27 August - Meeting on AI, ML, and Computer Vision

Description

Join our virtual meeting to hear presentations from experts on cutting-edge topics related to AI, ML, and computer vision. Date, time, and location: 27 August 2026, 9:00 AM - 11:00 AM PST Online. **[Register on Zoom!](https://voxel51.com/events/ai-ml-and-computer-vision-meetup-august-27-2026)** **Robust Protection of Concepts Against Image Editing and Diffusion-Based Personalization** Diffusion-based image editing and personalization models have significantly simplified the manipulation and replication of visual concepts based on just a few reference images. However, existing protection methods are often tailored to a single attack model and fail to generalize to a variety of editing processes. In this presentation, I will discuss recent advancements in concept protection for generative AI systems, focusing on strategies for intentional disruption and style-sensitive diffusion representations. I will also present experimental results in various editing and adaptation scenarios, highlighting challenges related to robustness, transferability, and invisibility in practical protection settings. Finally, I will discuss open problems and future directions towards trusted ownership of generative content. *About the speaker* [Qiuyu Tang](https://www.linkedin.com/in/qiuyutang/) is a Ph.D. student in Computer Science and Engineering at Lehigh University. Her research focuses on trustworthy AI, media forensics, and robust protection methods against image editing and diffusion-based personalization systems. Her recent work investigates concept protection, style safeguarding, semantic image manipulation, and generative AI robustness. She has contributed to numerous publications in computer vision and AI security, including research on diffusion model protection and manipulation detection, and has organised workshops at conferences. **From Pixels to Planet: Building Scalable and Fact-Based AI for Science** AI has exhibited many new capabilities, from writing emails to editing and generating images. The effectiveness of AI models is primarily based on a standard machine learning process where data is fed into models to obtain representations and predictions, and performance is evaluated using controlled benchmarks and metrics. However, discrepancies arise when we try to translate this process into real-world interaction and leverage AI for scientific discovery. Beyond decisions in a closed set, scientists wish to explore new categories and propose new hypotheses. In this presentation, I will share how I address AI challenges for science from the perspective of data-centric methods and interpretability approaches. *About the speaker* [Jianyang Gu](https://www.linkedin.com/in/jianyang-gu-9235151b9/) is a Ph.D. researcher at Ohio State University. His research focuses on leveraging data-centric methods to construct scalable and interpretable foundational models for science. **Beyond the Barn: Non-Invasive Detection of Acidosis in Dairy Cattle using Multimodal Gas Emission Intelligence** Rumen acidosis quietly costs the global dairy industry billions each year and threatens animal welfare, yet existing detection methods remain invasive, delayed, and impractical at scale. Our lab has developed a completely new method: recording and analyzing patterns of CO₂ and CH₄ gas emissions through synchronized RGB-thermal imaging, turning every breath into a diagnostic signal. We developed DualGasNet, a deep learning architecture with dual streams featuring attention fusion, which detects acidosis in a non-invasive and real-time manner, achieving industry-leading accuracy on a purpose-built gas emission dataset from cattle. Aiming for explainable, farm-ready AI, we integrate vision and language models—CLIP and LLaVA-1.5—enabling zero-shot diagnostic reasoning that bridges the gap between deep learning predictions and actionable veterinary insights. This presentation will walk through the entire process from creating a custom dataset to multimodal fusion and VLM-assisted interpretation, offering the audience an intriguing case study on using computer vision to tackle high-impact real-world problems beyond traditional benchmarks. *About the speaker* [Taminul Islam](https://www.linkedin.com/in/taminul-islam/) is a Ph.D. candidate at Southern Illinois University Carbondale, with over 40 publications, more than 740 citations, and an h-index of 16 - with papers in CVPR 2026, WACV 2026 (oral), ICCV 2025, and Nature Scientific Reports, including a highly cited publication for 2024–25. **Building Real-World Computer Vision Systems with Voxel51** This presentation will explore the practical workflows associated with building, evaluating, and improving modern computer vision systems. We will address real-world approaches to data curation, model analysis, multimodal AI workflows, and production-ready vision pipelines using open-source technology. The session is designed for engineers, researchers, and AI practitioners who want to understand better how teams are developing and scaling computer vision applications today. Expect hands-on demonstrations, technical insights, and discussions on the evolving ecosystem of AI tools. *About the speaker* [Daniel Gural](https://www.linkedin.com/in/daniel-gural/) is an expert in Physical AI and has been working in this field for over 8 years. Working in healthcare, he has experience in both operational applications and using Visual AI as support in psychological applications.