Camelia D. Brumar

Visual Analytics Lab, Tufts University · Medford, MA · camelia_daniela.brumar[at]tufts[dot]edu

I am a Ph.D. student in Computer Science at Tufts University. I am working with Dr. Remco Chang in the Visual Analytics Lab ([v]alt) and my research areas of interest are Data Visualization, Machine Learning and Statistics.

Website last modified during Spring 2021. For last updates, please see my CV.


Publications

PIXAL: Visualizing Explainable Anomalies through Predicate Induction

In submission.

Brian Montabault, Camelia D. Brumar, Michael Behrisch, Remco Chang.
TVCG 2021

A Log-Rectilinear Transformation for Foveated 360-degree Video Streaming

Honorable mention.

David Li, Ruofei Du, Adharsh Babu, Camelia D. Brumar, Amitabh Varshney.
[paper]
IEEE VR - TVCG 2020

Application of Approximate Matrix Multiplication to Neural Networks and Distributed SLAM

Co-first author.

Co-first authors: Brian Plancher (Harvard), Camelia D. Brumar (Worcester Polytechnic Institute), Iulian Brumar (Harvard), Lillian Pentecost (Harvard), Saketh Rama (Harvard), David Brooks (Harvard)
[paper]
IEEE HPEC 2019

Experience

Data Science Intern

Alife Health, San Francisco, CA - Remote

  • Working on deep neural networks explainability, interpretability and visualization using methods such as integrated gradients, occlusion sensitivity, Grad-CAM, SmoothGrad, etc.

June 2021 - Present

Graduate Research Assistant

Visual Analytics Lab ([v]alt), Tufts University

  • Researching how to visualize semi-structured data such as network graphs using Graph Neural Networks, and how to create a tool that allows users to prototype such models without any coding involved. This project will potentially have applications in biology, COVID’19 data visualization, software visualization, etc.
  • Co-authoring a paper based on the creation of a visual analytic system for explainable anomaly detection. Designed the visualization which is able to display a multi-dimensional data set and detect where the anomalies most probably occurred by using a parallel coordinates visualization integrated with a violin plot for each dimension.
  • Research assistant on NSF Collaborative Research: Converging Genomics, Phenomics, and Environments Using Interpretable Machine Learning Models [read more]. Managing two research assistants/interns through the research and design process of integrating our platform.
  • Research assistant for The Walmart Foundation, designing an experiment on gamifying a shopping website.

May 2020 - Present

Undergraduate Research Assistant

AR/VR Lab, University of Maryland

  • Worked with Prof. Amitabh Varshney
  • on the paper "A Log-Rectilinear Transformation for Foveated 360-degree Video Streaming", which was submitted and accepted to IEEE VR 2020 conference with a honorable mention.
  • Worked on a computer vision project. Developed a user interface for creating animations starting from an individual image (photograph or painting). Reproduced results observed in the paper "SinGAN: Learning a Generative Model from a Single Natural Image".
  • Worked with the following tools: SinGAN library, Slurm Worload Manager, Node.js, PHP and the Laravel framework.
November 2019 - May 2020

Software Engineering Intern

Bose Corporation

  • Developed a demo Android App and a Python Dockerized microservice in the Bose cloud. Prototyped a new Dynamic App UI experience by fetching dynamic resources and configuration from the cloud to display them in a mobile app.
  • Worked with the automation team on a research project about how to port the code that automates the UI tests for Bose Music App from Python to Kotlin.
May 2019 - August 2019

Undergraduate Research Assistant

Worcester Polytechnic Institute

March 2019 - July 2019

Education

Tufts University

Ph.D. in Computer Science
Currently

University of Maryland, College Park (UMD)

Bachelor of Science, Mathematics
Graduated with honors, GPA: 3.68.
June 2020

Projects

Dota 2 Counters

3rd Prize Overall, Data Day Grindx, Major League Hacking

Dota 2 is a highly complex, highly acclaimed computer game. Before each game 10 players, distributed in two opposing teams, are involved in a several minutes-long drafting phase in which they choose heroes. Both novice and expert players of this game struggle to learn new ability combos that will surprise their opponents and help them earn victories and game experience. We have developed a dataset and a visualization tool that is capable of rendering the over 7000 relationships between the 119 heroes and guide each one of the trillions of possible Dota 2 draft phases (Combinations(119, 10) = 10^14). We focus on the hard problem omitted by other Dota 2 pickers which is "why a hero is more effective than another" rather than just create a ranking of possibly effective heroes. For more info, check out the associated Devpost publication, or my Devpost portfolio.



