Accessibility AI
Developing LLM-based navigation systems for individuals with visual impairments
Aydin Ayanzadeh
Ph.D. Student in Computer Science
University of Maryland, Baltimore County
About Me
I am a Ph.D. student in Computer Science at UMBC (M.S. 2025). I work on computer vision, medical image analysis, and multimodal learning — with a focus on AI that is useful in clinics, the field, and assistive technology.
That includes navigation systems for people with blindness or low vision, vision-language models for wildfire detection, and medical image segmentation under real data constraints.
Research Interests
Perception & models
- Computer Vision
- Medical Image Analysis
- Multimodal Learning
- Large Language Models
- Vision-Language Models
People & planet
- Accessibility Technology
- Environmental AI
Methods
- Knowledge Distillation
Current Focus
Environmental AI
Creating Vision-Language Models for early wildfire detection and environmental monitoring
Medical AI
Advancing deep learning techniques for medical image analysis and diagnosis
Education
May 2025 - Present
Ph.D. in Computer Science
University of Maryland, Baltimore County
GPA: 3.78/4.00
Courses: Advanced Algorithms, Knowledge Graphs, Machine Learning, Computer Vision, Data Privacy
Feb 2022 - May 2025
M.S. in Computer Science
University of Maryland, Baltimore County
GPA: 3.78/4.00
Courses: Advanced Algorithms, Knowledge Graphs, Machine Learning, Computer Vision, Data Privacy
Sept 2018 - Sept 2020
M.Sc. in Applied Informatics
Istanbul Technical University, Turkey
GPA: 3.75/4.00
Courses: Image Processing, Applied Informatics in Structural Biology, Fuzzy Logic
Sept 2011 - Apr 2016
B.Sc. in Computer Science
University of Tabriz, Iran
GPA: 3.11/4.00
Publications
Accepted
Characterizing an On-Device Depth-Aware Multimodal Workload in Assistive Navigation for Blind and Low-Vision Users
Accepted - IISWC 2026
Accepted
LLM-Guided Agentic Floor Plan Parsing for Accessible Indoor Navigation of Blind and Low-Vision People
Accepted - CVPR 2026
Accepted
WildfireVLM: AI-Powered Analysis for Early Wildfire Detection Using Satellite Imagery
Accepted - IGARSS 2026
Accepted
Floorplan2Guide: LLM-Guided Floorplan Parsing for BLV Indoor Navigation
IEEE Big Data 2025, arXiv preprint arXiv:2512.12177
Featured
PURSUhInT: In Search of Informative Hint Points Based on Layer Clustering for Knowledge Distillation
Expert Systems with Applications, vol. 213, 119040, 2023
Improved cell segmentation using deep learning in label-free optical microscopy images
Turkish Journal of Electrical Engineering and Computer Sciences, vol. 29, no. 8, pp. 2855-2868, 2021
Representation learning using graph autoencoders with residual connections
arXiv preprint arXiv:2105.00695, 2021
Deep Learning based Segmentation Pipeline for Label-Free Phase-Contrast Microscopy Images
2020 28th Signal Processing and Communications Applications Conference (SIU), pp. 1-4, IEEE, 2020
Graph Embedding For Link Prediction Using Residual Variational Graph Autoencoders
2020 28th Signal Processing and Communications Applications Conference (SIU), pp. 1-4, IEEE, 2020
A New Class of Scaling Matrices for Scaled Trust Region Algorithms
arXiv preprint arXiv:1904.09209, 2019
Cell Segmentation of 2D Phase-Contrast Microscopy Images with Deep Learning Method
2019 Medical Technologies Congress (TIPTEKNO), pp. 1-4, IEEE, 2019
Automated Segmentation of Cells in Phase Contrast Optical Microscopy Time Series Images
2019 Medical Technologies Congress (TIPTEKNO), pp. 1-4, IEEE, 2019
A Modified Ant colony Based Approach to Digital Image Edge Detection
2015 2nd International Conference on Knowledge-Based Engineering and Innovation (KBEI), pp. 726-729, IEEE, 2015
Gaussian Three-Dimensional kernel SVM for Edge Detection Applications
International Conference on New Research Findings in Electrical Engineering and Computer Science, Tehran, 2015
Book Chapter
Automated analysis of phase-contrast optical microscopy time-lapse images: application to wound healing and cell motility assays of breast cancer
In Diagnostic Biomedical Signal and Image Processing Applications with Deep Learning Methods, pp. 137-154, Academic Press, 2023
Journal Reviewer
AI & Machine Learning Journals
Expert Systems with Applications (Elsevier)
Jan 2024
Neurocomputing (Elsevier)
Nov 2024
Scientific Reports (Nature)
May 2024
Multimedia Systems (Springer Nature)
May 2024
Computer Vision & Signal Processing
International Journal of Machine Learning & Cybernetics
(Springer Nature)
June 2024
Signal Processing: Image Communication (Elsevier)
June 2024
For a complete list of publications and citations, please visit my Google Scholar profile.
