Get my full resume here.

Publications

  • Lilly Kumari, Sunny Dhamnani, Akshat Bhatnagar, Atanu R. Sinha, and Ritwik Sinha. 2016. Audience Prism: Segmentation and Early Classification of Visitors Based on Reading Interests . In Proceedings of the 3rd IKDD Conference on Data Science, 2016 (CODS '16). ACM, Pune, India [report]

Patents

  • Classification of Website Sessions using One-Class Labeling Techniques - Sunny Dhamnani, Lilly Kumari, Ritwik Sinha, Vishwa Vinay (Filed in US PTO office, Application - 15/793,001)


  • Detecting robotic internet activity across domains utilizing one-class and domain adaptation machine learning models - Sunny Dhamnani, Vishwa Vinay, Lilly Kumari, Ritwik Sinha, Margarita Savova, David Weinstein (Filed in US PTO office, Application - 15/982,393)


  • Makeup identification using deep learning - Niyati Chhaya, Lilly Kumari, Nitin Rathor, Vineet Vinayak, Rutuj Jugade (Filed in US PTO office, Application - 15/994,837)

Work Projects

  • Skill Tree Prediction for Adobe Photoshop Tasks (Jul'17 - Dec'17)

    Summary: Designed a LSTM based Recurrent Neural Network that predicts and ranks skill sets required by an end-user for completing a Photoshop tutorial/task. Work to be submitted for peer review in NAACL, 2018.

  • Face2Tools: Deconstructing Image Effects using Online Text Tutorials (May'17 - July'17)

    Summary: Developed a technology that detects makeup/edits given an image and recommends relevant PhotoShop tools & actions to achieve similar effects on a new image. Modified VGG16-Net for facial makeup detection/removal tasks (multiclass classification), built knowledge base of Photoshop tutorials, designed a novel algorithm for tool and effect recognition in tutorials and evaluated methods based on embedding, semantic, syntactic and concordance features for tuple retrieval and ranking. Secured the Best Creative Project Award .

  • Detecting Non-Human Traffic in Web Analytics Data with Labels from One Class (Jan'17 - Sept'17)

    Summary: A novel approach to solve the problem of unavailability of negatively labeled points in the domain of bot detection. Proposed modifications to existing Positive-Unlabeled (PU) learning techniques (one class classification) to tackle the conditional independence assumption for labeled subset selection.

  • Fashion Attribute Prediction for Image Based Similarity in ECommerce Domain (Nov'17 - July'18)

    Summary: Developing an algorithm for content-based representation of apparel images in the form of category & attribute tags for application in similar image retrieval (visual search).

  • CTR Prediction for Display Advertising (Jab'17 - Oct'17)

    Summary: Experimented with logistic regression, random forest classifier, XGBoost and Field Aware Factorization machine (FFM) to predict Click through rate (CTR) on a highly unbalanced click data. Developed an end-to-end service in PySpark and C for FFM based CTR model using word embeddings (Glove) for categorical features

  • Sentiment Analysis & Classification of Reviews about Adobe Products on Social Media Channels (July'16 - Dec'16)

    Summary: Developed an ensemble based approach using committee machines for aspect based sentiment analysis of customer reviews, reported improvements over a single model.

Academic Projects

  • Sentence Classification using Convolutional Neural Networks (Jan'17 - Feb'17)

    Summary: Implemented Yoon Kim's Paper in Keras and compared the model's performance with respect to word2vec & Glove embedding methods in both static and non-static environment

  • Study of Electrical Activity of Human Brain through Intervention (Aug'15 - May'16) [report]

    Summary: Researched about changes that occur in a human brain while being in natural versus urban landscapes. Built an logistic regression based algorithm that predicts the percentage by which a person’s cortical activity increases in natural landscape. Demonstrated positive restorative changes in brain while being in natural environments.

  • A Genetic Algorithm based Back Propagation Neural Network for Weather forecasting (Feb'15 - Mar'15)[report]

    Summary: Developed a BP based genetic algorithm in MATLAB and C++ for weather forecasting (temperature, humidity and rainfall prediction)