Triplet loss is a loss function commonly used in metric learning (e.g., face recognition) to learn an embedding space where similar examples are closer together and dissimilar examples are farther apart​.It operates on triplets of samples: an anchor (the reference sample), a positive (a sample of the same class or otherwise similar to the anchor), and a negative (a sample of a different class or dissimilar to the anchor)​.The triplet loss is formulated to ensure the distance between the anchor and positive embeddings is smaller (by a margin) than the distance between the anchor and negative embeddings. Mathematically, if f(x) is the embedding function and we use a distance like Euclidean distance, triplet loss aims to satisfy: d(f(anchor), f(positive)) + α < d(f(anchor), f(negative)), for a chosen margin α​By training with triplet loss, the model learns a representation where intra-class similarity is high and inter-class similarity is low, which is vital for tasks like clustering or identification without explicit classification.