From NVIDIA’s holistic approach to improving the MLOps stack, to Edge Impulse’s aggressively streamlined model training/deployment platform — the door is wide open for developers looking to bring AI at the edge.
Tel Aviv’s ChipEx2021 offered a smörgåsbord of short and sweet technical sessions, including those in “Track E” — sponsored by Edge Impulse — which was dedicated to AI/ML and TinyML. As an engineering student and developer set to start his first job in the world of embedded ML I was excited to experience a cross section of the international effort to bring AI down from the Clouds and nudge it into the extremities of our hardware.
Among the presenters was Dr. Deepu Talla, VP of Embedded & Edge computing at NVIDIA. Rather than presenting a new technology or framework, Dr. Talla emphasized the need for a comprehensive approach to developing the Edge computing stack. He spoke about the importance of all aspects of AI development advancing in parallel, starting from data preprocessing and through to building applications. He suggested that too much emphasis is placed on the moment of inference, and not enough on the stages before and after — which can be the most time consuming bottlenecks to development.
The highlight for me (being a shameless fanboy of Edge Impulse and all they produce) was Amir Sherman’s presentation. Amir, no stranger to silicon, was until recently the GM of Arrow Electronics and now runs his own business development agency, while also serving as Israel’s Country Manager for Edge Impulse.
Amir spoke about the prickly nature of the traditional MLOps pipeline, in which data acquisition, model training, deployment, testing, and model adjustment all occurred in separate environments, posing a tremendous challenge to a Joe Blow developer interested in using AI in his designs. Amir showed how every step of the process could be achieved using Edge Impulse’s free, hardware agnostic platform.
To sum up, I left ChipEx2021 feeling excited to be entering the world of EmbeddedML. It’s clear to me that technology is maturing in such a way that less time will be wasted on Sisyphean nitpicking and more time can be spent doing what developers actually want to do: Make the world better with technology.
