Triplet Loss

• Triplet loss is the improvement over the Contrastive Loss
• Triplet loss expects one anchor, one positive and one negative point
• It simultaneously minimizes the Euclidian Distance between positive and anchor
• AND maximizes the Euclidian Distance between negative and anchor
• As it works simultaneously it needs less amount of sample for convergence

[!def] Triplet Loss Formula
$$
loss = max(0, |x-x^+|^2 - |x-x^-|^2 + m)
$$