Fig. 6: Multi-task classification performance. | Nature Communications

Fig. 6: Multi-task classification performance.

From: Wearable in-sensor reservoir computing using optoelectronic polymers with through-space charge-transport characteristics for multi-task learning

Fig. 6

a The stacked feature vectors of each class (a combination of sub-tasks). Each column represents the feature of a sample retrieved by the organic optical reservoir. b The details of the features of two different classes. c The 3D visualization of the feature vector distribution using linear discriminative analysis (LDA) for dimensionality reduction. d Comparison of the accuracy and number of trainable weights between our optoelectronic reservoir-computing system, a single-layered ANN, and a double-layered ANN. Our system achieves comparable performance with drastically reduced number of trainable parameters. e Confusion matrices of the reservoir-computing system on the three sub-tasks. The system achieves sound accuracy in all sub-tasks where matrix rows are dominated by the diagonal elements. f Comparison of training and inference costs among our optoelectronic RC system, a single-layered ANN, and a double-layered ANN. Our optoelectronic RC system features the least MAC operations in both training and inference.

Back to article page