Nov. 26th, 2020
The process of training a neural network consists (roughly speaking) of presenting an artificial intelligence (AI) system with a large amount of sample data to enable it to make preliminary general conclusions about objects of interest. For example, if we want to teach an AI system to recognize human faces, we do not provide a geometric pattern for the oval of the face, but we show the system thousands of photos with markings such as: “this part of the photo is a face”, “these two faces are similar, and the others are not” or “this is an eye.” This method produces excellent results, but its use in practice has certain limitations. First, to train an AI, we need to collect thousands of photos representing the objects of interest. For highly specialized applications, such large sets of data may prove extremely difficult to collect. In addition, we need a huge amount of labour to make accurate photo markings that show the AI system clearly what the objects are. This makes training AIs long and costly, as the human factor is an indispensable part of the process.
Sky Engine has developed innovative methods to train neural networks quickly and with greater accuracy, changing the current paradigm and eliminating the above-mentioned limitations. The methods developed by Sky Engine AI combine into one coherent system the deep learning module and the rendering module that produces the images needed for learning (image data are generated automatically and also automatically tagged). By combining graphics technologies to create images based on their 3D digital models with the use of graphics processing units (GPUs) and advanced methods of adapting this image data for deep learning, it has become possible to use them for training neural networks. Thanks to this, the system trains AIs extremely fast and accurately.
Lately, Sky Engine concluded an investment agreement with the Krakow-based INNOventure fund, which in the past has already made investments in companies developing technologies related to image analysis and artificial intelligence. “At INNOventure, we value and support ground-breaking technological projects – just take a look at our investment portfolio. We particularly like those whose practical usefulness and market potential have already been positively verified. In the case of Sky Engine, serious implementations such as Scania AB speak for themselves. So we are sure that the project will be successful” – comments Marcin Bielówka, managing partner in the INNOventure fund.
More information about Sky Engine can be found in our portfolio entry.