eduzhai > Applied Sciences > Engineering >

Layer-Wise Adaptive Updating for Few-Shot Image Classification

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 7 pages

Abstract: Few-shot image classification (FSIC), which requires a model to recognize newcategories via learning from few images of these categories, has attracted lotsof attention. Recently, meta-learning based methods have been shown as apromising direction for FSIC. Commonly, they train a meta-learner(meta-learning model) to learn easy fine-tuning weight, and when solving anFSIC task, the meta-learner efficiently fine-tunes itself to a task-specificmodel by updating itself on few images of the task. In this paper, we propose anovel meta-learning based layer-wise adaptive updating (LWAU) method for FSIC.LWAU is inspired by an interesting finding that compared with common deepmodels, the meta-learner pays much more attention to update its top layer whenlearning from few images. According to this finding, we assume that themeta-learner may greatly prefer updating its top layer to updating its bottomlayers for better FSIC performance. Therefore, in LWAU, the meta-learner istrained to learn not only the easy fine-tuning model but also its favoritelayer-wise adaptive updating rule to improve its learning efficiency. Extensiveexperiments show that with the layer-wise adaptive updating rule, the proposedLWAU: 1) outperforms existing few-shot classification methods with a clearmargin; 2) learns from few images more efficiently by at least 5 times thanexisting meta-learners when solving FSIC.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×