Colin Shea-Blymyer

Machine Learning Research | PhD Candidate | Oregon State University

sheablyc@oregonstate.edu

ABOUT

Colin is a PhD student of computer science and artificial intelligence at Oregon State University. He researches how to ensure the machines of the future behave ethically with Houssam Abbas. This research uses formal logic to express ethical specifications, machine learning to search through specifications, reinforcement learning to train agents, and formal methods to prove an agent abides by a specification.

Colin's research interests include ethics in artificial intelligence, robustness in machine learning, automated scientific discovery, data mining, and logic.
You can read his CV here.

In the past, Colin has researched the automation of scientific discovery under Dr. Benjamin Jantzen in Virginia Tech's Digital Philosophy Lab. He has worked in industry on projects in security for machine learning, policy and artificial intelligence, cyber-security, and in simulation validation, and data analysis.

PUBLICATIONS

Model Checking the Optimal Behavior of Big Markov Processes
Describing and verifying on-policy behaviors in large Markov decision processes
Workshop Paper

FoMo - Formula and Model Generation for Learning-Based Formal Methods
Generate data to train a neural network for jointly embedding formulas and system models
Workshop Paper

Generating Deontic Obligations from Utility-Maximizing Systems
Describing, verifying, and discovering ethical obligations in Markov decision processes
Conference Paper

A General Metric for the Similarity of Both Stochastic and Deterministic System Dynamics
Measuring differences in symmetry of system changes
Journal Paper

Entropy 23(9) — 2021
READ PDF

Algorithmic Ethics: Formalization and Verification of Autonomous Vehicle Systems
Formal verification of the ethics in automata
Journal Paper

ACM TCPS 5(4) — 2021
READ PDF

Learning A Robot's Social Obligations from Comparisons of Observed Behavior
Democratizing the specification of ethical constraints for robots
Conference Paper

ARSO 2021
READ PDF

A Deontic Logic Analysis of Autonomous Systems' Safety
Using deontic logic to model, verify, and ensure system obligations
Conference Paper

HSCC 2020
READ PDF

Differentiation of Collective Behavior Based on Automated Discovery of Dynamical Kinds
Analyzing swarm behavior with a general metric for system dynamics similarity
Conference Paper

DSCC 2018
DOWNLOAD PDF

Exploration of Extraterrestrial Planets
Using automated intelligent systems for remote geological science
Workshop Paper

NASEC 2014
DOWNLOAD PDF



PROJECTS

Deontic Logic for Normative Autonomous Systems
Review of logics and problems

Qualifying Exam Paper - OSU 2021
DOWNLOAD PDF

Inspectable Incrementality in Minimum Feedback Arc Set Solving
Leveraging features of SMT solvers

Theory of Computation - OSU 2021
DOWNLOAD PDF

Dynamic Robot Barriers
Reinforcement learning to coordinate robots that increase flow in crowd simulation

Multiagent Systems - OSU 2020
DOWNLOAD PDF

Forecasting of Spatio-temporal Chaotic Dynamics with Self-Attention
Comparing self-attention mechanics to reservoir computing in deep forecasting of chaos

Deep Learning - OSU 2020
DOWNLOAD PDF

Peeking Behind the Mask: Modeling Belief in a Game of Mascarade
Examining the structure of belief models in a board game

Probabilistic Graphical Models - OSU 2020
DOWNLOAD PDF

The Colors of Love: Topic/Image Color Mining via Clustering
Exploring the associations between topics and colors through clustering algorithms

Convex Optimization - OSU 2019
DOWNLOAD PDF

Adversarial Control of Neural Network Policies
Optimizing images to force autonomous vehicles to misbehave

Machine Learning and Security - VT 2017
DOWNLOAD PDF

Comparative Analysis of Causal and Random DAGs
Searching for patterns in how we search for causality

Data Mining Large Networks and Time Series - VT 2017
READ MORE

Chaos and Biodiversity in Stochastic Lotka-Volterra Models
Exploring the behavior of chaotic regimes in a stochastic population model

Computational Cellular Biology - VT 2017
DOWNLOAD PDF

Drug Repositioning
Finding new uses for drugs with machine learning

Data Analytics - VT 2016
DOWNLOAD PDF

RESEARCH AREAS

ETHICS IN ARTIFICIAL INTELLIGENCE

I am currently researching techniques to ensure AI acts ethically. I am designing a logic of ethics to specify what an RL agent's obligations are; I am writing code to prove that an agent has a given obligation; I am studying algorithms to explore the space of ethical obligations; and I am creating methods to synthesize control for an agent that respects ethical guidelines or population preferences. This work led to, and is partially supported by Dr. Abbas's NSF CAREER award.

AUTOMATION OF SCIENTIFIC DISCOVERY

I developed a suite of algorithms that determines similarity between dynamical systems by comparing how a system evolves under different conditions. This work led to, and was partially supported by Dr. Jantzen's NSF CAREER award.
Read More Here

ADVERSARIAL MACHINE LEARNING

As a research intern with MITRE, I helped develop a taxonomy and dictionary of adversarial machine learning techniques, included in a draft NIST Interagency Report. I wrote the groundwork documentation for the development of an AML testing lab, and researched existing policy and best practices on AML testing.



CONTACT

Email
sheablyc[at]oregonstate[dot]edu

Social Links

Research Gate: LinkedIn: