SciMargin
In press · ICAI 2026Where do decision boundaries expose groups to error?
Class-conditioned boundary-exposure diagnostics for studying fairness degradation under distribution shift. Evaluated on Adult, COMPAS, and German Credit.
PHD RESEARCHER · UNIVERSITY OF CENTRAL FLORIDA
Mayra Bornacelly Castaneda
I study how neural decision systems behave when the data changes, and what that means for reliability and fairness.
Conceptual illustration, not experimental results.
At the intersection of
Machine learningIndustrial engineeringHuman factors01 / RESEARCH
My work examines model vulnerability and subgroup error patterns through gradients, learned representations, and decision-boundary geometry.
I connect these diagnostics with risk-aware human–AI decision support. Three complementary lines of research shape that work.
Where do decision boundaries expose groups to error?
Class-conditioned boundary-exposure diagnostics for studying fairness degradation under distribution shift. Evaluated on Adult, COMPAS, and German Credit.
What can gradients reveal after a model has converged?
A co-developed, supervised diagnostic of post-convergence optimization pressure, measured with model parameters held fixed, to study reliability under changing data conditions.
Can a shared perturbation explain different group errors?
A geometric and finite-sample framework for testing whether a specified perturbation in representation space can jointly explain group-specific error patterns.
PyTorch · Automatic differentiation · GPU experimentation · Representation analysis · Subgroup error analysis
02 / SCHOLARSHIP
Published work, accepted manuscripts,
and conference contributions.
Statuses reflect the November 2026 CV. In-press work and accepted presentations are listed separately from published articles.
03 / TEACHING & COMMUNITY
IN THE CLASSROOM
STA 3032 · University of Central Florida
As an Instructor of Record and Graduate Teaching Associate, I develop assessments, instructional materials, and applied course projects.
May 2026–present · Previously Graduate Teaching Assistant, August 2025–May 2026
UPCOMING · SPRING 2027
In personSTA 3032H · Section 0202 · University of Central Florida
Scheduled instructor for the in-person honors section in Spring 2027.
MENTORSHIP & LEADERSHIP
INFORMS Mentor Match mentor, with a focus on machine learning, undergraduate research, and preparation for summer opportunities.
Graduate Advisor to the IEOM Student Chapter at UCF, supporting recruitment, faculty coordination, and academic and professional activities.
Previously: In2science science mentor, University of Melbourne.
SELECTED ACADEMIC SERVICE
SUMMER 2026 / STUDENT FEEDBACK
4.72/ 5
Course & Content
4.61/ 5
Instructor & Instruction
Source: UCF Summer 2026 SPI report. These are separate category scores, not a combined overall rating.
SELECTED INVITED CONTRIBUTIONS
Advanced Manufacturing and Industrial AI Innovation · Irvine, USA
Redefining Problem Solving in the Digital Era · IEOM North America, Milwaukee
Southern Arkansas University
04 / ACADEMIC PROFILE
My path brings together computer science, engineering management, economics, and data science, alongside professional experience in data and IT operations.
University of Central Florida · USA
Cumulative GPA: 4.00 / 4.00University of Melbourne · Australia
Pontificia Universidad Javeriana · Colombia
Graduated with honorsUniversidad de La Sabana · Colombia
Research-based degreeUniversidad San Martín · Colombia
Graduated with honors05 / LET’S CONNECT
I’m interested in US research fellowships and academic collaborations focused on reliable, fair, and risk-aware machine learning.
Find me on ORCIDMayra Bornacelly Castaneda
PhD Researcher · Industrial Engineering