2019 Symposium

Sponsors

  1. Amazon Robotics
  2. Mitsubishi

Judges

  1. Dan Potter, Director of Industry Relations, Brown University Data Science Initiative
  2. Ellie Pavlick, Assistant Professor, Brown University Computer Science
  3. Daniel Ritchie, Assistant Professor, Brown University Computer Science
  4. Jeff Huang, Assistant Professor, Brown University Computer Science

Winners

  1. Writing Robot by Atsunobu Kotani
    Imagine that you are asked to write 'hello' in your handwriting. The model I trained can predict how you would write other words such as 'world' in your handwriting style, without ever asking you to write that characters
  2. Phonologically Informed Low-Resource Speech Recognition for Foreign Accents by Alex Jang
    By using phonology to inform a model, we can remove the need for foreign accented training data to deal with foreign accents in automatic speech recognition
  3. DBPal by Rohin Bhushan, Shekar Ramaswamy, & Nathaniel Weir
    Using a deep model to translate natural language statements to SQL, we present a novel data exploration tool with a natural language interface
  4. (Audience Award) Examining Lexical Differences In South America by Alexander Chase
    This project is an exploration into how data science can be applied to visualize how language is being used and provide a unique perspective of the language variation in South America

Entrants

  1. Bridging the Semantic Gap for Robots: Action-Oriented Semantic Maps via Mixed Reality by Nishanth Kumar
    Our research aims to provide an interface for robots to acquire knowledge directly from humans via Mixed Reality and describe a formalism for how to act in an environment given the necessary information

  2. Value Preserving State-Action Abstractions by Nathan Umbanhowar
    Our research seeks to develop a theoretically rigorous framework for joint state-action abstraction, and show how such a framework can underlie hierarchical reinforcement learning

  3. Topological data analysis of morphological transitions observed in 3-D cell culture by Zachary Neronha
    We have developed an automated method for detecting transitions in morphological phenotypes associated with metastasis to quantify tumor progression using topological data analysis

  4. A Comparison of Probabilistic Rating Systems for Club Soccer by Ozan Adiguzel
    Using English Premier League data, we optimized the parameters for probabilistic rating systems based on a predictive validation approach and compared their predictive capabilities

  5. Computational models of collective cell behavior in tumorigenesis and metastasis by Subhanik Purkayastha
    My lab is interested in using computational models of collective cell behavior to identify the key factors responsible for the abnormal behavior observed in carcinomas and cancer progenitors resulting in the formation a tumor and spread of cancer cells to distal sites

  6. Learning Feature Extraction for Transfer from Simulation to Reality by Josh Roy
    For deep reinforcement learning, I focus on the task of transferring learned policies between tasks with different visual features using three different methods

  7. Phonologically Informed Low-Resource Speech Recognition for Foreign Accents by Alex Jang
    By using phonology to inform a model, we can remove the need for foreign accented training data to deal with foreign accents in automatic speech recognition

  8. Examining Lexical Differences In South America by Alexander Chase
    This project is an exploration into how data science can be applied to visualize how language is being used and provide a unique perspective of the language variation in South America

  9. Sochiatirst: Using Direct Messaging Data for Clinical Psychology  by Grant Fong & Varun Mathur
    Extracting Messaging data is hard, but has promising use cases in clinical psychology

  10. Computational Study of the Efficiency of Road Networks by Christopher Wolfram
    By comparing a distance function defined by travel distance against the great-circle distance, we can evaluate the efficiency of road networks, and observe some of their general properties

  11. LibFilter: Debloating Dynamically-Linked Libraries through Binary Recompilation by Benjamin Shteinfeld
    To help prevent against code reuse attacks,  we implement LibFilter and found that it can reduce the number of functions available in the address space by 76.5% in Coreutils and 52.7% in SQLite

  12. DBPal by Rohin Bhushan, Shekar Ramaswamy, & Nathaniel Weir
    Using a deep model to translate natural language statements to SQL, we present a novel data exploration tool with a natural language interface

  13. Using Unsupervised Clustering to Improve 30-day  ICU Readmission Prediction by Aansh Shah & Jerome Ramos
    We propose a model that first clusters patients based on their diagnoses and then trains separate classifiers for each cluster to predict readmission to the ICU

  14. Transparency Models for Projection and Rendering in Deep Learning by Henry Stone
    While machine learning tools exist to operate on both 2D and 3D representations of the world, the question still remains how to properly close the loop and allow machine learning models to work simultaneously in 2D and 3D domains connected by our physical knowledge of light

  15. Fairness Interventions in the Real World: Quantifying Disparate Impact by Jessica Dai
    How well do machine learning interventions for fairness perform in real-world situations, particularly in comparison to the pre-algorithm data?

