Arnav
Pal

This work tackles detecting adversarial patches — maliciously crafted perturbations engineered to fool deployed vision systems. We propose a detection framework leveraging gradient probing to expose anomalous activation patterns, combined with saliency alignment techniques that surface spatial inconsistencies introduced by adversarial inputs.

Our approach demonstrates strong detection performance across standard benchmarks, advancing the robustness of real-world computer vision deployments.

• —Designed data pipelines for large-scale neuroscience datasets
• —Automated preprocessing, cutting manual overhead significantly
• —Bridged CS and cognitive neuroscience in interdisciplinary team

• —Developed gradient probing to expose adversarial patterns in deep vision models
• —Implemented saliency alignment improving explainability and robustness
• —Co-authored paper on adversarial patch detection

• —Deployed CMS adopted across multiple city departments
• —Gathered stakeholder requirements and iterated features
• —Delivered production-ready software in enterprise environment

I'm a CS student at Western University specializing in ML/AI. Co-author of a CVPR paper on adversarial patch detection.

I work at the intersection of rigorous research and real-world engineering — from publishing at top conferences to shipping apps with 150+ users.

Let's build
something.