From colorizing old photos to becoming more efficient with Deep Learning

The technology key in improving the present future

The ability to synthesize sensory data, while preserving the desired statistical properties, is currently proving to be a great success in different industries.

Many examples are based on this concept and apply it to various industries using technology. One of the most prominent is the case of DeOldify, an artificial intelligence program that translates black and white images into color, or Nvidia, with its proposal to create realistic images of fake landscapes or non-existent people’s faces, from semantic sketches.

GANs: The most interesting idea in Machine Learning of the last decade

These systems are based on a very specific neural network architecture, called Generative Adversarial Network (GAN), an artificial intelligence algorithm based on the fact that synthesized data must maintain both statistical properties and be indistinguishable from real data, a process similar to a Touring test for data.

Such concepts have their origins in the past, where the comparison was made through simple visual inspection. Nowadays, classification models, called discriminators, are used to distinguish between synthesized data and real data. In a more intuitive way, this network can be understood as two competing networks: the first one is in charge of generating candidates, synthesized data, while the second one evaluates whether the data is real or synthesized. The goal of training is to increase the error rate of the discriminating network, i.e. to deceive the discriminator.

Monitoring, deep learning and its business benefits

The solutions described above are adapted to various sectors such as “Smart”, “Industry 4.0” and “Energy” among others. Real-time asset monitoring software is starting to use these technological advances to solve common problems, such as, for example, connection failures. It often happens that some sensors sending data are partially disconnected, which could have been avoided if a generator-discriminator model had entered the game, replacing the missing data with synthetic data. From here we could consider the possibility of replacing some sensors completely with synthetic parts, which would ensure the highest possible quality and reduce the hardware infrastructure required for effective monitoring.

Currently, Spanish companies such as CIC Consulting Informático, with its asset monitoring product IDboxRT Inteligencia Operacional, consider Deep Learning as a tool to make their customers’ lives easier.

Example of predictive visualization of the value that each variable will have in the next 15-minute period.
Example of predictive visualization of the value that each variable will have in the next 15-minute period.

The series of measures in the field of Deep Learning in monitoring, developed by CIC Consulting Informático, leads to significant positive results at several levels. First of all, it provides favorable economic conditions, allowing savings in the operation and maintenance of specific equipment, while avoiding serious losses of information. For this reason, there are advantages associated with energy efficiency, such as the reduction in energy consumption resulting from the reduction in the number of physical components.

Back to the future with deep learning

Deep Learning is expected to have a revolutionary effect on the way companies operate in the near future, making them more efficient in terms of consumption and profitability, optimizing all their processes and achieving tangible results on a global scale.