One-shot learning seems to work really well in many application domains. Are there any major (or even minor) drawbacks of using one-shot learning? Does it have flaws that could prevent it from being used in certain image identification scenarios?

In this case, I'm specifically referring to the Siamese Neural Network and Memory Augmented Neural Network approaches to one-shot learning.

  • $\begingroup$ It depends on what you mean by "one-shot learning". If the population is changing, then one-shot learning would still need to be applied every time it changes. But I would say that learning from as few examples as possible is the ultimate goal (in terms of machine learning, i.e. learning from data). For example, would you choose a person who learns faster, because it uses fewer examples, or a person which requires more time to learn, if you had to hire it for a job (assuming equality/fairness in all other aspects)? $\endgroup$ – nbro Dec 29 '18 at 17:11
  • 1
    $\begingroup$ @nbro I have edited the question to clarify this. $\endgroup$ – Aditya Radhakrishnan Dec 30 '18 at 1:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.