Paper on Activity Recognition wins Best Paper Award

Brehler MCSoC Best Paper Award Lamarr - Lamarr Institute for Machine Learning (ML) and Artificial Intelligence (AI)

Marius Brehler, together with his colleague Lucas Camphausen, has been awarded with a Best Paper Award at the 2023 IEEE 16th International Symposium on Embedded Multicore/Many-core Systems-on-Chip (MCSoC). Their paper “Combining Decision Tree and Convolutional Neural Network for Energy Efficient On-Device Activity Recognition” explores a hybrid, hierarchical architecture that successfully reduced energy consumption for activity recognition tasks.

Activity recognition is the process of automatically identifying and classifying activities. This and the sub-problem of motion classification can be performed based on sensor data provided by inertial measurements units. One particular field of interest is the detection of human activities, which has attracted a lot of attention for a broad range of applications. Among other things, it can be used for medical applications. For many applications, it is of high interest to process the data in an energy efficient manner. This can be realized by processing the data close to the sensor on a low-cost, low power embedded device.

By combining a decision tree with a convolutional neural network in a hierarchical architecture, the scientist’s hybrid model successfully reduced the amount of required energy for motion classification tasks on a simple dataset with eight different motions and the popular UCI HAPT dataset for human activity recognition. At the same time, the demonstrated results underlined the importance of evaluating the accuracy, defined as the percentage of correctly classified motions, of the combined classifier and not limiting the evaluation to the accuracy of the decision tree.

The annual IEEE International Symposium on Embedded Multicore/Many-core Systems-on-Chip (MCSoC) focuses on topics related to embedded Artificial Intelligence, Machine Learning, and many-core neuro-inspired computing and systems. The 2023 edition of the symposium was held at the Singapore University of Technology and Design (SUTD).


More news