📚 node [[candidate_sampling|candidate sampling]]
Welcome! Nobody has contributed anything to 'candidate_sampling|candidate sampling' yet. You can:
  • Write something in the document below!
    • There is at least one public document in every node in the Agora. Whatever you write in it will be integrated and made available for the next visitor to read and edit.
  • Write to the Agora from social media.
    • If you follow Agora bot on a supported platform and include the wikilink [[candidate_sampling|candidate sampling]] in a post, the Agora will link it here and optionally integrate your writing.
  • Sign up as a full Agora user.
    • As a full user you will be able to contribute your personal notes and resources directly to this knowledge commons. Some setup required :)
⥅ related node [[candidate_sampling]]
⥅ node [[candidate_sampling]] pulled by Agora
📓 garden/KGBicheno/Artificial Intelligence/Introduction to AI/Week 3 - Introduction/Definitions/Candidate_Sampling.md by @KGBicheno

candidate sampling

Go back to the [[AI Glossary]]

A training-time optimization in which a probability is calculated for all the positive labels, using, for example, softmax, but only for a random sample of negative labels. For example, if we have an example labeled beagle and dog candidate sampling computes the predicted probabilities and corresponding loss terms for the beagle and dog class outputs in addition to a random subset of the remaining classes (cat, lollipop, fence). The idea is that the negative classes can learn from less frequent negative reinforcement as long as positive classes always get proper positive reinforcement, and this is indeed observed empirically. The motivation for candidate sampling is a computational efficiency win from not computing predictions for all negatives.
📖 stoas
⥱ context
← back
ai glossary

↑ pushing here
(none)

↓ pulling this
(none)
→ forward
(none)
🔎 full text search for 'candidate_sampling|candidate sampling'