Fig. 5: Handwritten letter recognition and neural machine translation. | Nature Communications

Fig. 5: Handwritten letter recognition and neural machine translation.

From: Intrinsic ferroelectric CuVP2S6 for potential applications in neuromorphic recognition and translation

Fig. 5

a The convolutional neural network architecture of the OCR module achieves image-to-text conversion. b The Transformer network architecture of the NMT module accomplishes English text to Chinese text translation. c The accuracy of OCR module in recognizing the 26 letters of “a-z” normalized confusion matrices. d, e Simulation results of the NMT module for ideal and device, showing BERT-scores for translations of 100 English sentences across various training epochs. f Comparison of BERT scores for the translations of sentences numbered 90 –100, assessing the final simulations of the NMT modules in both ideal and device-based scenarios.

Back to article page