Mehrshad Zandigohar
Contact
https://www.linkedin.com/in/cmehrshad/
With ESL
- PhD 2018 - Current
- MSc 2018 - 2021
Biography
Mehrshad Zandigohar is a PhD candidate in Computer Engineering at Northeastern University, specializing in high-level hardware-software system design and deep learning technologies. With an educational foundation from both Northeastern University and Sharif University of Technology, Mehrshad has excelled in areas like prosthetic hand robotics, machine learning and deep learning and their acceleration on embedded devices. His significant contributions to research include publications in top-tier journals and conferences, with notable work in multimodal fusion for prosthetic hand control and model accelration for embedded devices. Mehrshad’s industry experience includes a role at Qualcomm Technologies, where he developed new power models for next-generation connectivity chips, showcasing his ability to bridge theoretical machine leaning research with practical, impactful industry solutions.
Contributions
- Grasp-LLaVA:
- Introduced semantic projection as a measure of a model’s ability to generalize grasp types to unseen objects
- A Grasp Vision Language Model that enables human-like reasoning for improved grasp prediction, achieving 49% accuracy on unseen objects (vs. 36.7% for existing models)
- Multimodal Fusion of EMG and Vision for Human Grasp Intent Inference
- HANDSv2 Dataset: Synchronized EMG and vision dataset for grasp types
- Supporting Aging-in-place Through Multimodal Sensing and Reasoning
- CatNet: Ensemble of categorized neural networks to improve model robustness against adversarial attacks.
- NetCut: Pruning problem-specific features of ConvNets to improve measured inference latency.
- Deployable Grasp Type Probability Estimation