Multi-objective Design of Hardware Accelerators for Levodopa-Induced Dyskinesia Classifiers

Taking levodopa, a drug used to treat symptoms of Parkinson's disease, is often connected with severe side effects, known as Levodopa-induced dyskinesia (LID). It can fluctuate in severity throughout the day and thus is difficult to classify during a short period of a physician's visit. A low-power wearable classifier enabling long-term and continuous LID classification would thus significantly help with LID detection and dosage adjustment. This paper deals with a co-evolutionary design of energy-efficient hardware accelerators of LID classifiers that can be implemented in wearable devices. The accelerator consists of a feature extractor and a classifier co-evolved using cartesian genetic programming (CGP). We introduce and evaluate a fast and accurate energy consumption estimation method for the target architecture of considered classifiers. The proposed energy estimation method allows for a multi-objective design enabled by introducing energy constraints. With the introduction of variable data representation bit width, the proposed method achieves a good trade-off between accuracy (AUC) and energy consumption.