Muhammad Arbab Arshad

I am a Ph.D. candidate in Computer Science at Iowa State University, working as a Research Assistant at the AI Institute for Resilient Agriculture (AIIRA). Currently, I am a Graduate Software Engineering Intern at Intel, applying Computer Vision and Large Language Models to UI automation, graphics validation, and hardware enablement. My work interests lie at the intersection of Large Language Models, Computer Vision, and MLOps.

Email  /  Scholar  /  Github

🏆 Latest Highlights

Projects & Publications

I focus on developing AI systems that have real-world impact. My projects aim to bridge the gap between cutting-edge research and practical applications in agriculture and beyond.

• Proposed Assisted Few-Shot Learning approach to enhance VLMs for image classification with limited agricultural data.
• Optimized few-shot example selection using encoders (ViT, ResNet-50, CLIP) and cosine similarity.
• Improved performance in 6/7 agricultural stress phenotyping tasks, increasing avg. F1 score from 68.68% to 80.45% (ViT).

• Developed AgEval, a 12-task benchmark for evaluating Vision Language Models (VLMs) in agricultural plant stress phenotyping.
• Assessed zero-shot and few-shot (up to 8-shot) in-context learning performance of models like Claude, GPT, Gemini, and LLaVA.
• Established baselines, analyzed VLM adaptability, and achieved up to 73.37% F1 score with minimal examples.

• Proposed deepmufl, a novel mutation-based technique for fault localization in DNNs, outperforming prior static and dynamic methods.
• Contributed to evaluation by executing automated repair tools on GPU clusters, optimizing parallel execution (16x speedup across 40 clusters).
• Deepmufl localized 53/109 real-world bugs to the top-1 ranked layer in evaluation.

Developed AgLLM, a system providing precise, regionally-appropriate agricultural recommendations for 90 pest species across North America, Africa, and India. Integrates deep learning vision models (InsectNet/WeedNet) with a species-specific, region-aware Retrieval Augmented Generation (RAG) pipeline using expert-verified data, significantly improving information relevance.

Core recommendation engine in use at: Pest-ID

As an intern, contributed to this cloud-based data & analytics platform by:

• Implementing auto-scaling AWS Fargate deployment and developing an end-to-end stress testing pipeline (JMeter/Blazemeter).
• Customizing the GitLab CI/CD pipeline to integrate testing seamlessly.
• Received recognition in two consecutive sprints for implementing comprehensive load testing frameworks that ensured application stability under high-stress conditions.

• Evaluated Neural Radiance Field (NeRF) techniques for 3D plant reconstruction in varied environments (indoor to field).
• Achieved 74.6% F1 score in challenging outdoor field conditions compared to LiDAR ground truth.
• Developed an early stopping optimization technique reducing NeRF training time by ~50% with minimal accuracy loss (7.4% F1 reduction).

Experience

Applying Computer Vision and Large Language Models to UI automation, graphics validation, and hardware enablement.

Developing AI applications (LLMs, CV, NeRFs) for agricultural challenges as part of doctoral research.

Contributed to cloud deployment, CI/CD enhancement, and load testing for the Data Refinery Platform.

Contributed to research on mutation-based fault localization for Deep Neural Networks (ASE '23 Distinguished Paper).

Feel free to steal this website's source code. Do not scrape the HTML from this page itself, as it includes analytics tags that you do not want on your own website — use the github code instead. Also, consider using Leonid Keselman's Jekyll fork of this page.