Neuroscience and Machine Learning

Sara Rostami

I study how the human brain makes sense of what we see, and I build AI systems that are interpretable, calibrated and worth trusting.

MSc Neuroscience · Specialization in Machine Learning in Health and Biomedical Sciences Western University · Vector Institute · London, ON

One foundation in computer engineering, two branches. Pick one to tailor the page, or keep both.

About

Two paths, one question

How does intelligence compute? I come at it from both sides: studying it in the brain, and building it in machines.

For academia

Neuroscience & NeuroAI research

I'm a Master's student in Neuroscience at Western University, in the Specialization in Machine Learning in Health and Biomedical Sciences. I work in the Mohsenzadeh Lab, a Vector Institute–affiliated lab, supervised by Dr. Yalda Mohsenzadeh and Dr. Emma Duerden. My thesis uses simultaneous EEG–fNIRS to study how the brain perceives images and what makes an image memorable, combining the timing of electrical activity with the slower hemodynamic response.

Before Western, I completed an MSc in AI & Robotics at the University of Tehran with Dr. Mohammadreza Abolghasemi Dehaqani. My thesis used an RSVP paradigm and 128-channel EEG to separate repetition and expectation effects in the visual mismatch negativity, within the predictive-coding framework.

Computational neuroscience
NeuroAI
Vision science
Interpretable AI

Academic CV Research program

For industry

Applied & trustworthy AI

I'm an ML researcher with a computer-engineering foundation and 4+ years of building end-to-end pipelines on high-dimensional, noisy real-world data. My focus is making model outputs trustworthy, through interpretability, uncertainty quantification and calibration, across deep networks in PyTorch and LLM-based systems.

Recently I've been building working with foundation models. I'm looking for Applied AI Research Scientist and Data Scientist roles in Toronto / GTA.

LLMs & agents
RAG
Calibration & uncertainty
Explainable AI
PyTorch
Python · SQL

Résumé Applied AI

Academia

Research program

Why are we exceptionally good at some tasks yet struggle with others? Can understanding neural mechanisms help us design better artificial systems?

Perception & image memorability

Simultaneous EEG and fNIRS to capture both the timing and the hemodynamics of how we perceive images, and of why some images are remembered while others are forgotten.

MSc thesis · Mohsenzadeh Lab · 2025–

Prediction & expectation in vision

How repetition and violated expectations shape the brain's automatic change detection (vMMN), framed by predictive coding.

MSc thesis · University of Tehran · 2022–2024

NeuroAI

Neuro-inspired, interpretable deep networks, designed in PyTorch and benchmarked against what we measure in the brain.

Ongoing · Mohsenzadeh Lab

Methods & tools

EEG (32 & 128 channel)
fNIRS
ERP analysis
MVPA / decoding
RSA
Permutation testing
Predictive coding
EEGLAB
FieldTrip
MNE-Python
PsychoPy
MATLAB
Python
PyTorch

2025–Mohsenzadeh LabWestern University & Vector Institute
2022–25Convergent Technologies Research CenterUniversity of Tehran
2022–23Genzel LabDonders Institute, Radboud University

Industry

Applied AI, built to be trusted

Foundation models are only as useful as our ability to know when to trust them. That layer is where I focus.

The stack I work across

ApplicationsClinical triage · VQA · image memorability · analytics
Agents & orchestrationLangChain · LangGraph · RAG · FastAPI
Trust layer · my focusCalibration · uncertainty · explainability · evaluation
Foundation modelsVLMs (BLIP-2) · LLMs (Gemini, OpenAI) · BERT / BEiT
Data & infrastructurePyTorch · Hugging Face · PostgreSQL · pandas · Git

Where it applies

Projects

Selected work

Click any image to open the project. Some have interactive figures.

Experience

Where I've worked

Recognition

Awards, teaching & talks

Awards

Teaching & talks

Contact

Let's talk

Open to Applied AI Research Scientist and Data Scientist roles in Toronto / GTA, and to research collaborations in NeuroAI and trustworthy machine learning.