  16. Markov Inference Attacks by Jamie DeMaria
     My research is to understand the power and limitations of Markov Inference Attacks in terms of their success rate and efficiency so that we can use these insights to design better encrypted search algorithms

  17. DeepMellow: Removing the Need for a Target Network in Deep Q-Learning by Seungchan Kim
    My research aims to remove the need for a target network in a deep Q-network reinforcement learning algorithm and still ensure stable learning and good performance

  18. Machine Learning for ELVO Stroke Detection by Charlene Wang
    Given that time is crucial in diagnosing and treating ELVOs, we develop deep learning models to automate ELVO detection, accelerate downstream care delivery, and improve patient outcomes

  19. Detecting Deep Fakes Using Multi-Stream Video Authentication by Eleanor Avril
    This project researches if there is a reliable way to detect deep fakes without using neural nets, no matter how realistic the deep fakes are

  20. A Novel Wireless EEG for Seizure Prediction by Nathaniel Goodman, Sumaiya Sayeed, & Isaac Nathoo
    We are attempting to develop a closed-loop, ambulatory EEG system that employs deep learning software to predict and classify seizures in patients with epilepsy

  21. Writing Robot by Atsunobu Kotani
    Imagine that you are asked to write 'hello' in your handwriting. The model I trained can predict how you would write other words such as 'world' in your handwriting style, without ever asking you to write that characters

  22. Shape From Tracing by Purvi Goel, Louden Cohen
    We are trying to solve the shape from shading problem in a new and unconventional way--differentiable pathtracing

  23. Training a General Agent to Play Levels in Super Mario World by Bridging Genetic Algorithms and Supervised Learning by Benjamin Spiegel
    Our goal is to attain a human level of success on complex levels in Super Mario World on which our agent was not trained, showing that we have designed an algorithm that achieves some degree of generalization in the form of knowledge transfer

  24. Variational Temporal Difference Models for Planning  by Evan Cater
    Planning under uncertaintly is difficult. Our research extends on a proposed solution (TD-VAE) to this problem

  25. Wayfair Interior Design 3D Dataset by Montana Fowler
     We are building a pipeline to clean up interior design scenes from Wayfair and make it usable for machine learning research on style, as these scenes were all created by professional designers

  26. Reinforcement Learning with Geometric Linear Temporal Logic in Atari by Kevin Du
    My research extends geometric linear temporal logic-based reinforcement learning to the Atari domain with more complicated algorithms like deep Q-networks

  27. ShareAid: Intelligent Assistive Sharing via Personal Informatics by Andrew Kim & Mounika Dandu
    Our research investigates the use of an intelligent software assistant to substitute for hand-posting content from our lives to social media

  28. Flight, Camera, Action! Using Natural Language and Mixed Reality to Control a Drone by Deniz Bayazit
    In this research, we offer a mixed reality interface for controlling a drone with natural language

  29. SleepCoacher: Flexible Self-Experiments for Sleep by Lisa Wang & Cintia Araujo
    We investigate how people choose a self-experiment, how long they want to conduct it for, whether maximizing user agency makes self-experimentation appealing to novices, and what can be further improved to make the framework more effective

  30. Personalized Eduction @ Scale by Madeline Griswold
    We are aiming to create a platform that will schedule when users should review terms to maximize memory retention over time

  31. heyPillow by Valerie Nguon & Claire Chong
    Can our fabricated sensing pillow system encourage people to sleep in different positions to improve their sleep or affect sleep behavior when unconscious?

  32. Projection and Deformation of High Detail Geometry onto Low Detail Surfaces by James Guesman
    The question we posed is, "what is the best way to convincingly project and deform high-detail geometry onto a low-detail real-world object?"