Hi! I'm Thom

I'm a research data scientist at RTI International. I have experience in applied machine learning research and development, working largely in healthcare and education. I enjoy making mixed and multimedia art.

I'm interested in machine (and biological) learning, realtime multi-modal sensor networks, image and signal processing, generative systems, and human-computer interaction.

Research and Development

Applied machine learning research

EqualAIs: Empowering People and Thwarting Machines 2018

Website | Github | FOIA press release | Twitter

Collaborated with a team of machine learning engineers to investigate and build a tool that embeds adversarial attacks in images, preventing face detection systems from finding faces in images. Built an open-source repository providing documentation and instructions for training a face detection adversarial attack. Submitted FOIA to the U.S. Department of Homeland Security seeking information about facial recognition technology and systems in use by the CBP agency and TSA. This project was part of the 2018 Assembly program on Artificial Intelligence and Governance at the Berkman Klein Center at Harvard and the MIT Media Lab.

Note: The API running the EqualAIs tool is currently offline.

Social media user classification 2015 to 2017

Source | APHA 2016 Presentation

Worked with social scientists and data scientists to collect over 15 million e-cigarette related tweets, and developed performant classification models for differentiating types of users who tweet about e-cigarettes. Used NLP techniques to derive behaviorally characteristic features, which were significant in distinguishing types of users.

Kim A., Miano T., Chew R., Eggers M., Nonnemaker J.. Classification of Twitter Users Who Tweet About E-Cigarettes. JMIR Public Health Surveillance 2017;3(3):e63. DOI: 10.2196/publichealth.8060

Zika vaccination site placement optimization 2016

Description | Executive summary

Designed a data science research project for and mentored two students in the Duke iid Data+ program. The team developed a model for disease spread based on transmission characteristics of Zika virus, using a simulated population of Durham County, NC. They determined optimal placement of clinics for minimizing public health impact using Cover Tree assignment, recursive removal of vaccinated residents in the population, and geospatial proximity to clusters of residents.
Endoscopy procedure performance evaluation 2014 to 2015


Collaborated with endoscopists at the University of Virginia Health System to develop models for evaluating procedure performance according to established standards. Contributed to the literature, finding that the presence of a trainee endoscopist accounted for 34% of variation in polyp detection rate in endoscopies. Developed an interactive dashboard for the monitoring of physician performance. Results were presented at the 2015 IEEE Systems and Information Engineering Design Symposium.

Charles, M., Miano, T. N., Zhang, X., Barnes, L. E., & Lobo, J. M. (2015). Monitoring quality indicators for screening colonoscopies. In Systems and Information Engineering Design Symposium (SIEDS), 171–175.

Software and app development

Phosphorescent aerosol image detection 2017 to 2018
Worked with a team of engineers, software developers, and data scientists using SCRUM and test-driven development to build a user-friendly, automated imaging system that quantifies aerosol concentration on skin and provides evaluation statistics on tested protective garments.
Automated lexicography of localized language content 2015 to 2017

Interactive tool

Collaborated with team of epidemiologists and data scientists to build a word2vec model on over 70 million cannabis-related tweets. This tool was built to help researchers stay up-to-date on current trends, providing them with a mechanism for discovering unknown terms of interest.


Empowering People and Thwarting Machines 2018

Conference page | Slides

The EqualAIs project was presented to about 70 people at the 2018 Tom Tom Festival Applied Machine Learning Conference in Charlottesville, Virginia. The discussion raised the problems of the loss of individual privacy and power as well the lack of mechanisms for communicating preferences and consent in our data driven age. Adversarial attacks were briefly introduced as were the approaches EqualAIs took to addressing the aforementioned problems.

Sound-to-Image 2018

Presented at the 2018 Ars Incognita Symposium on Art and Artifical Intelligence in Córdoba, Spain. This project explored the use of CNNs in real-time sound classification and classified sound visualization, using a variety of techniques, like network activation backpropogation.

Artificial intelligence 2017


Recording part one | Recording part two

This was presented to about 150 people at RTI International. My goal was to provide a high-level introduction to artificial intelligence. Some of the questions addressed:

  • What are factors driving the revitalization of "AI"?
  • What is AI?
  • What is deep learning?
  • When is machine learning appropriate?

Ethics in data science and artificial intelligence 2015



This was an open-forum where we discussed ethical dilemmas in data science and machine learning. This was presented to about 30 people at RTI International. Some of the topics addressed:

  • Data vs ethics
  • Design thinking in algorithms
  • Bias in data, consent, and privacy
  • Machine learning and artificial intelligence

Case studies in design thinking, visualization, and communication 2017


This was held for the 2017 data visualization class in the University of Virginia School of Architecture.

Workshop in building interactive data visualizations for the web 2016


This was held for the 2016 data visualization class in the University of Virginia School of Architecture. We discussed the following:

  • Foundations: use of physical media, importance of play, color theory, and perception of space
  • Data: ethics, interpretability, vision, and storytelling
  • Coding: web development overview and coding exercise


Getting started with GPU-driven deep learning 2016

This is a three-part tutorial targeted at folks interested in getting started with deep learning but not quite sure where to begin.

  1. How to build a GPU deep learning machine
  2. Setting up Ubuntu and Docker for deep learning
  3. Using deep learning with style transfer


Immersive art in virtual reality 2016

This was part of the 2016 Datapalooza hosted by the Data Science Institute at the University of Virginia. I set up an HTC Vive and introduced dozens of people to Google Tilt Brush, giving participants an opportunity to experience 3D painting (many for the first time), with a live feed of their work being displayed on a monitor for all to see.



On nature, form, and aesthetics

Studies using physical and computational media. Explore by clicking a thumbnail below!

It would obviously take some getting accustomed, I think, if it should be a matter of taking into one’s eyes that which is up there outside the cave, in the light of the sun.

Plato, The Allegory of the Cave, Republic VII

© 2018 Thom Miano. All rights reserved.