Join “Machine Learning in IoT” in Technopark Zurich, Building Pauli, Room Cobol, 2nd Floor at 2 pm CET / 6 am PDT.

 

Raamadaas  Krishnadas  

The contouring control problem aims to increase the tracking accuracy while traversing the trajectory as fast as possible.  Model-based control algorithms such as Model Predictive Control have improved the tracking performance of contouring tools.  They are nevertheless subject to the limitations of the simplified linear system models.  The combination of control and machine learning has contributed to significant performance improvement.  Therefore, an online learning algorithm for contouring control problems using Bayesian Linear Regression (BLR) was developed.   The  objective  is  to  learn  online  and  adapt  locally  to  model  mismatches  while  tracking  the geometry.   BLR  is  a  simple  model  whose  major  advantage  compared  to  other  learning  models is that its parameters can be efficiently updated as new data points are collected.  The learning performance of BLR is evaluated for different mismatches in the system’s dynamics, and its ability for fast adaptation is discussed.

Guillaume Neol

I guess we all agree that single use containers don’t really make sense anymore. This is true for the coffee cup that you discard after sipping the last bit of coffee but also for the industrial packaging that is used to transport goods all around. However, the GHG footprint of reusable packaging is pretty high as they are usually made out of plastic compared to cardboard or wood for single use. Therefore, they need to be used continuously for a long period of time to end up having a lower environmental impact. Thanks to the Internet of Things, it is now possible to have cheap enough tracking devices attached to the packaging to ensure supply chain visibility. Problem solved you might think? Not quite, because IoT data isn’t of the best quality. Data bits are lost during communication, location accuracy isn’t always super accurate and the other characteristics of the data sources are extremely heterogeneous. It sounds bad, but nothing AI cannot help with!

Florentin Marty

Autonomous vehicles, intelligent systems, facial and speech recognition and diagnostic medical devices are part of our everyday lives. The underlying algorithms are versatile, powerful but also performance-hungry. IOT devices do not always have access to powerful servers. So how can AI run “at the edge”, i.e. embedded in the end device, without computing data in the cloud? In this talk, Florentin Marty uses a person-counting and a cell-counting AI to show how the concepts were implemented in customer projects.