Toggle theme

Sam Ghalayini

MSc Computer Science · Copenhagen, Denmark

Research in multiparty computation and neurosymbolic AI at DTU and ITU Copenhagen.

Groups DTU Compute Section for Cybersecurity Engineering ITU Center for Information Security and Trust
[email protected] | GitHub | LinkedIn
Sam Ghalayini

1. Current Work

Sept 2026–

Research Assistant, Technical University of Denmark

Privacy-Preserving Collaborative Machine Learning

Investigating how multiple parties can train machine learning models without compromising privacy, combining neuro-symbolic learning with secure multi-party computation. Part of Robust NIDS.

Supervisors: Professor Nicola Dragoni, Associate Professor Alessandro Bruni
Funding: DIREC Catalyst Project

Aug 2026–

Assistant Lecturer, IT University of Copenhagen

Course: AI Security · Copenhagen
  • Supporting teaching and hands-on exercises in adversarial machine learning, neural network verification, LLM security, and AI regulation.
  • Guiding students through laboratory work and providing feedback on assignments and projects.

2. Research

Feb–Aug 2026

Satellite Traffic Engineering & Network Optimization

Aalborg University, advised by Yan Kyaw Tun

Formulating multi-class resource allocation optimization problems for satellite networks with varying topology and orbital compute, exploring methods such as decision-focused learning.

3. Education

Feb–Apr 2026

Graduate coursework, University of Copenhagen

Online Reinforcement Learning
Feb–Apr 2026

Graduate coursework, Aalborg University

Reinforcement Learning with Generative AI
2025

MSc Computer Science, University of Colorado Boulder

Graduate Certificate in Artificial Intelligence · GPA 4.0
2021

BA Computer Science, University of Colorado Boulder

4. Projects

Predicting Soil Moisture Dynamics with Physics-Informed Neural Networks

Scientific machine learning · precision agriculture

A PINN predicting continuous 1D soil moisture profiles for precision irrigation, integrating the Richards equation as a physics constraint. Achieved PDE residuals of 1×10−7 and was validated against sensor data from 57 stations across 8 global soil moisture networks (RMSE 0.018 m3/m3). Transfer learning cut training time by more than half while enabling 24–72 hour predictions where traditional numerical solvers failed to converge.

Keywords: PyTorch, Julia, PINNs, Richards equation, transfer learning.

Adversarial Backdoor Injection in LLMs

AI safety · adversarial machine learning

An investigation of backdoor persistence in fine-tuned language models. Implemented a modular injection pipeline on Llama 3.2-3B using LoRA with 4-bit quantization, and an evaluation framework measuring attack success rate, clean accuracy, trigger robustness, and poison efficiency across varying configurations.

Keywords: PyTorch, LoRA, LLM security, Llama 3.2, RunPod.

BitShogi — A Japanese Chess Game Engine

Game engine · bitboard architecture

A complete mini shogi ("the game of generals") engine in Julia using bitboard representations for efficient move generation and board state management. Includes a REST API server and a React/TypeScript frontend playable at bitshogi.com. Featured in the Julia newsletter as a community project.

Keywords: Julia, bitboards, React, TypeScript, Docker.

Urban Management Model for Ride-Sharing

Multi-agent simulation · optimization

A multi-agent, ant-colony-inspired route-finding simulation optimizing ride-sharing routing in complex urban environments. A multithreaded Python simulation models pheromone-based pathfinding that adapts in real time to changing traffic conditions across diverse street network topologies.

Keywords: Python, multi-agent systems, ant colony optimization, concurrency.

5. Volunteering

Jan–Aug 2026

DANSIC – Danish Social Innovation Club

AI ethics & sustainability

Leading AI workshops with DANSIC, a student-driven social innovation non-profit in Copenhagen. Bridging technical and non-technical audiences on topics including local AI deployment, data privacy, and algorithmic bias.

Oct–Dec 2025

Coding Pirates

Tutor

Teaching children programming fundamentals and hands-on microcontroller projects.

2018–2019

Engineers Without Borders

CAD design

Drew CAD plans for a water conveyance bridge in rural Nicaragua, enabling gravity-fed water transport across mountainous terrain to support community access to clean water.