Shreya Rajpal Resume 2g4d20

Shreya Rajpal +91-9999805820

Education Bachelor of Technology - Production and Industrial Engineering (8.225/10.0) Minor - Computer Science (Coursework GPA - 9.235/10.0) Indian Institute of Technology, Delhi

[email protected]

Expected: May 2016

Publications [1] Shreya Rajpal, Karan Goel, Mausam. POMDP-Based Worker Pool Selection for Crowdsourcing. CrowdML Workshop, International Conference on Machine Learning (ICML). Lille, . July 2015. Relevant Coursework Computer Science: Introduction to Computer Science [topped among 350+ students], Data Structures, Analysis and Design of Algorithms, Artificial Intelligence, Probabilistic Graphical Models Electrical Engineering: Digital Electronics (+ Lab) Industrial Engineering: Probability and Stochastic Processes, Operations Research Research Experience Cost-Quality-Time Optimization in Crowdsourcing [report | slides] with Prof. Mausam (IIT-Delhi)

July 2015 - Present

• Modeled the optimization of a 3-variable reward function, in a multi-agent setting using Markov Decision Processes (MDPs). Agents make online decisions for task pricing, prioritization and termination. • Utilized a Thinned Non-Homogenous Poisson Process to model worker arrivals to the marketplace, with a Discrete Choice Model to estimate task selection. • Exploring the application of Decentralized Partially Observable MDPs (DEC-POMDPs) to this paradigm. Modeling Human Perception for Discovering Consensus Rankings [report] with Prof. Aditya Parameswaran (UIUC), Collaborator: Prof. David Forsyth (UIUC) July 2015 - Present • Proved NP-Hardness for the problem of discovering multiple latent consensus rankings when humans sort a set of items, by reducing from the Minimum Clique Cover problem in graphs. • Conducted experiments on Amazon MTurk to study perception of depth with ambiguous visual cues, by adapting Cornell’s OpenSurfaces interface to our application, and deploying it on Amazon EC2. • Developing a Conditional Random Field model to classify workers based on similarity of perception. Worker Pool Selection in Crowdsourcing [1] [paper | poster | slides] with Prof. Mausam (IIT-Delhi) January 2015 - May 2015 • Developed an online task routing algorithm using Partially Observable MDPs (POMDPs), with a novel approach that leverages an organization of workers based on their qualifications. The model systematically trades off push and pull crowdsourced approaches in a single, theoretical framework. • Experiments in simulation and on real Amazon MTurk data demonstrated improvement over baselines. • Accepted at the International Conference on Machine Learning’s CrowdML workshop. Teaching Experience Artificial Intelligence, Undergraduate and Graduate-Bridge Course [website] with Prof. Mausam (IIT-Delhi) July 2015 - Present • Developed, implemented and graded 2 assignments on developing AI game bots. • Taught 125+ students concepts in MDPs and POMDPs after creating problem sets for them. • Undertook several one-on-one teaching sessions for a student with a hearing disability.

Internships Algorithm Performance Metrics for Foreign Exchange Currencies at Deutsche Bank Center, Mumbai, India

May 2015 - July 2015

• Developed a performance metric for High Frequency Trading algorithms and utilized this metric to analyze the performance of frequently used algorithms and find issues with inefficient algorithms. • Developed a heuristic for selecting optimal HFT algorithm given trade volume, using the metric. • Developed an efficient querying scheme using graph pruning algorithms for a massive dataset. Outlier Detection in Large Datasets at LIRIS Lab, University of Lyon-1,

December 2013

• Implemented a Principal Component Analysis based algorithm for anomaly detection, and analyzed its performance with respect to the state-of-the-art. Projects OCR Word Prediction using Conditional Random Fields (CRFs) [report | code 1 | code 2] • Performed Variable Elimination on a CRF and used the Min-Fill heuristic to generate a Clique Tree. • Implemented Sum-Product Message ing, Gibbs Sampling and Loopy Belief Propagation to find the Max-Marginal Assignment and also implemented their Most Probable Explanation counterparts. • Increased model complexity by increasing CRF connectivity to compare algorithm performance. POS and NER Tagging for Tweets using MALLET [code] • Engineered features to improve POS and NER tagging in a CRF model. Grouped similar sounding words via metaphone similarity to tweets and added Twitter specific features. • Studied CRF performance by performing an ablation study by adding features. Viterbi Algorithm for Genomic Sequences [code] • Implemented a Hidden Markov Model to identify G islands in Genomic Sequences. Learned model parameters using Maximum-Likelihood estimation and used the Viterbi Algorithm for prediction. A Review of Convolutional Codes and the Viterbi Algorithm [paper] • Performed a literature survey on the Viterbi Algorithm and its faster, less optimal extensions. • Compared two popular hardware implementation of the Viterbi algorithm. AI for Connect M-N-K • Created two game bots using Minimax with Alpha-Beta pruning, and Monte Carlo Tree Search. • Enabled flexible board configuration handling for a general Connect-4 game board. Skills Languages: C/C++, Python, SQL, Javascript Software: MATLAB, AutoCAD, LATEX Development: HTML, CSS, JQuery, Ajax Activities • Institute journalist for the Board of Student Publications. • Performed stage and street theater for inter-college events. • Regularly act as master of ceremony for institute-wide events.