Figure 2
From: An extremely simple macroscale electronic skin realized by deep machine learning
The dataset structure. (a) Training dataset structure consists of a green box and a red slab; green box represents input data involving 15,000 (=1,500 × 10) 8-dimensional vectors (blue bars), the 8 resistance values constituting each vector were instantaneously measured from 8 probe terminals connected to the CNT-PPDMS sheet, and red slab denotes displacement (pressure) matrix, wherein red dots are pressure values corresponding each input vector. The total dataset consists of 36 (6 × 6) sectoral datasets and thereby 540,000 resistance input vectors and pressure values were used for the DNN training. (b) The test dataset also has the same dataset structure but not the same shape and dimension exhibiting a green slab and a red thinner slab, each of which represents 3,000 (=1,500 × 2) 8-dimensional vectors (blue bars) and the corresponding pressure values, respectively. Note that the most important point here is that the training dataset was made by a wooden bar but the test dataset was made by a toy finger.
