eduzhai > Applied Sciences > Engineering >

ResOT Resource-Efficient Oblique Trees for Neural Signal Classification

  • Save

... pages left unread,continue reading

Document pages: 13 pages

Abstract: Classifiers that can be implemented on chip with minimal computational andmemory resources are essential for edge computing in emerging applications suchas medical and IoT devices. This paper introduces a machine learning modelbased on oblique decision trees to enable resource-efficient classification ona neural implant. By integrating model compression with probabilistic routingand implementing cost-aware learning, our proposed model could significantlyreduce the memory and hardware cost compared to state-of-the-art models, whilemaintaining the classification accuracy. We trained the resource-efficientoblique tree with power-efficient regularization (ResOT-PE) on three neuralclassification tasks to evaluate the performance, memory, and hardwarerequirements. On seizure detection task, we were able to reduce the model sizeby 3.4X and the feature extraction cost by 14.6X compared to the ensemble ofboosted trees, using the intracranial EEG from 10 epilepsy patients. In asecond experiment, we tested the ResOT-PE model on tremor detection forParkinson s disease, using the local field potentials from 12 patientsimplanted with a deep-brain stimulation (DBS) device. We achieved a comparableclassification performance as the state-of-the-art boosted tree ensemble, whilereducing the model size and feature extraction cost by 10.6X and 6.8X,respectively. We also tested on a 6-class finger movement detection task usingECoG recordings from 9 subjects, reducing the model size by 17.6X and featurecomputation cost by 5.1X. The proposed model can enable a low-power andmemory-efficient implementation of classifiers for real-time neurologicaldisease detection and motor decoding.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×