Machine Learning Researcher

Area of interests: Machine Learning, Large Language Models, Natural language processing, Text Extraction, Computer Vision

CV

Download my CV

Blog

Blog

Latest

• April 2025: Gave presenation on Evaluation in Retrieval Augmented Generation (RAG) system in NIU AI/ML Seminar
• April 2025: Attended and Presenting Poster on Midwest Speech and Langugage Day (MSLD) 2025 at University of Notre Dame, Indiana
• March 2025: Succesfully defended PhD Qualifying.
• March 2025: Succesfully defended Masters thesis
• February 2025: Gave presenation on Large Language Models in NIU AI/ML Seminar
• December 2024: Got 3MT best paper presentation award in 2024 ACM/IEEE Joint Conference on Digital Libraries (JCDL), 2024, Hong Kong, titled “Limitation Generation of Reseresarch Paper using LLM and RAG with Text Evaluation framework”.
• October 2024: Our work “Mitigating Visual Limitations of Research Papers” has been accepted in 2024 IEEE International Conference on Big Data (IEEE BigData 2024) as a poster.
• September 2024: Our work “LimTopic: LLM based Topic Modeling and Text Summarization for Analyzing Scientific Articles Limitations” accepted in 2024 ACM/IEEE Joint Conference on Digital Libraries (JCDL), 2024, Hong Kong.
• August 2024: Our work “Limitation Generation of Research Papers with LLms and Retrieval Augmented Generation (RAG)” has been accepted in 11th IEEE International Conference on Data Science and Advanced Analytics (DSAA 2024), San Diego, CA as a poster.

Education

• PhD in Computer Science, Northern Illinois University, Dekalb, United States (August 2023 - Present)
• MS in Computer Science, Northern Illinois University, Dekalb, United States (August 2023 - Present)
• BSc in Computer Science, Sylhet Engineering College, Sylhet, Bangladesh (March 2016 - March 2021)

Research Interests:

• Information Extraction (text and images) in Science of Science, utilizing Large Language Models (LLMs) and Retrieval Augmented Generation (RAG).
• Utilizing BERTopic, LLM with zero shot, few shots, chain/tree/graph of thoughts, self-consistency, prompt tuning, prefix tuning, fine-tuning with LoRa/QLora/Dora.
• Jailbreaking, Evaluation, and Hallucination of LLM and Tiny LLm.
• Optimizing LLM performance: time complexity 0(n) with Mamba (state space models), Flash Attention, KV cache, xLSTM, Deja Vu, Sparse LLM, Galore/Q-Galore, and scalable matrix multiplication.
• Open Source LLM: Llama, OLMO, Ollama, Dolma, Paloma.
• Vision: LLM (Llava), Diffusion model, CLIP, BLIP,GAN, Variational Autoencoder, Vision Transformer, Swithc Transformer, a Mixture of Experts, and multi-modal LLM.

Published

• I. A. Azher, S. Ahmed, M. S. Islam, and A. Khatun, ”Identifying Author in Bengali Literature by Bi‑LSTM with Attention Mechanism,” 2021 24th International Conference on Computer and Information Technology (ICCIT), Dhaka, Bangladesh, 2021, pp. 1‑6, doi: 10.1109/ICCIT54785.2021.9689840. Description: Proposed a Bi‑LSTM model with a self‑attention mechanism by Glove embedding to identify the Author. Used Deep Learning models like CNN, RNN, LSTM, GRU, BiLSTM, and Fast text and showed proposed model outperforms.
• I. A. Azher, V. Reddy, A. P. Akella, H. Alhoori, “LimTopic: LLM-based Topic Modeling and Text Summarization for Analyzing Scientific Articles Limitations”, 2024 ACM/IEEE Joint Conference on Digital Libraries (JCDL), 2024, Hong Kong. We extracted the scientific article’s limitations sections using a parsing tool. Here, each topic contains the title and ‘Topic Summary.’ This study focuses on effectively extracting and understanding these limitations through topic modeling and text summarization, utilizing the capabilities of LLMs. We extracted limitations from research articles and applied an LLM-based topic modeling integrated with the BERtopic approach to generate a title for each topic and ‘Topic Sentences.’ To enhance comprehension and accessibility, we employed LLM-based text summarization to create concise and generalizable summaries for each topic’s Topic Sentences and produce a ‘Topic Summary.’ Our experimentation involved prompt engineering, fine-tuning LLM and BERTopic, and integrating BERTopic with LLM to generate topics, titles, and a summary.
• I. A. Azher, H. Alhoori, “Generating Suggestive Limitations from Research Articles Using LLM and Graph-Based Approach”. 11th IEEE International Conference on Data Science and Advanced Analytics (DSAA 2024). We are generating ‘Limitation’ based on other section texts such as ‘Abstract,’ ‘Introduction,’ ‘Methodology,’ ‘Related Work,’ ‘Experiment’, and ’Conclusion.’ We experimented with various LLMs such as BART, T5, Pegasus, GPT 3.5, GPT 4, Gemini, and added Retrieval Augmented Generation (RAG) with GPT 3.5 to generate the ‘limitation’ of research papers. We found that GPT 3.5 with RAG performs better than other models. We will extend our work by incorporating graph neural networks to generate limitations.
• I. A. Azher, H. Alhoori, “Mitigating Visual Limitations of Research Papers” has been accepted at 2024 IEEE International Conference on Big Data (IEEE BigData 2024). In this work, we focus on visual-related limitations, which refer to issues like the clarity of charts, diagrams, unclear captions or descriptions, or methodological constraints in generating such visual and tabular data using multimodal LLM.
• Christy Muasher-Kerwin, M Courtney Hughes, Michelle Foster, Ibrahim Al Azher, Hamed Alhoori. ``Exploring Large Language Models for Summarizing and Interpreting an Online Brain Tumor Support Forum.” Sage Jounral.

