Extended Data Fig. 5: The VAU. | Nature Electronics

Extended Data Fig. 5: The VAU.

From: An index-free sparse neural network using two-dimensional semiconductor ferroelectric field-effect transistors

Extended Data Fig. 5

a, Two steps (by column vectors or rows vectors) of training process with VAU. b, The mean square error (MSE) of VAU as a function of sparsity, derived from 50 runs of VAU. Errors are generated by comparing the delta weights from VAU and the accurate ones from the cell-by-cell method. On each box, the central mark indicates the median of MSE for 50 runs, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The outliers are plotted individually using the ‘o’ marker symbol outside the bounds of maximum and minimum. c, The training accuracy of sparse neural network (SPNN) with conventional cell-by-cell method and VAU (with 10% write noise and 4.25% read noise introduced), derived from 20 runs of corresponding algorithms. Here, only the maximums and minimums are plotted in this box plot. See more details in Methods.

Back to article page