Technology
Nowadays Artificial Intelligence (AI) is everywhere. Our cars are detecting traffic signs, we talk naturally to Siri, Amazon tells us what products we will buy next and we get beaten in almost every game.
„AI“ is maybe the most popular buzzword. But what does it actually mean? The span of use is extensive. From simple systems whose "intelligence" consists of executing processes in a predefined sequence to autonomous vehicles or robots. In science, this is generally understood to mean a system that learns correlations or derives patterns based on training data provided. These learned rules can then be used for new predictions. So unlike as our human intelligence, where we can learn all those skills with “one algorithm” - our brain, all the before mentioned applications are fundamental different.
Every AI is different
What does this mean? Depending on the application there are different requirements that need to be fulfilled in order to give the algorithm the chance to learn properly and to give meaningful results.
Let’s take a system for traffic sign detection as an example. When you feed an image into this system the first steps are processing the pixel data and transforming it into smaller features. This happens in the so called Convolutional Layers of a Neural Network. Based on the existence and relations of those features the decision what object it is can be made.
If, on the other hand, one wants to have an algorithm that predicts the future course of a stock, so-called Recurrent Neural Networks are used, for example. In these networks, not only the individual layers are interconnected, but also individual neurons of different layers. Thus, the historical sequence of the data also has an influence on the learned algorithm.
How can AI be trained?
To train AI modules, the following steps are usually performed. The first step is the selection of the algorithm category. Depending on whether something is to be classified, numerical values predicted or entire sequences of actions determined, the set of available algorithms must be narrowed down.
Then, based on the algorithm category and the use case itself, relevant training data can be collected and processed. Next, training parameters are set and the algorithm is trained. Finally, the performance of the algorithm will be measured on a subset of the available data.
The sequence from the selection & compilation of the training data to the evaluation of the trained algorithm is a very iterative process.
Once a sufficiently accurate algorithm has been trained, it can be exported to the target system. There it is fed with new data and the predicted results can be used.
The challenge
Smartphones, and apps in particular, lend themselves very well to making recommendations to us humans during the day. We always have them with us, and thanks to the computing power now available in these devices, even complex algorithms can be run on them.
However, one question remains unanswered. How can we make these algorithms and apps so individual that the recommendations made are always best suited to each individual? If you use a generic approach as described in the previous section, you lose the advantage of individuality. On the other hand, it is also not possible to manually adjust the intelligence module for each individual.
Our solution - the individual Artificial Intelligence (iAI) Module
Our solution is the "individual AI" module. In this module different scenarios of the described pipeline can be executed and evaluated. This means that for each user the best composition and preparation of the training data, the training parameters, the algorithm and the parameterization of the algorithm can be created. Thanks to an intelligent selection of these compilations, they can be evaluated quasi in real time. This allows us to give each user the best individual recommendations! Your habits change over time? You suddenly have more data available, for example through a smartwatch? No problem! Each time you run the iAI module, the best parameters are determined from the front.
The best one? Both this parameterization of the iAI module and the continuous training based on the personal data is done exclusively on the smartphone. No data leaves the device. Besides the best possible result quality, we can also promise the best possible security of your personal data! A big concern of us!