Master’s Thesis Intern at Forschungszentrum Jülich

Designed and implemented a multimodal lip-reading model using state-space sequence models optimized for long-range temporal dependencies. Leveraged event-based camera data for low-latency, high-temporal-resolution input capturing subtle lip movements. Preprocessed asynchronous event streams into frame-like representations for deep learning model training. Achieved improved accuracy and computational efficiency compared to traditional CNN-RNN based lip reading systems.

Student Arbeit at Universität Siegen

Conducted a comprehensive project on Speech Emotion Recognition (SER) systems. Developed models using Machine Learning and Deep Learning architectures. Achieved state-of-the-art results, particularly with the CNN-BiLSTM model.

Researcher - Human Activity Recognition at Universität Siegen

Conducted data preprocessing and engineered the sliding window method on multivariate time-series data to extract features for Human Activity Recognition classification. Trained and evaluated Hybrid and N-BEATS models using K-Fold cross-validation to improve accuracy over previous CNN-LSTM models.