A closer look at the problems I work on. Click a project to see what it asks, how I approached it, and what I found.
Zero-Shot Cross-City Generalization in End-to-End Autonomous Driving: Self-Supervised versus Supervised Representations
Does a driving model trained in one city still drive well in another? A controlled zero-shot stress test of how visual pretraining (I-JEPA, DINOv2, MAE vs. supervised ImageNet backbones) affects end-to-end planners under strict geographic splits.
Data-Efficient Adaptation of Self-Supervised Encoders for Driving Perception
Which few labeled images should you pick, and how much of the network should you tune, when adapting a self-supervised encoder to driving perception with limited labels?
TALKtoME: Educational Materials for Speech and Language Acquisition in Autism
A curated visual dataset of real-world images and videos of action verbs and verb–noun pairs, designed to support speech and language acquisition in autistic children and adults.
Mapping Human Grasping to 3-Finger Grippers: A Deep Learning Perspective
Learning where a three-finger robotic gripper should grasp an object from how humans grasp it: a new dataset of human three-finger grasps and a deep model that predicts grasp points from RGB images and finger keypoints. B.Sc. thesis at TaarLab, University of Tehran.