Right from the start, Deeplite’s mission has been to make AI deep neural networks faster, smaller and more energy-efficient. We want to enable AI for everyday life on any device and in the cloud.
tinyML® Foundation is focused on bringing together embedded hardware, software, and machine learning practitioners to discuss collaborations across these disciplines. Founded in 2018 by members from ARM, Google and Qualcomm, it has organized the leading annual conference in ultra-low power machine learning in the Bay Area. Today the tinyML Foundation has over 3,300 members worldwide and counting, which makes it one of the fastest growing ML and AI communities across the globe. With backing from the tinyML Foundation, Deeplite has been heading the organizing committee of the tinyML: Enabling low Power ML at the edge Meetup in Montreal. It has been an inspiring experience hosting the meetups, which led us to become a sponsor for the increasingly popular tinyML Talks Webcast Series!
With this webcast series, the Deeplite team hopes to connect the professional community with the industries that can benefit from leveraging the tinyML technology, such as IoT, robotics, semiconductors, automotive, security and surveillance. We want to further stimulate knowledge sharing, networking and research in the exciting deep learning optimization space to accelerate the productization of intelligent devices and AI applications!
The webcast series only began in March this year, but the tinyML community has already had great success among hardware architects, software engineers, systems engineers, ASIC designers, algorithms and application developers, low power sensor providers and other interested professionals. The webcasts include 2 speakers and take place twice a month on Tuesdays at 8 am PT. The presenters are world-leading experts in machine learning technologies at the edge, dedicating their talks to educating the attendees on the latest and greatest in tinyML. To make sure no tinyML enthusiast misses a talk, the foundation is publishing all past sessions on the TinyML YouTube Channel.
Our own co-founder & CTO, Ehsan Saboori, will be holding a talk on November 10th at 8am PDT.
Talk title: Networks within Networks: Novel CNN design space exploration for resource-limited devices
In this talk, Ehsan will introduce how Deeplite approximates the layers within a neural network using a smaller neural network to find highly accurate, highly compact CNN topologies that satisfy strict constraints on model size and acceptable accuracy.
We'll showcase our novel, automated method for compressing convolutional neural networks to maintain maximum accuracy on the target application by utilizing a “network within a network” paradigm. We will also discuss the implications of this approach for tinyML applications and future developments in the field.
Register to attend the session and learn more about ultra-low power and low-latency deep learning models, computing hardware, and systems for inference on edge devices that are designed to preserve model accuracy after extreme model compression!