On 13 March 2019 the international IPEC - Integrated Plant Engineering Conference 2019 will take place under the topic "edge analytics" at the IHK Academy in Nuremberg. Our expert, Stefano Signoriello, will give a lecture on "Edge Analytics through Artificial Neural Networks" at 10:05 am. For the exchange of ideas and experiences you can visit our team of experts at the accompanying exhibition.
Abstract of the lecture „Edge Analytics through Artificial Neural Networks“:
In the talk we are considering the problem of learning nonlinear low dimensional latent representations for high dimensional data through unsupervised training of artificial neural networks. Here, the principal idea is to build a nonlinear encoder as well as a nonlinear decoder, both modelled by artificial neural networks, where the encoder model maps high dimensional data to points of a low dimensional space and the decoder model maps these points back to high dimensional data space. The model obtained by chaining encoder and decoder together, called an autoencoder, is then trained to reproduce input data by minimizing reconstruction error. Ideally, the learned encoder produces low dimensional but information rich latent features that can be used for further analysis like classification or anomaly detection. High but meaningful data specific compression can greatly reduce the need for storage, network bandwidth as well as computing power for further data processing. In this sense, edge analytics can really benefit from learning good latent representations.