Some experiments with the SinGAN library

Graphics and Visual Informatics Laboratory, UMD

I am currently playing with SinGAN, an unconditional generative model that can be learned from a single natural image. It turns out that an image contains enough information that can be captured by SinGAN, which will create new unique samples of it that carry the same visual content as the original picture. In my case, I am focusing on producing animations from a single image.



CS 4731 Computer Graphics

Worcester Polytechnic Institute
Link to the description of this course.

(To properly see the animations, please open this website on a computer.)

I personally enjoyed two projects from this course. In the first project, the objective was to load a mesh stored in the .ply file format, render it as a 3D wireframe model using Vertex Buffer Objects and also add keyboard control that lets us move the object around.

The task in the second project was to model and animate a kinetic sculpture (Alexander Calder's hanging mobiles). To do so, we explored hierarchical modeling of 3D meshes, lighting and shading. We got used to the transformations and hierarchical modeling in WebGL using a matrix stack. The keys that can be used to play with the hanging mobile--I call it MegaMobile5000- are:

kinetic_structure

Know Yourself

Independent Project · Machine Learning + Mobile Application

During my first two years of college, I created a stress-detection wearable application, which I called Know Yourself. The objective of the application is to predict the well being of its user as correlated with the day of the week, hour of the day, number of hours slept and as related to the weather (temperature, pressure, humidity, etc.) and other variables. The challenge was to fit an inference model into my wearable device with a good balance of speed and accuracy. I’ve converged upon a Random Forest implementation which is not as computationally intensive as a deep neural net and could exploit the non-linear nature of the problem.

CS 480X Data Visualization

Worcester Polytechnic Institute
Link to the course website.

In this course I learned to evaluate the effectiveness of visualizations for specific data, task, and user types. I implemented visualization algorithms and undertook projects involving the use of commercial and public-domain visualization tools. In this course we focused on the usage of the d3.js library. Here are some highlights of my work:

  • Mini animation.
  • 10 ways of showing the same data.
  • Improved the Number Of Rockets Launched Chart
  • Final Project - (GitHub repo) I worked with my team on the final project, which aims to provide an informative and exploratory way of presenting information about Wikipedia articles, links and citations. Through the assistance of freely available APIs, we created a website that can be used to browse specific articles of Wikipedia, view similar articles to the ones searched for, and provide exploratory analysis of which of these articles were cited the most. We hope to have provided a tool that can assist in research while also giving freedom to discover new links and connections between information in one of the internet’s most popular encyclopedic websites.

The Party Donator

HackUPC Winter Hackathon, Barcelona, Spain

Someone once said: "discipline is the bridge between goals and accomplishments". So you decide to go out, but of course tomorrow you have work to do. You'd like to get back home early, but you also know how easy it is to lose track of time. When you go out with your friends, you know when you leave but you also need a way to com home in time. Luckily for you, the Party Donator has been developed to help you. You only have to tell Alexa that you're going out and when you expect to return. She will remember that time. For each half an hour you're late, she will donate 2 euros to a predefined NGO. Therefore, you will want to get home early for an extra reason, but still if you're late you'll be helping people.


Skills

Programming Languages & Tools
  • PYTHON. Used in the Data Analytics and Statistical Learning course at WPI. My team and I worked on a project which had as a goal gaining insight into what qualities make songs popular in different parts of the world. More specifically, we aimed to predict the popularity of a song in a certain country based on the features from our dataset.
  • JAVASCRIPT, HTML and WEBGL. Used in the 4000 level courses Computer Graphics and Data Visualization at WPI. More details about the projects sone in these courses can be found under the Projects section.
  • ANDROID DEVELOPMENT IN KOTLIN/JAVA. Learned both by independently working on personal projects, such as The Wear Minesweeper and Know Yourself, and by interning at Bose Corporation.
  • C++ DEVELOPMENT. Learned during contest preparations and courses taken at UPC such as the freshman "Numerical Analysis" and the sophomore "Algorithms" courses. Solved 158 Jutge.org programming problems.
  • MATLAB. Used mainly in the courses Numerical Analysis at UPC and Numerical Analysis of Differential Equations at WPI, and in my research with Prof. Zhang.

*WPI: Worcester Politechnic Institute.

**UPC: Universitat Politècnica de Catalunya.



Interests

Apart from being a research student, I enjoy my free time being indoors and outdoors. I LOVE indoor rock climbing. I would say I am more of a bouldering person (max: V4 for now), but I like belaying (max: 5.10+), too. Otherwise, I can easily get competitive in ping pong and squash.

In the winter, I am an avid skier. During the warmer months, I enjoy hiking (I hicked La Pica d'Estats, the highest mountain in Catalonia, Spain!).

Besides sports, I like reading DC and Marvel comics, and drawing 2D animations.