Research Interests

We specialize in developing and applying both classical machine learning and deep learning algorithms, with a strong emphasis on efficiency—achieving superior speed and prediction accuracies.

We solve practically relevant problems by excellent theory.

Our areas of expertise include:

• Optimization: We develop efficient optimization algorithms with proven guarantees for training various machine learning models on single CPUs, GPUs, and distributed systems (federated learning).
• Understanding Deep Learning Architectures: We conduct in-depth analysis of deep neural networks to understand their core mechanisms, predictions, and functionalities. This includes dissecting various architectures, such as transformers used in large language models, designing new architectures with proven guarantees, and debunking erroneous claims in the deep learning literature.
• Quantum Computing: We explore efficient algorithms for quantum computing, focusing on solving discrete optimization problems like QUBOs.

We share our findings through publications, code, and web services.