our services
IDEATION
We are offering advisory services to explore and formalize your opportunities that can be found in high performing and explainable AI. The approach is based on the principles of service design, providing an in-depth understanding of the problem and a context of the solution.
Problem understanding
In-depth understanding of the problem and empathy for the individuals affected by it . The problem definition is based on whom, what and why the problem should be solved.
zero-based design
Starting from a blank page we generate and explore ideas that address the problems. The solution landscape is initially formed witout any constraint in mind and forms the basis for the vision. The hypothesis and solution are set in the context of the value-chain and overall processing.
MVP SCOPE AND GOALS
Existing best practices and latest AI research within the specific area are reviewed to understand current possibilities and constraints. The scope for the first minimum viable product (MVP) is defined as a subset of the overall vision. Goals, targets and metrics are defined that reflects the different benefits that the solution should provide such as speed, accuracy and transparency. With this in place the solution hypothesis have been formalized and lays the foundation for requirements to the machine learning modelling phase.
MACHINE LEARNING MODELLING
Our core belief lies in rapid prototyping. It is not about perfecting each stage in the cycle the first time, but rather to learn fast and prove if the ideas are worthwhile to pursue. Alternatively there might be a need to validate existing solutions via an end-2-end due diligence.
Data acquisition
Data is acquired through inhouse and/or external sourcing. The data type/size along with runtime performance and integration requirements dictate where storage, training and inference are performed. The range spans from local servers to distributed cloud and edge computing respectively.
Data exploration
Data is explored to get an quantitative as well as qualitative understanding. Visualization is key to explore distributions, correlations, datatypes, missing data and noise. Graph analytics can be applied to reveal relationsships between entities in the data.
data preparation
Based on the data exploration, the data is cleaned and processed. In order to achieve high performing solutions it is vital to embed domain knowledge at this stage. Another key aspect is to ensure that training and validation data is representative, using stratificaiton and hypothesis testing. Feature engineering is followed by feature selection using filter or wrapper methods.
machine learning modelling
A number of relevant machine learning models are identified, implemented/configured and fitted to the training data. Each model has it's hyper parameters tuned and are evaluated based on the specific target metrics. The models can be further analysed and improved using an explainability layer: see the XAI dashboard
deploy & improve
Deployment of machine learning models is one of the more challenging steps and there are many aspects to consider. The solution needs to be continually monitored in production to understand how the model performs on new, unseen data and any retraining needs to be carefully planned.
MODEL VALIDATION
& EXPLAINABILITY
Model validation helps to ensure that your trained models are performing in a robust, predictable way. Explainability (XAI) is a way to help make the model predictions more transparent and understandable based on the perspective of a human domain expert.
Validation & explainability
XAI supports various stakeholders to understand predictive models through a comprehensive set of aggregation and visualizations. The inner workings of the models are presented on a global, regional as well as local level. See our use case for how XAI could be applied.