Kaoutar El Maghraoui

Principal Research Staff Member – AI Engineering, IBM Research AI

Kaoutar El Maghraoui Photo

Invited Talk: Hardware-aware Automated AI for Efficient Deep Learning across Hybrid Deployments: Current Landscape and Future Directions

Slides

Video

Abstract:

Automating the decisions required for creating and setting up a deep learning (DL) pipeline has become a key direction in both research and industry to speed up the complex, slow and error-prone process of designing novel DL architectures across many modalities and tasks. Neural architecture search (NAS) has been a growing trend that aims at automating the design of neural networks that are on par or even outperform hand-designed architectures. While the holy grail for evaluating the quality of an DL model in NAS techniques has largely been focused on getting models with the highest possible accuracy, additional metrics become of paramount importance as AI models meet constrained devices (e.g., mobile and AR/VR systems) or emerging specialized hardware accelerators. Such metrics include speed or computation time, power/energy consumption, memory footprint and model size. To address these additional constraints, there is Cambrian explosion of new research directions: designing optimized deep learning hardware such as low-bit quantization and analog accelerators , efficient neural architecture search algorithms for specialized DL accelerators, and hardware (latency, energy) aware neural network architectures search targeted for constrained devices. This talk uncovers the evolving landscape of hardware-aware neural architecture search and its future directions. It also highlights IBM Research suite of techniques towards the design & build of optimized deep learning hardware as part of IBM’s AI hardware Center initiative.

Biography:

Dr. Kaoutar El Maghraoui is a principal research staff member at the IBM T.J Watson Research Center working on innovations at the intersection of systems and artificial intelligence. She leads the AI testbed of IBM Research’s AI Hardware Center, a global research hub focusing on enabling next-generation accelerators and systems for AI workloads. Her current research interests include hardware aware neural architecture search, performance optimization of AI systems, and operationalizing AI workloads in hybrid cloud environments. Kaoutar has co-authored several patents and publications in the areas of AI, distributed systems, and high-performance computing. Kaoutar holds a PhD. from Rensselaer Polytechnic Institute, USA and a master’s degree in Computer Networks from Al Akhawayn University, Morocco. She received several awards including the the American Association of University Women fellowship, the Robert McNaughton Award for best thesis in computer science, IBM’s Eminence and Excellence award for leadership in increasing Women’s presence in STEM, IBM’s award for contributions to the foundational POWER software technologies and promoting these systems in Africa, 2 IBM’s outstanding accomplishment awards for contributions to building cognitive virtual agents, and IBM Research accomplishment for the AI Center Testbed.


FastPath 2021 Program