Active projects

• Caravaner: Campus Rideshares

  Spend less on fast, flexible transit between campus and the airport. A self-serve rideshare matching platform for university students publicly launching at Stanford soon.

Active research

• Classifying customer support requests with GPT-3 data augmentation (natural language processing, machine learning)

  Mentored by Silei Xu, Sina Semnani, and Prof. Monica Lam

• CORELS: Certifiably Optimal Rule Lists (interpretable machine learning, systems optimization)

  Mentored by Prof. Margo Seltzer and Prof. Cynthia Rudin
  CORELS is a machine learning algorithm that provides optimal human-interpretable models called rule lists, or one-sided decision trees.
  My contributions:
  • Designed web user interface (source code)
  • Created R language API
  • Key contributor to parallel implementation of algorithm
  • Publications: Co-author on upcoming and published papers on performance optimizations

Past research

• KnowBias: Predicting Political Bias in Text (natural language processing, machine learning)

  A tool that displays the degree of political bias in content such as tweets, online articles, and plaintext files.
  • Winner of the 2018 Congressional App Challenge for Massachusetts's 5th Congressional District
  • Web version available at knowbias.ml and extensions available for Firefox and Chrome
  • Publications: short paper, long paper



• Revisiting Ensembles in an Adversarial Context: Improving Natural Accuracy (adversarial machine learning)

  Mentored by Guillaume Leclerc and Prof. Aleksander Mądry through MIT PRIMES
  Evaluating the efficacy of ensembling deep learning models for image classification in the context of adversarial robustness.
  • Developed ensemble schemes that maintain same level of adversarial robustness as a single model but improve natural accuracy
  • Publications: final report



• Concache: Concurrent Hashmaps in Rust (parallel computing, systems optimization)

  Mentored by Jon Gjengset and Prof. Frans Kaashoek through MIT PRIMES
  A lock-free, linked-list based concurrent hashmap in Rust using epoch-based memory reclamation.
  • Implemented two hashmaps released as Rust crates
  • Presented analysis of various Rust type-system properties that help and hurt developers writing safe and correct concurrent code
  • Work featured at September 2018 Boston Rust Meetup and at the October 2018 MIT PRIMES Computer Science Conference
  • Publications: final report

Publications

• Revisiting Ensembles in an Adversarial Context: Improving Natural Accuracy (February 2020)

  Aditya Saligrama, Guillaume Leclerc

  Presented at ICLR 2020 Workshop on Towards Trustworthy ML: Rethinking Security and Privacy for ML, April 26, 2020.
  Available at arXiv:2002.11572.

• KnowBias: Detecting Political Polarity in Long Text Content (September 2019)

  Aditya Saligrama

  Presented at AAAI 2020 Student Abstract and Poster Program, February 9, 2020. Last revised November 2019.
  Available at arXiv:1909.12230

• KnowBias: A Novel AI Method to Detect Polarity in Online Content (May 2019)

  Aditya Saligrama

  Available at arXiv:1905.00724. Last revised November 2019.

• A Practical Analysis of Rust's Concurrency Story (April 2019)

  Aditya Saligrama, Andrew Shen, Jon Gjengset

  Available at arXiv:1904.12210

• Systems Optimizations for Learning Certifiably Optimal Rule Lists (February 2018)

  Nicholas Larus-Stone, Elaine Angelino, Daniel Alabi, Margo Seltzer, Vassilios Kaxiras, Aditya Saligrama, Cynthia Rudin

  Presented at SysML Conference 2018