Emerging Machine Learning Using Siamese and Triplet Neural Networks

Machine learning (ML) systems often encounter difficulties in executing and assessing using a paucity of training data. Emerging ML schemes, namely, multi-branch neural networks (NNs), are a promising solution for performing tasks like similarity/dissimilarity with limited pre-known information. This chapter comprehensively introduces the use of two popular multi-branch NNs (i.e., Siamese and Triplet networks) for emerging ML systems. Siamese and Triplet networks and their training and inference processes are initially reviewed. Hardware-efficient schemes for implementing these networks are then discussed; furthermore, reliable learning in multi-branch NN systems under different types of hardware errors are analyzed and evaluated. Finally, the latest error-tolerant techniques utilized for these networks are reviewed.