Advanced Signal Processing Across Neural, Audio, and Biomedical Domains
The Signal Modelling and Analysis Laboratory (SIGMA Lab) , founded by lecturer Ana Neacșu is now part of the CAMPUS Research Institute, Politehnica Bucharest. The laboratory originates from SpeeD group, the first speech research group within the Faculty of Electronics, Telecommunications and Information Technology. The laboratory focuses on advancing foundational and applied signal processing across domains such as neural networks, audio, and biomedical systems.
Our research on neural network robustness develops optimization methods, adversarial training strategies, and regularization techniques to improve the safety, interpretability, and resilience of AI systems. In the field of audio analysis, we enhance feature extraction and spatial rendering through deep learning, signal processing algorithms, and Head-Related Transfer Function (HRTF) measurements. The biomedical analysis stream focuses on extracting meaningful insights from physiological signals, combining machine learning with artifact removal and working closely with clinicians on applications in mental health, diagnostics, and rehabilitation.
SIGMA Lab actively partners with academia and industry to ensure that its research leads to impactful, real-world applications.
Robustness of Neural Networks
Neural networks, while achieving exceptional performance across various tasks, are notably susceptible to adversarial manipulations. Small, intentional alterations to input data can lead to incorrect model predictions, posing risks in critical applications like autonomous vehicles and healthcare. At SIGMA Lab, we are dedicated to enhancing the robustness of neural networks during the training phase. Our research focuses on developing advanced optimization methods, adversarial training strategies, and regularization techniques. By fortifying models against adversarial attacks, we aim to improve the safety, interpretability, and overall resilience of AI systems, ensuring their real-world applicability.
Audio analysis
Our work in audio analysis seeks to push the boundaries of sound quality and spatial audio rendering. We concentrate on advanced denoising techniques, audio source separation, and upmixing, which are crucial for both consumer and professional audio environments. SIGMA Lab has developed successful upmixing and audio source separation models that have already found applications in the industry. By leveraging deep learning and innovative signal processing algorithms, our methods enhance audio clarity and immersive experiences, paving the way for next-generation audio technology that significantly improves user experience.
Biomedical analysis
In the realm of biomedical analysis, we focus on Automatic EEG analysis, particularly in the automatic diagnosis of neurological diseases such as epilepsy, autism, and ADHD. Our research integrates machine learning with artifact removal techniques to extract meaningful insights from complex physiological signals. Currently, we are collaborating with Victor Gomoiu Hospital to develop a tool that aids clinicians in diagnosing and monitoring these conditions more effectively. By working closely with medical professionals, SIGMA Lab strives to create innovative solutions that improve patient outcomes and advance diagnostic methodologies in healthcare.