Deep learning
Our experts develop and train deep learning models, harnessing the full power of artificial intelligence to analyse data and provide trustworthy results. A recent success story is the development of a CE marked deep learning model that automatically detects diabetic retinopathy on fundus images. This is accompanied by the successful submission of two patent applications, which resulted in the creation of a new spin-off company called MONA.
Another example is our Characterise-to-Sort (CtS) technology, an inline characterisation device that identifies each individual waste particle on a conveyor belt. VITO has combined a variety of sensors within the device and developed a smart software package to analyse the scans. The CtS technology provides insights in the materials that can be found in a waste stream, that can be used to determine the value of the waste and optimise recycling processes.
Reinforcement learning
At the core of reinforcement learning is independent decision making. The learner or agent is put in an environment, where it must learn the optimal behaviour to achieve its goal and to maximise the reward. Contrary to supervised learning, where the learner is fed information, and unsupervised learning, where the learner finds structure in complex data, reinforcement learning is not based on direct examples. The reward signals from the environment guide the learner in choosing the optimal action for every situation.
A classic application for reinforcement learning is optimisation for price, to keep costs low. At VITO, we develop reinforcement learning algorithms for various applications, such as peak shaving in district heating networks, to maximise financial and environmental gain. Reinforcement learning was used in the now completed TEMPO project (Temperature Optimisation for Low Temperature District Heating across Europe), funded by the European Union’s Horizon 2020 Programme for Research and Innovation.
Heuristics and mathematical optimisation
In comparison to descriptive analytics (providing insight in the past and the present) and predictive analytics (statistics models and machine learning to predict the future), heuristics and mathematical optimization, the core of prescriptive analytics, support to decide what you should do in order to achieve business objectives and improve business operations.
With our MooV service (Mobilisation & Optimisation Of Value chains), we provide decision support in supply chain design and optimisation. By combining geodata analytics, network supply chain design and mathematical optimisation, we create a digital twin of a business’ supply chain considering the key constraints and objectives.
In this way, Moov supports to optimise your supply chain for the best performance, simulate the impact of potential changes and critical decisions (e.g. new storage) and experiment with the impact of alternative strategies (e.g. merges) within the virtual supply chain.
Forecasting
As part of a bigger picture, forecasting is all about making accurate predictions and taking the right decisions with limited information and data sets. Data science and AI help dig up useful information to make predictions and to keep on improving those predictions.
When it comes to the application areas of energy and health, VITO pushes the envelope with brand new techniques that can be applied immediately in areas with social relevance. Low TRL, AI research into valorisable and applicable technologies and multidisciplinarity set us apart from the bulk of the research field.
One example is ADriaN (Active Distribution Networks), a long-term bilateral research cooperation between Fluvius, the Flemish Distribution Grid Operator, and VITO/EnergyVille.
Multi-agent systems
Multi-agent systems perform what we humans might call teamwork. These systems comprise of autonomous but connected instances that work towards a common result. It is a case of collaborative AI, performing multi-party computation without the need for central control. One example is the smart battery cell, developed by VITO & EnergyVille.
This innovation allows batteries to monitor the condition of their individual cells and monitor their operation, because some electronics are built in the battery cells as part of the next generation of battery management system and consequently it reduces the electronic footprint in the battery systems. In addition, VITO is involved in a Flanders AI Research project about multi-agent collaborative AI, where information management agents provide intelligent capabilities for personal data stores.