Natural Language Processing
Mechanistic interpretability and reasoning in language models — multihop biomedical reasoning (PathwayLM) and the robustness/security of LLM tools (impeding LLM-assisted cheating).
Saiful Islam Salim
Hello! I'm Saiful Islam Salim, a PhD candidate in Computer Science at the University of Arizona, advised by Prof. Mihai Surdeanu. I work at the intersection of natural language processing, machine learning, and security. Through my research, I aim to build language models we can trust: models whose reasoning can be traced and whose misuse can be anticipated and blunted.
Before Arizona, I completed my bachelor's and master's in Computer Science and Engineering at BUET, advised by Prof. A. B. M. Alim Al Islam. There I built low-cost sensing systems and traffic simulation tools for real-world problems in developing countries.
Outside research, I love traveling and playing cricket.
Degrees
Education
- Fall 2021–Present
Ph.D. in Computer Science
The University of Arizona · Arizona, USA
Focus: Natural Language Processing · Advisor: Dr. Mihai Surdeanu, Professor
- November 2018–August 2021
M.Sc. in Computer Science and Engineering
Bangladesh University of Engineering and Technology (BUET) · Dhaka, Bangladesh
Advisor: Dr. A. B. M. Alim Al Islam, Professor · Thesis: "Devising A Selective Sensor Deployment Strategy Leveraging NSGA-II for Detecting Derailments in the Railway System of Bangladesh"
- July 2014–October 2018
B.Sc. in Computer Science and Engineering
Bangladesh University of Engineering and Technology (BUET) · Dhaka, Bangladesh
Advisor: Dr. A. B. M. Alim Al Islam, Professor · Thesis: "Enhancing Fidelity of Quantum Cryptography using Maximally Entangled Qubits"
Appointments
Experience
- Fall 2021 – Present
Research Assistant, Department of Computer Science
The University of Arizona
Research Assistant terms: Fall 2022, Fall 2023.
- 2021 – Present
Teaching Assistant, Department of Computer Science
The University of Arizona
See Teaching page.
- November 2018 – July 2021
Research Assistant at NeC Lab, Dept. of CSE
Bangladesh University of Engineering and Technology
PI: Dr. A. B. M. Alim Al Islam, Professor.
What I work on
Research
Machine Learning
Applied ML for safety and low-resource settings, including adversarial perturbations that degrade LLM-assisted cheating while remaining human-legible.
Computer Networks & Security
Low-cost, long-range networking and sensing for the developing world — rail-track monitoring, structural vibration, dust/noise pollution, road traffic, and post-disaster rescue.
Optimization
Multi-objective optimization (NSGA-II) for selective sensor deployment and multi-sensor fusion in rail-track systems.
Publications
Featured work
-
PathwayLM: Multihop Mechanistic Pathways for Biomedical Language Model Reasoning
ACL (Under Review) · 2026 Under review
-
A cost-effective real-time rail track monitoring system leveraging multi-sensor fusion and multi-objective optimization
Scientific Reports, Nature — Q1, IF 3.9 · 2025
-
Impeding LLM-assisted Cheating in Introductory Programming Assignments via Adversarial Perturbation
EMNLP 2024, Miami · 2024 Oral · Best Paper Nominee
-
Long-Range Low-Cost Networking for Real-Time Monitoring of Rail Tracks in Developing Countries
ACM ICTD · 2022