Siamese Neural Networks for One-shot Image Recognition

Information Technology

Learning excellent features for machine learning applications can be computationally expensive, and it can be difficult in situations when there is limited data. The one-shot learning setting is a classic example of this, in which we must accurately make predictions based on only one sample of each new class. We look at a strategy for learning Siamese neural networks that uses a unique structure to automatically priorities similarity between inputs in this research. After tuning a network, we can use powerful discriminative features to extend the network's predictive capacity not merely to fresh data, but to totally new classes from unknown distributions. We can generate strong results using a convolutional design that outperforms the competition.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.