Siamese Network is a type of network architecture that contains two or more identical subnetwork used for generate feature vectors for each input and compare them.
A Siamese Neural Network is a class of neural network architectures that contain two or more identical subnetworks. Identical, here means, they have the same configuration with the same parameters and weights. Parameter updating is mirrored across both sub-networks. It is used to find the similarity of the inputs by comparing its feature vectors, so these networks are used in many applications
We will deal the case in whitch the SNN is to train using the Triplet Loss function. It is based on the distance on three inputs:
- Anchor (arbitrary class of points)
- Positive (same class of the Anchor)
- Negative (different class from the Anchor)
The main advantages of Siamese Networks are:
- More Robust to class Imbalance: With the aid of One-shot learning, given a few images per class is sufficient for Siamese Networks to recognize those images in the future.
- Nice to an ensemble with the best classifier: Given that its learning mechanism is somewhat different from Classification, simple averaging of it with a Classifier can do much better than average 2 correlated Supervised models (e.g. RF classifier)
- Learning from Semantic Similarity: Siamese focuses on learning embeddings (in the deeper layer) that place the same classes/concepts close together. Hence, can learn semantic similarity.
The downsides of the Siamese Networks can be:
- Needs more training time than normal networks: Since Siamese Networks involves quadratic pairs to learn from (to see all information available) it is slower than normal classification type of learning(pointwise learning).
- Doesn’t output probabilities: Since training involves pairwise learning, it won’t output the probabilities of the prediction, but the distance from each class
When the input are the same, and so d(A,P) = d(A,N)= 0, the loss i equal to zero. This is call trivial solution.
To prevent trivial output, we introduce a new term call margin, that pushes the anchor-positive pair and the anchor-negative pair further away from each other.
During training, if A,P and N are choose randomly, d(A,P) + margin <= d(A,N) is easily satisfied.
So we are interesting to choose triplets that are "hard" to train on.
In that case the learning algorithm has to try hard to keep these on the right order, so that there is at least a margin between the left and the right side. This increasing the efficiency of the learning algorithm.
Than we can apply the Gradient Descent to try to minimize the Cost function J.
Image similarity estimation using a Siamese Network with a triplet loss