Fig. 1
Network of learning. a Overview of the designed network in this study. Here, x, y, and t represent an input vector, an output vector, and a position of an amino acid sequence. In the squares, “Embed,” “Full connect,” and “Softmax” stand for a word embedding operation, a fully connected network, and a softmax function layer, respectively. The solid and broken arrows represent a matrix operation and an array operation, respectively. The numbers at the bottom of panel (a) stand for a dimension of vectors of each layer. b Description of LSTM layer. Here, u, v, h, s, ×, +, dot, τ, σ, wa, wb, and b stand for an input vector to an LSTM unit, an output vector from an LSTM unit, a previous input vector, a unit for constant error, multiplication of matrices, summation of matrices, a Hadamard product calculation, a hyperbolic tangent, a sigmoid function, a weight matrix to be learned, another weight matrix, and a bias vector