Research Experience
Research roles spanning accessible AI, medical imaging, computer vision, and efficient machine learning.
-
CORAL LAB
Dec 2022 - PresentResearcher — Advisor: Tim Oates
Baltimore, MD
Role notes at CORAL LAB
- Developing a multi-task network for segmenting and classifying medical images on imbalanced datasets
-
SP4CING Lab
Sept 2018 - Dec 2021Research Assistant — Co-Advisors: Behcet Ugur Toreyin and Devrim Unay
Istanbul, Turkey
Role notes at SP4CING Lab
- Introduced an auto-encoder with a modified ResNet-18 encoder
-
Vodafone FutureLab
May 2019 - Jan 2022Research Fellowship — Supervisor: Mehmet Basaran
Istanbul, Turkey
Role notes at Vodafone FutureLab
- Introduced Res-VGAE, a variational graph auto-encoder with residual connections
- Proposed dynamic Word2Vec to examine social structure of Vodafone customers
-
Tubitak 1001 (Grant #119E578)
Oct 2020 - Jan 2022Research Assistant — Supervisor: Devrim Unay
Izmir, Turkey
Role notes on Tubitak 1001
Project: Image processing and machine learning tools for phase-contrast optical microscopy time series
- Analyzed cell morphology and movement in phase-contrast optical microscopy time series
- Designed an auto-encoder that sped up training, improved tracking robustness, and reduced overfitting on segmentation and tracking
-
Arcelik Global Co.
Dec 2019 - Dec 2020Researcher — Supervisor: Nazim Kemal Ure
Istanbul, Turkey
Role notes at Arcelik Global Co.
Project: Optimization of Multi-Task Network on Surveillance Cameras
- Quantized a multi-task model for edge devices, raising inference from 2 fps to 18 fps
Project: Model Compression for Efficient Video Processing on Edge Devices
- Built a multi-task people-monitoring network for surveillance cameras on edge devices
- Compressed the models with quantization, using Intel OpenVINO
-
SiMiT Lab
Apr 2017 - June 2018Researcher (Unpaid) — Supervisor: Hazim Kemal Ekenel
Istanbul, Turkey
Role notes at SiMiT Lab
Project: Kaggle Dog Breeds Identification with transfer learning
- Fine-tuned ImageNet-pretrained models for the Kaggle dog-breed identification dataset
Project: Google Cloud YouTube-8M Video Understanding Challenge
- Examined deep neural networks with skip connections for the Kaggle video understanding challenge
Teaching & Mentoring
Graduate Teaching Assistant at UMBC since Spring 2022 — systems, AI, data science, scripting, and intro programming.
-
University of Maryland, Baltimore County
Spring 2022 – PresentGraduate Teaching Assistant
Baltimore, MD
-
CMSC 313
Computer Organization and Assembly Language Programming Fall 2024, Spring 2025, Spring 2024, Spring 2022
-
CMSC 104
Problem Solving & Programming Summer 2024
-
CMSC 433
Scripting Languages Summer 2024
-
CMSC 671
Principles of Artificial Intelligence Fall 2023
-
CMSC 304
Social & Ethical Issues in IT Summer 2023
-
CMSC 691
Introduction to Data Science Fall 2022, Spring 2023
-
CMSC 341
Data Structures Summer 2022
Role notes
- Supported undergraduate and graduate sections across systems, AI, data science, scripting, ethics, and intro programming
- Ran labs and discussion support, held office hours, and graded assignments and exams
-
CMSC 313
Mentoring
- Undergraduate research in computer vision and machine learning
- Technical guidance on programming projects and assignments
- Career advising for research paths and graduate school
- Project supervision for capstones and independent study in AI and vision
News & Updates
15
May
2025
Achievement
M.S. Graduation & International Alumni Scholarship
Jan
2025
Conference
Presenting at CMD-IT/ACM Richard Tapia Conference
15
Dec
2024
Research
New Paper: Improved Autoencoder for Nuclei Segmentation
Oct
2024
Conference
Presented at STARS Celebration Conference
2024
Project
WildfireVLM: Environmental AI Research
2023
Publication
PURSUhInT Published in Expert Systems with Applications
May
2019
Award
Vodafone FutureLab Research Fellowship
Research Projects
Current directions and selected published work.
