Ankit Gupta
PhD Student
Department of Computer Science & Engineering
Michigan State University
PhD Student
Department of Computer Science & Engineering
Michigan State University
Hi there!
I am a PhD student in Computer Science & Engineering at Michigan State University.
My research focuses on AI safety in semantic search, agentic retrieval, and multimodal AI systems. Across my work on embedding models, retrieval-augmented generation (RAG), and vision-language models, I study where embedding similarity diverges from meaning and what that implies for robustness and security.
Earlier, as an undergraduate at the Indian Institute of Technology Kharagpur (IIT-KGP), I built an agent-based simulation platform synthesizing complex emergent behaviors from simple neural wiring with a multi-agent system of Braitenberg Vehicles. Alongside this, I had the opportunity to pursue research internships at the Graduate Institute of Mind, Brain, & Consciousness (where I later came back to do my MS in Cognitive Neuroscience), the Indian Institute of Science (IISc), and Harvard University. In addition, I also served as an open-source student developer with Google Summer of Code at Red Hen Lab. Taken together, these experiences provided early exposure to experimental research, computational modeling, and studying emergent behavior in intelligent systems.
For more about my profile, please check my CV. Also, here's my personal blog for more about my journey and some of my poetries!
Nov 2026 — My solo-authored paper NONSENSE was accepted to the EACL 2027 main conference in Athens, Greece! It shows how meaningless strings can outrank relevant documents in semantic search.
Oct 2026 — New preprint SPOOK shows that nearly blank images can outrank real images in text-to-image search with CLIP and image-to-image search with DINOv3.
Oct 2026 — New preprint STROKE shows that strokes covering at most 2% of an image fooled ResNet-50 in 98.4% of ImageNet tests, 102 times faster than NeurIPS 2025 baseline.
Jul 2026 — Our ALIFE 2026 work on AI and life detection drew wide media coverage, including stories in MSU Today, Popular Science, Mashable Science, and Space.com.
Jun 2026 — Our paper Can AI Detect Life? was accepted for an oral presentation at ALIFE 2026 in Waterloo, Canada. We tested whether AI could identify self-replicating digital sequences.
Dec 2025 — New preprint SPOOF shows how sparse edits to a blank image can make a classifier confidently predict an object that isn’t there.