ACCELERATING DEEP DATA ARCHITECTURES USING FOG AND EDGE COMPUTING

Businesses across all industries are having a hard time to effectively manage Big Data. Many of them do not have the know-how to extract relevant insights from the mass of information. Deep data is based on the premise that a fraction of essential knowledge can either carry the entire business requirements at a lower cost in money and time or enrich the existing Big Data with advanced capabilities and performances.
This presentation is about accelerating AI via generating Deep data using Edge and Fog Computing strategies.
The proposed strategies are applied with different acceleration techniques to improve speed in training and inference for data enrichment from the top level (algorithm) to low level (Processors) in Natural Language Processing and Object Recognition applications.
At the functional and algorithmic level, the presentation will introduce and differentiate various methods and models, namely Transformers vs Reformers in NLP and Inception vs. MobileNet in Object Recognition.
At the implementation and compiler level, the presentation will focus the currently available generations of AI environments, in particular the new coming JAX compared to TensorFlow, or TensorFlow Lite vs Tensor RT.
At the infrastructure level, the presentation will position the main paradigms available in the market, namely positioning the various processing architectures: CPU vs GPU vs TPU.
We then introduce Mind in a Box, an innovative, proprietary AI Edge and Fog Computing Integrated Solution which helps businesses achieve cost-effective performances for turn key, production ready applications.
Finally, we illustrate examples of industry use cases that are benefiting from using Mind in a Box in in implementing these different Deep Data strategies to streamline and transform data into real-time actionable information.

Jérémie Farret – Inmind Technologies coordinator
Dipl. Eng. National Institute of Applied Sciences (Lyon, France)
20 years of experience as CTO and Director of RSDE at SGDL Systems, Logik3D, Parallel Geometry Inc and Inmind Technologies since 2016
Relevant Experience :
High Performance Computing / GPGPU: Various publications (See below), functional architect and patent author for the Mind in a Box suite of products, integrated systems for Artificial Intelligence processing and acceleration (Listed in Invest AI 2020 quebec providers catalog)
Language processing: 17 years of project coordination based on Scheme / Lambda calculus programming, Natural Language implementations for solid modeling, planar and spatial geometry teaching (Schene Project, 2008), 3D printing (P4BUS, 2015) and smart contracts (Creativité Québec Program 2017/2018).
Supply Chain: Project manager for CNRC PRECARN USE (automate assembly chains from symbols and sketches, 2005), simulation and optimization of NBC sensors provisioning (Nizza airport, Thales, EADS, 2009), Centre Hospitalier Sud Francilien (Automated provisioning of 4000 hospital rooms for daily service attributions 2012), JADE (Jewelry mass customization online and additive manufacturing chain, 2015).
AI/ML: CRNC PRECARN USE Project Symbol Recognition software integration 2005, CNRC IRAP project coordination (Tensor Flow, TRAX, TPU and GPU implementation and optimization of NLP and object recognition approaches, Splunk Machine Learning Toolkit cybersecurity project coordination at Bell Solutions Techniques.