Kamran Razavi completed his disputation in computer science on September 23, 2024, defending his thesis titled "Resource Efficient Inference Serving With SLO Guarantee." His research was conducted under the supervision of Prof. Dr. Lin Wang and Prof. Dr. Max Mühlhäuser; the committee members included Prof. Dr. Matthias Hollick, Prof. Dr. Carsten Binnig, and Prof. Dr. Justus Thies.
His thesis tackles the challenges regarding resource efficiency in inference serving while complying with the end-to-end latency requirements. For this, he explored horizontal-based autoscalers to increase resource efficiency and vertical-based autoscalers to increase responsiveness and accuracy of inference serving systems. Additionally, he leveraged the programmable network devices to enable in-network intrusion detection inference serving.
