News

Talk by Dr.-Ing. Arya Mazaheri

ZITI is very happy to welcome Dr.-Ing. Arya Mazaheri for a visit on April 29 2024.
Dr.  Mazaheri will give a public talk at 4 pm in the ZITI lecture room (INF 350, room U014). 

Title: Full-stack AI: Towards efficient AI for next-level autonomous systems

Abstract: AI-based applications are revolutionizing the way we approach prediction tasks by leveraging deep neural networks, including recently emerged Transformers and large language models. However, the challenge of efficiently training and deploying these complex models—especially on smaller edge devices with limited processing power—remains significant. Addressing this, various methods have been developed to improve different aspects of model design and deployment, including accuracy, speed, energy efficiency, cost, and adaptable performance. On the other hand, full-stack AI takes a holistic approach, optimizing the entire AI development cycle to balance all these design concerns effectively. In my talk, I will demystify the layers of full-stack AI development and discuss each component in detail. I'll also share the results of my research, showcasing the practical improvements we can achieve in AI application performance.


Speaker: Dr. Arya Mazaheri is a senior researcher at TU Darmstadt and co-founder of the spin-off PanocularAI, working on the computer systems side of AI. He received his Ph.D. in Computer Science from TU Darmstadt in 2021 on the topic of performance engineering of data-intensive applications.  His research expertise covers HPC performance engineering and deep learning, with a special focus on developing multitask models that automatically achieve high task accuracy, low computational cost/energy, and strong robustness. He has published more than 12 research papers in prestigious HPC and AI/ML conferences. Since 2022, he is actively transferring his research results on accelerated AI into a spin-off, which has been awarded several times by hessian.AI and Germany Trade and Invest.