Matthew Mattina

Papers

Total Citations

H-Index

About

Matthew Mattina is a leading researcher at the intersection of robotics, computer vision, and efficient deep learning. His early work, notably "Recognizing Human Pose and Actions for Interactive Robots" (2007, 6 citations), pioneered neuro-inspired methods for monocular tracking and action recognition, enabling robots to imitate human movement by combining learned kinematic vocabularies with online trajectory estimation. This foundational contribution advanced human-robot interaction by making pose estimation modular and adaptive. More recently, Mattina has made significant strides in Bayesian deep learning, particularly with his 2021 paper "On the Effects of Quantisation on Model Uncertainty in Bayesian Neural Networks" (3 citations). This work critically examines how model compression techniques like quantization impact uncertainty estimation—a vital consideration for deploying reliable AI in safety-critical applications. By demonstrating that quantization can distort a model’s confidence, his research provides essential guidance for building trustworthy, resource-efficient neural networks. With a career spanning embodied AI and principled uncertainty quantification, Mattina’s work continues to influence how machines perceive, learn, and make decisions under real-world constraints.