Hello! I am a PhD student in the Image and Video Computing (IVC) Group at Boston University, advised by Professor Kate Saenko and Professor Bryan A. Plummer. I study the gap between training metrics and what we actually want models to learn—whether it shows up as spurious correlations, shallow reasoning, or training procedures that help in one context and hurt in another. I like diagnostic work: tracing a problem to its source and figuring out what to do about it.

I've recently been collaborating with NASA on geospatial foundation models—check out our work below.

Selected Publications

Spurious Correlations in Vision-Language Models

• Maan Qraitem, Piotr Teterwak, Kate Saenko, Bryan A. Plummer. Web Artifact Attacks Disrupt Vision Language Models. ICCV 2025. [PDF] [Code]
• Maan Qraitem, Nazia Tasnim, Piotr Teterwak, Kate Saenko, Bryan A. Plummer. Vision-LLMs Can Fool Themselves with Self-Generated Typographic Attacks. NeurIPS 2024 Workshop. [PDF] [Code]
• Maan Qraitem, Kate Saenko, Bryan A. Plummer. From Fake to Real: Pretraining on Balanced Synthetic Images to Prevent Spurious Correlations. ECCV 2024. (Oral) [PDF] [Code]
• Maan Qraitem, Kate Saenko, Bryan A. Plummer. Bias Mimicking: A Simple Sampling Approach for Bias Mitigation. CVPR 2023. [PDF] [Code]

Geospatial & Earth Observation AI

• Paulo A Arevalo, Maan Qraitem, Sujit Roy. Detecting Key Land Surface Phenometrics with the Prithvi Foundation Model and Harmonized Landsat Sentinel-2 Data. AGU 2025. [Code]
• Prithvi-EO: An Open-Access Geospatial Foundation Model Advancing Earth Science through Global Collaboration. AGU 2025. [40+ authors; contributed model adaptation pipelines for phenology forecasting and mining detection.]

Evaluation & Reasoning in Multimodal Models

• Eunice Yiu, Maan Qraitem, Charlie Wong, Anisa Noor Majhi, Yutong Bai, Shiry Ginosar, Alison Gopnik, Kate Saenko. KiVA: Kid-inspired Visual Analogies for Testing Large Multimodal Models. ICLR 2025. [PDF] [Code]
• Maan Qraitem, Kate Saenko, Bryan A. Plummer. Breaking the Assistant Mold: Modeling Behavioral Variation in LLM Based Procedural Character Generation. In Submission. [PDF] [Code]

Physics Informed Deep Learning

• Maan Qraitem, Dhanushka Kularatne, Eric Forgoston and M Ani Hsieh Bridging the gap: Machine learning to resolve improperly modeled dynamics. Phsycia D Journal 2019. [PDF]
• Bruce A Maxwell, Casey A Smith, Maan Qraitem, Ross Messing, Spencer Whitt, Nicolas Thien, Richard M Friedhoff Real-time physics-based removal of shadows and shading from road surfaces. CVPR Workshop 2019. [PDF]