7. 7
SWOT
Strengths
● We accelerate DCNN-based algorithms while maintaining
low power consumptions thanks to FPGAs (this w.r.t.
desktop GPUs which are widely used for inference)
● Scalability improvement by using a multi-PYNQ system
● Multi-job handling on embedded systems by with FARD
8. 8
SWOT
Weaknesses
● Since we deal with a distributed system we have to
handle data transfer between multiple nodes
● System failure handling on multiple nodes
● The design has an steeper learning curve with respect to
already existing solutions
9. 9
SWOT
Opportunities
● Given the growth of machine learning an optimization of
embedded systems is of interest to many companies
● The scalability of the system and its modularity allow
application on multiple scenarios
● Thanks to the multi-job system multiple users can use the
same architecture
10. 10
SWOT
Threats
● Research in machine learning is so extended that there is
a huge number of competitors
● Nowadays users tend to prioritize economic solutions
instead of the effective and modular solution we propose
11. 11
Thanks for the attention
Giorgia Fiscaletti <giorgia.fiscaletti@mail.polimi.it>
Marco Speziali <marco.speziali@mail.polimi.it>
Luca Stornaiuolo <luca.stornaiuolo@mail.polimi.it>
www.facebook.com/pynoliatnecst @PyNOLIatNECST