Our aim for the project Organic Machine Learning (OML) is to break the rigid approach of training Machine Learning Systems, as well as to develop a method for machine learning that mimics organic, espescially human learning. With such a method, systems would train/learn their whole life span, especially during their usage.
The way machines learn is supposed to be more organic. Instead of training with big, clear and good structured trainingdata, which is needed to be elaboratley prepared, OML's soon to be developed system will aim to train with heterogenic, less or not at all structured data, which is found in the environment around it and will rely on less training data. – similar to humans. Different sources – such as interactions with humans and the systems own experiences, will be multimodally combined and the training will be more focussed on cases of uncertainty. In order for that to be achieved the systems need to be able to recognise, in which cases they are uncertain or when further training is needed. In addition to that are learning systems not supposed to be a „black box“ anymore, in which it is hard to get insight to their function from the outside. Rather they should be able to explain how they arrived at a conlcusion and why they acted in certain ways. Through the ability of justification the decisions of the systems will be more accteptable to humans and their usage for many applications in the real world will be made possible.
At last the developed system will be merged with a robotic overall system. In a scenario of interactiv robotic programming the robot will be able to learn from the ground up new skills, through physical, visual and verbal interaction with the human, as well as through their own experiences - much like the apprentice learns from their master.