Our paper in the AI4Sys '24 conference at HPDC 2024 presents a novel technique using Generative AI (GenAI) to automate on-the-fly customizations of AI/ML solutions. The ECO-LLM system dynamically adjusts task placement between edge and cloud computing, resulting in minimal performance differences while significantly reducing manual effort and time in solving systems problems.
Our paper presents CLAP, a dynamic solution for optimizing microservice placement across edge and cloud computing in real-time applications. It addresses workload-induced latency issues and cost efficiency by utilizing Reinforcement Learning. Experiments on video analytics demonstrate significant cost reductions of 47% and 58% while maintaining acceptable latency levels.
Our paper, presented at PDP 2024, discusses a containerized distributed processing platform for Autonomous Surface Vehicles to enhance real-time data processing in marine environments. Utilizing microservice management with DataX and Kubernetes, it addresses challenges such as limited connectivity and energy constraints. Experiments demonstrate its effectiveness in marine litter detection.
Our paper, presented at POAT 2023 in Singapore, examines joint microservice scaling in Kubernetes, focusing on video analytics pipelines. It introduces DataX AutoScaleUp, which efficiently adjusts CPU resources while Horizontal Pod Autoscaler (HPA) operates. This method significantly enhances processing rates, achieving up to 1.45X improvement over traditional approaches.
Our Application in a Box (AnB) project, presented at SMARTCOMP 2023, simplifies the deployment of remote 5G applications. AnB includes pre-configured hardware and software, allowing quick setup without extensive technical knowledge. It features automated resource management for optimized performance, demonstrating real-world applications and significantly reducing deployment time from months to minutes.
Our Elixir is a system that enhances video analytics performance from multiple IoT cameras by adjusting camera settings through Multi-Objective Reinforcement Learning (MORL). In tests, Elixir significantly outperformed default settings and time-sharing methods, improving detection rates for cars, faces, persons, and license plates, addressing challenges in multi-analytical unit settings.
Giuseppe Coviello 