A team of researchers from Drexel University's Electrical and Computer
Engineering department has been awarded the Best Paper Award at the 21st
ACM International Conference on Computing Frontiers (CF'24), held in
Ischia, Italy, last month. The conference is a leading platform for
showcasing cutting-edge research that pushes the boundaries of computing.
The winning paper, co-authored by PhD student Ilknur Mustafazade;
Naga Kandasamy, PhD
, professor and department head; and
Anup Das, PhD
, associate professor, introduces a new approach to enhance the efficiency
of brain-inspired computing systems, known as neuromorphic hardware. These
systems aim to mimic the human brain's neural structure and offer an
energy-efficient way to run advanced machine learning models.
The challenge lies in optimally mapping these models onto the neuromorphic
hardware to ensure efficient use of resources and minimize communication
between different parts of the system. Previous methods often overlooked
the hardware's architecture when breaking down the models into smaller,
more manageable parts.
Mustafazade, Kandasamy, and Das developed a novel technique that takes the
hardware architecture into account during the process, guaranteeing that
the resulting subnetworks can be effectively mapped onto the system. They
created algorithms for two types of hardware configurations and tested
their approach using both synthetic and real-world machine learning models.
The results show that their architecture-aware method significantly
improves hardware utilization compared to existing techniques. The paper
also explores how the structure of the machine learning model affects the
solution's quality and proposes strategies for further enhancement.
This prestigious award highlights the groundbreaking research being carried
out by Drexel University's Electrical and Computer Engineering department
in the field of brain-inspired computing. As the need for energy-efficient
artificial intelligence grows, innovations like those presented in this
paper will be essential in shaping the future of computing technology.