To battle these severe cyber-physical risks, Dr. Capraru’s engineering research concentrates heavily on defense resilience. His works, such as "Overcoming catastrophic forgetting in radar and LiDAR object detection in rain via layer freezing and data augmentation," offer real-time neural network patches. By restricting weight adaptation to specific core sensor-processing layers and applying smart synthetic data augmentation, machines can sustain their perception accuracy across vastly different climates without needing completely fresh datasets. Seminal Publications and Contributions
[Insert information about impact, legacy, and any relevant recognition] richard capraru
When neural networks are optimized to handle clear weather, retraining them to recognize objects in rain often causes them to "forget" how to operate safely in sunshine. Dr. Capraru solved this paradox through advanced training methodologies utilizing: To battle these severe cyber-physical risks, Dr
Below is a blog post draft tailored to his professional focus. To battle these severe cyber-physical risks
Learning from Failures: LLM-Guided Safety-Critical Scenario Recommendation for Self-Evolving Autonomous Driving Robustness of 3D Deep Learning in an Adversarial Setting
Richard Capraru's research in this field has been conducted alongside academic peers such as Emil Lupu, Jian-Gang Wang, and Boon Hee Soong.
[Physical Adverse Weather (Rain)] + [Adversarial Spoofing (10-20 Points)] │ ▼ [Traditional ML Models Misled / Blinded] │ (Dr. Capraru's Countermeasures) ▼ [Robust Perceptual Defenses & Anti-Forgetting ML] 1. Unmasking LiDAR Vulnerabilities in Adverse Weather