MAPPE

The project (MAPPE) explores how to design and evaluate interpretable and data-intensive AI solutions for human action and performance in complex physical environments. The project features two interesting and distinctly different real-world scenarios: smart office environments and outdoor target practice exercises.

Introduction

In smart products and services (SPS), the keyword smart implies adaptiveness or intelligence. Many real-world applications of SPS also require the ability to make sense of the physical environment. We have witnessed the widespread adoption of machine learning (ML) in smart products and IoT solutions.

However, most machine learners are black box technologies: the software solves the prediction task, but the user is unable to comprehend why some input yields a particular output. Comprehension and explanation of predictions are paramount in many real-world applications to ensure system adoption and trust. The MAPPE project (“Mining Actionable Patterns from complex Physical Environments”) explores how to design interpretable machine learning solutions for sensor-based real-world applications.

Purpose

The purpose of this research project is to 1) review the state-of-the-art in interpretable machine learning and explainable AI, and to 2) propose and evaluate new methods. The project focuses on real-world scenarios where human activities in complex physical environments are recorded with various types of sensors.

Expected results

The project is expected to generate general principles for designing and evaluating explainable AI systems for real-world applications and, more concretely, new methods and datasets for sensor-based data mining and decision support. The project is conducted in co-production with two industrial partners who are interested in machine learning proof-of-concepts and prototypes in their respective domains.

Financiers

Do you want to know more about the project?

Do you want to develop the next project with us?