Current Research
Active
LLM-Based Indoor Navigation System for Individuals with Blindness or Low Vision
Developing an innovative indoor navigation system using Large Language Models to assist individuals with visual impairments. The system combines computer vision, natural language processing, and spatial reasoning to provide real-time navigation guidance and environmental awareness.
Large Language Models
Computer Vision
Accessibility
NLP
Collaborator: Prof. Tim Oates (UMBC)
View Project
Active
WildfireVLM: AI-Powered Analysis for Early Wildfire Detection and Risk Assessment
Developing a Vision-Language Model system for early wildfire detection using satellite imagery. The project focuses on creating an AI-powered solution that can analyze environmental data and provide early warning systems for wildfire prevention and management.
Vision-Language Models
Satellite Imagery
Environmental AI
Risk Assessment
Collaborators: Prakhar Dixit, Sadia Kamal, Prof. Milton Halem
(UMBC)
Accepted at: IEEE International Geoscience and Remote Sensing
Symposium (IGARSS) 2026
View Project
Completed Research
PURSUhInT: Knowledge Distillation for Efficient Deep Learning
Developed a novel approach for knowledge distillation by identifying informative hint points based on layer clustering. This method significantly improves the efficiency of knowledge transfer from large teacher models to smaller student models.
Knowledge Distillation
Deep Learning
Model Compression
Clustering
Published in: Expert Systems with Applications (Impact Factor: 8.5)
View Project
Deep Learning for Medical Image Segmentation
Developed advanced deep learning pipelines for cell segmentation in label-free optical microscopy images. The work includes both traditional CNN approaches and novel architectures for improved accuracy in medical imaging applications.
Deep Learning
Medical Imaging
Computer Vision
PyTorch
Published in: Turkish Journal of Electrical Engineering and
Computer Sciences, IEEE SIU 2020
View Project
Graph Autoencoders for Representation Learning
Investigated residual connections in graph autoencoders for improved representation learning on graph-structured data. The work contributes to the field of graph neural networks and unsupervised learning.
Graph Neural Networks
Autoencoders
Representation Learning
Unsupervised Learning
Published in: arXiv preprint, IEEE SIU 2020
View Project
Selected Projects
Technical implementations and research software, presented as a compact browsable collection.
Deep Learning
Medical Image Segmentation
Multi-task deep learning platform for automated medical image analysis with advanced handling of imbalanced datasets.
Read MoreLLM App
LLM Indoor Navigation System
AI-powered navigation system using GPT-4 and computer vision to assist visually impaired individuals in indoor environments.
Read MoreOptimization
Knowledge Distillation Framework
Novel hint-based knowledge distillation framework achieving 2.5x model compression with minimal accuracy loss.
Read MoreMultimodal AI
Vision-Language Medical Models
Custom CLIP-based architecture for medical imaging with zero-shot classification and report generation capabilities.
Read MoreGraph Learning
Graph Autoencoder Framework
Advanced GNN implementation with residual connections for improved representation learning on graph-structured data.
Read MoreMLOps
ML Data Pipeline System
Scalable data processing pipeline handling 10TB+ daily with automated preprocessing and feature engineering.
Read MoreEnvironmental AI
WildfireVLM
Vision-Language Model for early wildfire detection and risk assessment from satellite imagery, accepted at IGARSS 2026.
Read MoreDeveloper Tools
Claude Code Encyclopedia
Searchable reference for Claude Code, Codex CLI and GitHub CLI — commands, flags, settings and config paths in one page, with OS-aware examples.
Read MoreAssistive Tooling
ADHD Study Pack
A local-first focus tool for study sessions — one visible task, step breakdown, an adjustable focus block, and one keystroke to park a distracting thought.
Read MoreTechnical Skills
Languages
Frameworks & Libraries
Tools & Platforms
Awards & Honors
International Alumni Endowed Scholarship
University of Maryland, Baltimore County (UMBC), USA
May 2025Prestigious scholarship recognizing outstanding academic achievement and international student contributions to the UMBC community.
Research Fellowship
Vodafone FutureLab, Turkey
May 2019Competitive fellowship for advanced research in telecommunications and technology innovation.
Top 1% Ranking
Nationwide Universities Entrance Exam, Iran
Sept 2011Achieved top 1% ranking among 500,000+ students in the highly competitive national university entrance examination.
Highlighted Certificates
- AI for Medicine (3-course specialization) - deeplearning.ai on Coursera
- Deep Learning (5-course specialization) - deeplearning.ai on Coursera
- Image and Video Processing - Duke University on Coursera
Contact Me
Interested in research collaborations, academic opportunities, or discussing innovative ideas in AI and Computer Vision? I'd be happy to connect.