Portrait

Dylan Cashman

dcashm01 _at_ cs _dot_ tufts _dot_ edu

PhD Student

Tufts University

Computer Science

I am a PhD candidate under Remco Chang in the Visual Analytics Lab at Tufts (VALT), with an expected graduation in May 2020. I am also teaching DS 4200 - Information Presentation and Visualization at Northeastern University as a part-time lecturer in the Spring semester of 2020. I live in beautiful Lynn, Massachusetts with my wife and wonderful baby son.

I received my Masters of Science in Computer Science at Tufts University in 2016, and my Bachelor of Science in Mathematics at Brown University in 2010.

In the winter of 2017, I passed my Qualification Exams, taking tests in Visualization, Human Computer Interaction, and Statistical Pattern Recognition.

Before grad school, I built Ruby on Rails systems for healthcare nonprofits in the Boston area at Annkissam. During my graduate study, I have interned at MIT Lincoln Laboratory, Palo Alto Research Center, and the MIT IBM Watson AI Lab.

Github

C.V.

Visual Analytics Lab at Tufts

Current Projects

  • The Human Role in Classical Model Selection

    PAC Learning for Model Selection

    It can be difficult to pin down exactly what the role of the user is in human-in-the-loop systems for machine learning. We know that users gain trust and understanding by being involved in the process. And we know that users can be valuable in labeling unlabeled data and in identifying data cleaning issues. But there are theoretical underpinnings of machine learning that can help us classify the types of errors that machine learning can have when generalizing to their deployed settings. These include model mismatches, changes in data distribution, or dependences within the data sampling. Looking into learning theory may provide theoretical foundations for why visualization is such a useful aspect of model selection.


  • Visual Analytics for Neural Architecture Search

    VA for NAS

    Model selection is a traditional task in machine learning in which chooses a learning algorithm and a hyperparameter setting that optimizes a cost function over a given training (and optionally, validation) dataset. In order to use neural networks, there is an additional complication in the need to choose an architecture for the network. By visualizing the search and providing affordances for controlling the search, we can empower the user to steer the discovery of architectures for their usage scenario.

Selected Publications

  • REMAP screenshot

    D. Cashman, A. Perer, R. Chang, H. Strobelt, "Ablate, variate, and contemplate: Visual analytics for discovering neural architectures. Transactions on Visualization and Computer Graphics (TVCG), 2019.

    Paper

    Code

    Slides

    Talk (Vimeo)

  • Snowcat screenshot

    D. Cashman, S. Humayoun, F. Heimerl, K. Park, S. Das, J. Thompson, B. Saket, A. Mosca, J. Stasko, A. Endert, M. Gleicher, R. Chang. A User-based Visual Analytics Workflow for Exploratory Model Analysis. Computer Graphics Forum (CGF), 2019.

    Paper

    Slides

    Video Fast Forward

  • BEAMES screenshot

    S. Das, D. Cashman, R. Chang, A. Endert, "BEAMES: Interactive Multi-Model Steering, Selection, and Inspection for Regression Tasks " Symposium on Visualization in Data Science (at IEEE VIS), 2018. Best Paper Award

    Best Paper Award

    Paper

  • Vanishing GradientVanishing Gradient

    D. Cashman, G. Patterson, A. Mosca, N. Watts, S. Robinson, R. Chang, "RNNbow: Visualizing Learning via Backpropagation Gradients in RNNs" IEEE Computer Graphics and Applications, 2018.

    D. Cashman, G. Patterson, A. Mosca, R. Chang, "RNNbow: Visualizing Learning via Backpropagation Gradients in Recurrent Neural Networks" Workshop on Visual Analytics for Deep Learning (at IEEE VIS), 2017.

    Best Paper Award

    Paper

    Demo

    Slides

    Talk (Vimeo)

  • Bayesian Detection

    B. Price, L. Price, D. Cashman, M. Nabi, "Efficient Bayesian Detection of Disease Onset in Truncated Medical Data" IEEE International Conference on Healthcare Informatics, 2017.

