Arjun Chandrasekaran RESEARCHER · COMPUTER VISION · MACHINE LEARNING · NATURAL LANGUAGE PROCESSING 1025 Atlantic Drive, Atlanta, Georgia, USA.  (+1) 323-337-4884 |  [email protected] |  filebox.ece.vt.edu/ carjun |  carjun Education Georgia Tech Atlanta, GA, USA PH.D IN COMPUTER SCIENCE Jan. 2017 - Jun. 2019 • Objective: Make human-AI interaction more natural and collaboration more efficient. • Specific research topics include computational humor, human-AI collaboration.

Virginia Tech Blacksburg, VA, USA PH.D IN COMPUTER ENGINEERING (GPA: 3.96) Aug. 2014 - Dec. 2016 • Worked on visual humor and temporal common sense.

Bangalore Institute ofTechnology Bangalore, India B.E IN ELECTRONICS AND COMMUNICATION ENGINEERING (AGGREGATE: 79%) Sep. 2009 - Mar. 2013 • Designed and implemented a VLSI based video decoder based on the H.264 video encoding standard. Publications We Are Humor Beings: Understanding and Predicting Visual Humor Las Vegas, USA CVPR (CONFERENCE IN COMPUTER VISION AND PATTERN RECOGNITION) July 2016 • Given an abstract scene, we train a model to predict how funny it is. • We also train a model to alter the funniness of a scene – to make a boring abstract scene funny, and vice versa. • Scenes made funny by our model is funnier than the corresponding original human-created scenes 28% of the time.

SortStory: Sorting Jumbled Images and Captions into Stories Austin, USA EMNLP (EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING) Nov. 2016 • We train different models to learn temporal common sense, i.e., the typical temporal sequence of events in the world. • Task: Given a set of jumbled, aligned image-caption pairs, we train a model to sort these into the correct temporal ordering. Preprint Punny Captions: Witty Wordplay in Image Descriptions ARXIV April 2017 • Inspired by a cognitive account of humor appreciation, we employ puns to produce witty descriptions of an image.

It Takes Two to Tango: Towards Theory of AI’s Mind ARXIV March 2017 • For good performance in human-AI teams, we argue that it is important for the human to understand the AI. • We find that with familiarity, humans better understand an AI. Surprisingly, access to its internal states doesn’t improve understanding. Skills Programming Python, MATLAB, C/C++, VHDL, Verilog, Assembly Level, HTML, Javascript. Work Experience Toyota Technological Institute Chicago, USA RESEARCH INTERN, ADVISED BY PROF. MOHIT BANSAL June 2016 - Aug. 2016 • Worked on multi-modal AI projects involving learning “temporal common sense”, and generating funny image captions.

Robert Bosch (RBEI) Bangalore, India ASSOCIATE SOFTWARE ENGINEER, COMPUTER VISION TEAM Aug. 2013 - July 2014 • Worked on a camera ECU for driver assistance functions like lane detection, road sign recognition, pedestrian detection, etc. • Tested camera exposure, calibration code (embedded C++), automated builds (Python), prototyped vision algorithms (MATLAB).

ARJUN CHANDRASEKARAN · RÉSUMÉ