Reza Arabpour

Quantitative machine learning · Toronto

Reza Arabpour.

Machine learning.
Grounded in finance.

I connect mathematical research with production machine learning. At BMO’s AI Research team, I build systems for forecasting, stress testing, and financial intelligence.

Explore my research View résumé
GitHub LinkedIn Scholar
Across industry & research
BMO Capital Markets & AI ResearchVector Institute Research, 2023–2025McMaster University M.Sc., 2025

01 Selected research

Theory with a
practical purpose.

Financial dynamics, the geometry of learning, and making powerful models more accessible.

Google Scholar

Work in progress2026 · Manuscript in preparation

FinAttn: Interpretable Financial Attention with Closed-Form Risk-Neutral Pricing Heads

Reza Arabpour, Mojtaba Valipour, Mozhgan Saeidi, Anastasis Kratsios

Forecasting return distributions across interpretable stochastic regimes, with a separate risk-neutral component for derivatives pricing.

Research overview
This ongoing work brings financial structure into transformer-based forecasting. The model separates the task of predicting return distributions from a closed-form risk-neutral pricing adapter. The manuscript is in preparation; a public preprint is not yet available.
Analysis and Applications2026 · Accepted / in press

Low-dimensional approximations of the conditional law of Volterra processes: a non-positive curvature approach

Luca Galimberti, Reza Arabpour, John Armstrong, Anastasis Kratsios, Giulia Livieri

Using geometric structure and hypernetworks to approximate the evolving distributions of financial processes with memory.

Read paper
Abstract
Volterra processes capture memory effects that make financial dynamics difficult to model. This work combines dimension reduction onto a statistical manifold of non-positive curvature with a geometry-aware learning model. A hypernetwork updates the model over time to represent changing conditional distributions.
ICML 2025 · ES-FoMo Workshop2025

LoRA Fine-Tuning Without GPUs: A CPU-Efficient Meta-Generation Framework for LLMs

Reza Arabpour, Haitz Sáez de Ocáriz Borde, Anastasis Kratsios

Generating task-specific language-model adapters on a CPU by combining a bank of pretrained LoRAs.

Read paper Code
Abstract
The framework uses relationships between datasets to construct new adapters from existing ones, without fresh gradient-based adapter training. Adapter generation runs on a CPU; the large-scale model inference and evaluation experiments use GPU resources. The resulting adapters improve on the base model, while GPU-trained adapters remain stronger.

02 Experience

From ideas
to implementation.

My work spans financial research, large-scale modelling, and the engineering that brings models into everyday use.

Full experience in my résumé
Currently building at BMO
AI Research (ARC)
BMO · AI ResearchNov 2025 — Present

Quantitative Machine Learning Engineer

Production ML for forecasting, stress testing, and financial workflows.

  • Own a production mortgage-application forecasting model and execution workflows, combining transformer and statistical models.
  • Lead economic-intelligence stress testing and RAG-based agents for scenario analysis.
  • Develop small-language-model tooling and benchmarks, alongside production modelling and agent systems.
Time seriesFinancial AIProduction ML
BMO Capital MarketsMay — Nov 2025

Machine Learning Quantitative Researcher

Research internship · Mitacs & McMaster University

  • Built synthetic financial-data generation using a regime-switching rough Bergomi model with Hawkes jumps.
  • Designed and benchmarked financially structured forecasting architectures, including the FinAttn research model.
Stochastic volatilityTransformersDerivatives
Vector Institute & McMasterSep 2023 — Dec 2025

Research Assistant

Supervised by Dr. Anastasis Kratsios

  • Researched geometric deep learning, conditional laws of Volterra processes, and CPU-efficient LoRA adapter generation.
  • Parallelized hypernetwork computation across GPU nodes on Vector Institute clusters, achieving a 60× speedup.
Geometric deep learningHypernetworksHPC
Independent trading researchJan 2021 — Sep 2022

Quantitative Researcher & Co-Founder

Researched statistical arbitrage in foreign exchange and gold, combining 24 trading agents with genetic algorithms, fuzzy logic, and Monte Carlo simulation.

03 Selected code

Research you
can look inside.

Public implementations, experiments, and the fundamentals underneath.

All repositories on GitHub
Python

Options pricing
with machine learning

A modular pipeline for European vanilla and worst-off options: synthetic data, feature engineering, model comparison, and evaluation.

Computational financeModel evaluation
Explore repository
Python · PyTorch

LoRA generation
without GPUs

The research implementation behind our ICML workshop paper, with adapter construction, dataset similarity, and evaluation workflows.

Efficient AIOpen research
Explore repository
C++

Fundamental
graph algorithms

Direct implementations of graph algorithms, written to explain the mechanics and support students learning competitive programming.

AlgorithmsTeaching
Explore repository

04 A wider view

Mathematics,
curiosity & craft.

I’m interested in what happens when mathematical structure meets a practical modelling problem: financial processes with memory, interpretable forecasts, or learning with fewer resources.

My background is in mathematics and computational science. Alongside research, I’ve supported graduate teaching in computational finance, risk management, and numerical methods at McMaster.

Away from the keyboard

Usually a good cup of coffee, a walk through the city, a camera, or something by Beethoven.

Education

2023 — 2025

M.Sc. Computational Science
& Engineering

McMaster University

3.9 / 4.0 GPA · Full scholarship

Geometric Deep Learning for Financial Data & Efficient Fine-Tuning of Foundation Models

2018 — 2023

Mathematics & Applications

University of Tehran

Recognition & community

2018

ACM ICPC Bronze Medal
West Asia regional contest

2024

Research presentation
Eastern Conference on Mathematical Finance, Fields Institute

2024

DLRL Summer School
CIFAR & Vector Institute

Let’s connect

Good work starts
with a conversation.

I’m always happy to talk about quantitative research,
financial machine learning, and thoughtful engineering.

Arabporr@gmail.com
LinkedIn GitHub

Copy citation

Automatic copying is unavailable in this browser. Select and copy the text below.