InfoNCE Loss

  • InfoNCE loss is the improvement over the Contrastive Loss
  • In InfoNCE, there is one anchor, one positive and multiple negative samples for each instance
  • It converge faster than the Triplet Loss

[!def] InfoNCE Loss Formula
loss = -E_x (log \frac{f(x^+)}{f(x^+) + \sum_{i \in N} f(x_i^-) })


  1. Need to revise and understand the loss function
  2. Especially the expected value part