"Transfer Learning will be the next driver of Machine Learning commercial success"
Andrew Ng, Stanford University and Head of AI at Baidu
The power of Deep Learning….
Deep Learning is the most powerful modelling technique available today, and can:
Reliably model non-linear behaviors:
in process industry the quality of a product depends in a complex way from the machine parameters and the interaction among them. Within AI,, Deep Learning solutions can model complex systems with unmatched performance.
Integrate all data available, not only aggregates:
building a model that can integrate process all the important data available is crucial to reach maximal predictive performance, in particular in complex processes where even small deviations can impact the quality of a product.
Model time-series with unmatched performance:
modelling the behavior of a process over time is fundamental for predicting quality. Deep Learning is the most powerful existing solution to model and predict time series behavior.
...On real life datasets
Transfer learning enables the training of models able to support variations in:
Types of machines.
Often in manufacturing industry several machines perform the same or very similar process. Transfer Learning allows the use of data coming from different machines to improve the predictive performance of the model.
Types of product.
The introduction of a new product can drastically decrease the prediction performance of an AI models. Transfer Learning techniques ensure the model remains performant even during the ramp-up of a new product.
Parameters setting for a given machine.
In many process industries machine parameters are often updated to take into account machines drift. Transfer Learning techniques allow us to build predictive models that support machine drifts and resetting of parameters.