Romina Mahinpei

I am a fully-funded Master of Science & Engineering (MSE) student at Princeton University’s Computer Science Department.

I received a Bachelor’s of Science (BSc) degree in Honours Computer Science along with a Mathematics Minor at the University of British Columbia (UBC). During my undergrad, I had the pleasure of working with the Systopia lab and the Numerical Linear Algebra lab at UBC’s Computer Science Department.

My research experience spans human-computer interaction, applied machine learning & artificial intelligence, and computer systems (focusing on high-performance computing). My current research interests are in human-centred AI, social computing, and educational technologies.

Email | CV | LinkedIn | OrcID

Projects

Leveraging Collaborative Filtering for Personalized Practice in Computer-Based Assessments [Code] [Paper]

Explored collaborative filtering (CF)-based recommender systems for personalizing question selection in computer-based assessments by predicting student performance on new, unseen questions using past scores.
Developed and evaluated six CF models against baseline model using Dietterich’s 5×2 cross-validation method on two datasets from an undergraduate computer systems course and assessed model performance using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), with statistical significance determined through a paired t-test.

Mixed Precision Minimal Residual (MINRES) Method [Code] [Paper]

Developed two mixed precision variants of the Minimal Residual (MINRES) method for solving sparse saddle-point linear systems, allowing users to adjust the precision of preconditioner solves and matrix-vector products.
Implemented these variants and a baseline version using CUDA C and conducted a comparative performance study on NVIDIA’s GeForce RTX 3070 Ti, evaluating speed-up across three Maxwell and three Stokes saddle-point problems.

Low Precision Training of Deep Learning Models [Code] [Paper]

Analyzed the effects of four low precision training schemes on both the training time and classification accuracy of four deep learning models from the domains of image and text classification.
Implemented these low precision variants and a baseline version using TensorFlow and conducted a comparative performance study on Google Colab’s Tesla T4 GPU to evaluate the trade-offs in speed and accuracy for each model.

A Generalized Framework for Describing Question Randomization [Code] [Paper]

Conducted a thematic analysis of computer-based assessment questions in undergraduate CS courses to investigate how instructors integrate randomization into their assessment questions.
Developed a six-level framework to systematically categorize these randomization strategies and validated it by applying the framework to questions from an undergraduate data structures course (N = 86) and an undergraduate computer systems course (N = 100), assessing inter-rater reliability using Gwet’s AC1 coefficient.

Enhancing Transparent Model Reporting with Automated Model Cards in Tribuo [Code] [Paper]

Conducted semi-structured interviews with ML and data professionals from academia (N = 6) and industry (N = 6) to assess current workflows and explore how data provenance could streamline these processes.
Identified ML model documentation as a key pain point for many participants and subsequently designed, implemented, and tested the Model Card package for Tribuo, Oracle’s open-source Java ML library, to facilitate the creation of partially automated documentation cards for ML models.

Publications

Romina Mahinpei †, Iris Xu †, Steven Wolfman, and Firas Moosvi. 2024. A Generalized Framework for Describing Question Randomization. In Proceedings of the 55th ACM Technical Symposium on Computer Science Education V. 2 (SIGCSE 2024). Association for Computing Machinery, New York, NY, USA, 1736–1737. https://doi.org/10.1145/3626253.3635599.

† Both authors contributed equally to this work.

Teaching

As a teaching assistant, I’ve had the pleasure of helping with the following courses:

CPSC 313 Hardware & Operating Systems, UBC (2023-2024)
CPSC 210 Software Construction & Development, UBC (2022-2023)
Science One Differential & Integral Calculus, UBC (2021-2022)