In Progress

• “Scientific Image as a Limitation.” We are working on images of scientific articles and using algorithms like superpixels or other stuff to improve the quality.
• “Can LLM predict citations ?” Take the scientific article abstract and other information and predict the citations using LLM.
• “Software vulnerability detection with LLM.” Take the code from various sources and check whether the code is vulnerable or not.

Projects

• Medical datasets: MIMIC IV. Predicting mortality, hospital length of stay MIMIC

• Amazon Review Analysis: Using data preprocessing, various machine learning and deep learning models are applied to classify the customer review for the product. ARA
• Image classification: Using CNN, Graph ConvNet, Inception Network, CNN, LeNet, AlexNet, GoogleNet, VGG16, and VGG19. Image

Experience

Research Assistant, Northern Illinois University, Dekalb, IL (May 2024 - Continue):

• Extracting texts and images from scientific articles using large language models. Incorporating LLM techniques such as zero-shot, few-shot, chain of thoughts, self-consistency, and fine-tuning.
• LLM is used for code vulnerability detection and image analysis with large vision language models, superpixels, diffusion models, vision transformers, etc.

Teaching Assistant, Northern Illinois University, Dekalb, IL (Aug 2023 - May 2024)

• C++, Data Structures, and Algorithm Analysis
• Conducted weekly lab sessions for over 50 students, facilitating hands-on exercises in Data structures (BFS, DFS, Linked list) and Algorithms.

Software Engineer @Bysl Global Technology Group, Dhaka, Bangladesh (June 2021 ‑ February 2022)

• Works on scalable projects from design to coding, testing, and installation using Python, XML, Javascript, PostgreSQL, and jQuery in Linux OS. Designed and developed various ERP‑based business software and modules like sales, inventory, purchase, website, and e‑commerce using Python language in Odoo

Junior Software Engineer, @Unisoft Systems Limited, Dhaka, Bangladesh (December 2020 ‑ May 2021)

• Research, assess, and lead the initiation of new technologies to maximize performance. Interact with clients, analyze requirements, and develop an application in Python. I have written various SQL to generate numerous reports and created various modules using the OOP Database model

Skills:

• Programming Languages: C, C++, Python, Javascript, Java, PHP
• Framework and Technologies: Django, Laravel, HTML, CSS, Bootstrap, Odoo, Django, Django REST
• Open Source Library: Keras, Tensorflow, Pytorch, numpy, pandas, sci-kit learn, matplotlib,
• Database: Oracle SQL, MySql, Postgresql
• (Development tools): Git, Github, CodeBlocks, Pycharm, Visual Studio, Anaconda, Jupyter Notebook, Latex, Overleaf
• Operating system: Linux OS, Windows

Honors & Awards

• Google GCP Grant Award, Google — Feb 2024-Feb 2025

More Project

• Exam marks management (PHP): A website project that allows teachers to enter the grades from various examinations and class quizzes, which are then automatically converted with GPA. Students can only see their grades. EMM
• Newspaper (Python, Django): Developed a fully web-based project using Python, Django, and Javascript, allowing anyone to read the news with a monthly subscription. Use a monthly payment subscription with Stripe. Newspaper
• MyShop: Sell and purchase items from a shop. Using HTML, CSS, Python, and Django. MyShop

Machine Learning:

• Experienced in Data Analysis, Linear Algebra, Probability, and Statistics behind Machine Learning Models (Linear Regression, Logistic, SVM, DT, SVM, NB). Also, Feature Importance, Dimensionality Reduction, Weight Initialization, Batch Normalization, Backpropagation, Activation functions (SGD, ADAM, NAG, Adagrad, Adadelta, RMSProp)

Programming

ACM Problem Solving: Over 350 problems were solved by various online judges. Over 200 problems from Codeforces, 100 problems from UVa Online Judge

Seminar

I am thrilled to get a chance to do a presentation in a seminar organized by the NIU CS department. I tried to give a brief overview of Large Language Models (LLMs), such as

1. Zero/few-shot learning, 2. Reasoning, Chain/Tree/Graph of thoughts, RLHF, RLAIF 3. Fine-tuning, Prompt tuning, 4. Flash Attention, 5. LLM as a Judge. I also talked about the current limitations of LLM, how RAG (Retrieval Augmented Generation) overcomes those problems, and what types of issues still exist in RAG. For example, RAG retrieves noisy information, which may not align with the LLM generator.

full video on youtube: YouTube seminar link: Seminar

Conferences