eduzhai > Applied Sciences > Engineering >

Batch-Incremental Triplet Sampling for Training Triplet Networks Using Bayesian Updating Theorem

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 7 pages

Abstract: Variants of Triplet networks are robust entities for learning adiscriminative embedding subspace. There exist different triplet miningapproaches for selecting the most suitable training triplets. Some of thesemining methods rely on the extreme distances between instances, and some othersmake use of sampling. However, sampling from stochastic distributions of datarather than sampling merely from the existing embedding instances can providemore discriminative information. In this work, we sample triplets fromdistributions of data rather than from existing instances. We consider amultivariate normal distribution for the embedding of each class. UsingBayesian updating and conjugate priors, we update the distributions of classesdynamically by receiving the new mini-batches of training data. The proposedtriplet mining with Bayesian updating can be used with any triplet-based lossfunction, e.g., triplet-loss or Neighborhood Component Analysis (NCA) loss.Accordingly, Our triplet mining approaches are called Bayesian Updating Triplet(BUT) and Bayesian Updating NCA (BUNCA), depending on which loss function isbeing used. Experimental results on two public datasets, namely MNIST andhistopathology colorectal cancer (CRC), substantiate the effectiveness of theproposed triplet mining method.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×