Preprints

  • Interface of NNCubes

    Z. Wang, D. Cashman, M. Li, J. Li, M. Berger, J.A. Levine, R. Chang, C. Scheidegger, "NNCubes: Learned Structures for Visual Data Exploration" arXiv preprint arXiv:1808.08983, 2018.

    Paper

Posters and Workshop Papers

  • Inferential Tasks

    D. Cashman, Y. Wu, R. Chang, A. Ottley, "Inferential Tasks as a Data-Rich Evaluation Method for Visualization" Workshop on Evaluation of Interactive Visual Machine Learning Systems at IEEE VIS, 2019.

    Paper

    Slides

  • Clipped Projections

    B. Kang, D. Cashman, R. Chang, J. Lijffijt, T. De Bie, "CLIPPR: Maximally Informative CLIPped PRojections with Bounding Regions" Posters for IEEE Conference on Visual Analytics for Science and Technology, 2018.

  • Big Data, Bigger Audience

    D. Cashman, S. Kelley, D. Staheli, C. Fulcher, M. Procopio, R. Chang, "Big Data, Bigger Audience: A Meta-algorithm for Making Machine Learning Actionable for Analysts" Posters for VizSec, 2016. Also presented at MIT Lincoln Labs Cyber and Netcentric Workshop CNW 2017

    Extended Abstract

    Poster (PPT)

Awards

  • Best Paper, Symposium on Visualization for Data Science, IEEE Conference on Visualization, Berlin, Germany, October 2018.

  • 3rd Place, Tufts Graduate Research Symposium, Tufts University, 2018.

  • Best Paper, Workshop on Visual Analytics for Deep Learning, IEEE Conference on Visualization, Phoenix, AZ, October 2017.

  • Provost's Fellowship, Tufts University, 2016-2018.

Talks

  • D. Cashman, G. Patterson, A. Mosca, N. Watts, S. Robinson, R. Chang, "RNNbow: Visualizing Learning via Backpropagation Gradients in Recurrent Neural Networks" Tufts Graduate Research Symposium , 2018

    3rd Place

    Slides

  • D. Cashman, F. Yang, J. Chandler, A. Mosca, M. Iori, T. August, R. Chang, "Chasing Waldo: Implicit Recovery of User Behavior and Intent from User Interaction Logs" Tufts Graduate Research Symposium , 2017
  • D. Cashman, "Color Spaces and Color Places" Tufts REU Lecture Series , Summer 2017
  • D. Cashman, "Big Data, Bigger Audience: A Method for Adapting Statistical Methods for a Wider Audience of Users" Tufts IGNITE , 2015
  • D. Cashman, "Introduction to Ruby" and "Models, Scaffolding, and Migrations", Railsbridge Boston , 2013

Teaching

Lecturer

  • DS 4200: Information Presentation and Visualization. Northeastern University. Spring 2020

Guest Lecturer

Teaching Assistant

  • COMP 40: Machine Structure and Assembly Language Programming. Tufts University. Fall 2016, Spring 2017
  • COMP 61: Discrete Math. Tufts University. Fall 2015
  • Math 0520: Linear Algebra. Brown University. Spring 2008
  • Math 0200: Multivariable Calculus. Brown University. Fall 2008, Fall 2009
  • Math 0190: Calculus II. Brown University. Fall 2007

Hobbies

I used to play classical upright bass; I still play guitar sometimes. I'm a big music guy and I try to constantly expand what I'm listening to, both in genre and in time period. I like reading and I try to alternate between something fun and something important. My brothers and cousins and I all have a scheduled night every two weeks to play some dumb online videogames together. I was really into pickup basketball, but I'm afraid I'll hurt my knees if I play too frequently.

Oh, and watching TV series over and over again. Way too many times.