• Delivered innovation talks and lab sessions at Adobe Tech Summit 2019.
• Published papers, filed patents, interviewed and mentored interns, and supervised new-hires.

Information-theoretic Evolution of Model Agnostic Global Explanations.

Sukriti Verma, Nikaash Puri, Haasith Guduru, Animesh Agrawal, Piyush Gupta, Balaji Krishnamurthy.
Under Review at the 2021 ACM Conference on Fairness, Accountability and Transparency. ACM FAccT 2021.

• A model agnostic approach to interpret a black box model into human understandable if-then rules.
• Outperforms existing state of the art on 9 out of 10 publicly available data sets.
• Deployed in Adobe Target as a feature to generate Insights Reports. (Adobe Blog)
• Patent Filed in US, UK, Germany, Australia, China. US15/812,991.

MixBoost: Synthetic Oversampling with Boosted Mixup for Handling Extreme Imbalance.

Anubha Kabra, Ayush Chopra, Nikaash Puri, Pinkesh Badjatiya, Sukriti Verma, Piyush Gupta, Balaji Krishnamurthy.
Accepted at the 20th IEEE International Conference on Data Mining. ICDM 2020. (Shortpaper: Oral, pdf)

• A novel approach to solve the problem of class imbalance.
• Outperforms existing state of the art on 17 out of 20 publicly available data sets.
• Patent application is in filing process. (P9215-US).

Explain Your Move: Understanding Agent Actions Using Specific and Relevant Feature Attribution.

Nikaash Puri, Sukriti Verma, Piyush Gupta, Sameer Singh, Dhruv Kayastha, Shripad Deshmukh, Balaji Krishnamurthy.
Accepted at International Conference on Learning Representations. ICLR 2020. (Poster: Oral, Video, Website, pdf, code)

• A model-agnostic approach to understand the behaviour of black-box deep RL agents using saliency maps.
• Outperforms existing state of the art on RL agents trained to play Atari, Chess and Go.
• Worked with experts to create the publicly available Chess Saliency Dataset.
• Generates saliency maps that are 22% more accurate than existing state of the art for Chess agents.
• Generates saliency maps that help humans solve Chess puzzles with a 77% increase in accuracy while taking lesser time than existing state of the art.

A Comparative Study of Evolutionary Methods for Feature Selection in Sentiment Analysis.

Shikhar Garg, Sukriti Verma
Accepted at the 11th International Joint Conference on Computational Intelligence - Volume 1: ECTA. IJCCI 2019. (Oral, pdf, code)

• Analysed the performance of Genetic, Binary Bat and Binary Grey Wolf Algorithm for feature selection.
• The analysis was done across 3 different classification algorithms (SVM, k-NN and Random Forest) and 2 publicly available data sets.
• Concluded Genetic Algorithm to be the most efficient approach in terms of feature reduction.
• Concluded Binary Bat to be the easiest to tune.

MAGIX: Model Agnostic Globally Interpretable Explanations

Piyush Gupta, Nikaash Puri, Sukriti Verma, Pratiksha Agarwal, Balaji Krishnamurthy.
Accepted at ICLR 2019 Workshop on Debugging Machine Learning Models. ICLR Workshop 2019. (Poster, pdf, Adobe Blog)

• A model agnostic approach to explain black box classification models

A Survey of Deep Learning Techniques in Speech Recognition.

Akshi Kumar, Sukriti Verma, Himanshu Mangla.
Accepted at the IEEE International Conference on Advances in Computing, Communication Control and Networking. ICACCCN 2018. (Oral, pdf)

• Survey of the application of Deep Belief Networks, Convolutional Neural Networks and Recurrent Neural Networks for speech recognition.

Extractive Summarization using Deep Learning.

Sukriti Verma, Vagisha Nidhi.
Accepted at the 18th International Conference on Computational Linguistics and Intelligent Text Processing. CICLING 2017. (Poster, pdf, code)

• A novel approach to summarize factual reports written in English.

”Rule Determination for Black-Box Machine-Learning Models”.

Piyush Gupta, Sukriti Verma, Pratiksha Agarwal, Nikaash Puri, Balaji Krishnamurthy.

• Filed in US, UK, Germany, Australia, and China. US15/812,991.

”Personalized Dynamic Content via Content Tagging and Transfer Learning”.

Dheeraj Bansal, Sukriti Verma, Pratiksha Agarwal, Piyush Gupta, Nikaash Puri, Vishal Wani, Balaji Krishnamurthy.

• Filed with the US Patent Office. US16/377,424.

”Minimizing opportunity cost of online exploration in customer journey management.”

Sukriti Verma, Jayakumar Subramanian, Shripad Deshmukh, Piyush Gupta, Nikaash Puri.

• In Filing Process with law firm Chau & Associates LLC. Internal No. P10405-US

Delhi Technological University, Delhi, India

(Formerly Delhi College of Engineering)

B.Tech in Computer Science and Engineering
2014 - 2018