Decebal Constantin Mocanu

Eindhoven University of Technology, Electro Optical Systems (Germany)

Papers

Total Citations

H-Index

About

Decebal Constantin Mocanu is a leading researcher in machine learning, with a primary focus on developing efficient, scalable, and biologically-plausible neural network architectures. His major contributions lie in advancing **sparse training** methods—techniques that enable deep neural networks to learn and maintain sparse connectivity from the outset, dramatically reducing computational and memory costs without sacrificing accuracy. This work has reshaped how researchers approach model efficiency, particularly in resource-constrained environments. Mocanu’s most cited paper, "Factored four way conditional restricted Boltzmann machines for activity recognition" (2015, 47 citations), introduced a novel probabilistic model for capturing complex temporal dependencies in sensor data, a key step toward robust human activity recognition. His earlier work on "Inexpensive user tracking using Boltzmann Machines" (2014, 9 citations) tackled the challenging problem of low-cost, probabilistic user localization. More recently, his research on "Automatic Noise Filtering with Dynamic Sparse Training in Deep Reinforcement Learning" (2023) extends sparse training to reinforcement learning, enabling robots to autonomously filter irrelevant information—a critical capability for real-world deployment. Mocanu’s contributions are foundational to the emerging field of efficient deep learning, with his methods influencing both academic research and practical AI systems.