DMN4: Few-Shot Learning via Discriminative Mutual Nearest Neighbor Neural Network
نویسندگان
چکیده
Few-shot learning (FSL) aims to classify images under low-data regimes, where the conventional pooled global feature is likely lose useful local characteristics. Recent work has achieved promising performances by using deep descriptors. They generally take all descriptors from neural networks into consideration while ignoring that some of them are useless in classification due their limited receptive field, e.g., task-irrelevant could be misleading and multiple aggregative background clutter even overwhelm object's presence. In this paper, we argue a Mutual Nearest Neighbor (MNN) relation should established explicitly select query most relevant each task discard less ones clutters FSL. Specifically, propose Discriminative Neural Network (DMN4) for Extensive experiments demonstrate our method outperforms existing state-of-the-arts on both fine-grained generalized datasets.
منابع مشابه
One-Shot Learning in Discriminative Neural Networks
Learning a concept from few examples remains a difficult problem in machine learning. Many of the breakthough successes have relied on applying gradient-based optimisation to high-capacity models. This has required large labelled datasets and long offline training times. However, there are many tasks where such data is not available and we need to quickly learn from only a few examples. In the ...
متن کاملFew-Shot Learning with Graph Neural Networks
We propose to study the problem of few-shot learning with the prism of inference on a partially observed graphical model, constructed from a collection of input images whose label can be either observed or not. By assimilating generic message-passing inference algorithms with their neural-network counterparts, we define a graph neural network architecture that generalizes several of the recentl...
متن کاملDiscriminative Learning of the Prototype Set for Nearest Neighbor Classification
The nearest neighbor rule is one of the most widely used models for classification and selecting a compact set of prototype instances is an important problem for its applications. Many existing approaches on the prototype selection problem rely on instance-based analyses and local criteria on the class distribution, which are intractable for numerical optimization techniques. In this paper, we ...
متن کاملNearest Neighbor Network Traversal
A mobile agent in a network wants to visit every node of an n-node network, using a small number of steps. We investigate the performance of the following \nearest neighbor" heuristic: always go to the nearest unvisited node. If the network graph never changes, then from (Rosenkrantz, Stearns and Lewis, 1977) and (Hurkens and Woeginger, 2004) it follows that (n logn) steps are necessary and su ...
متن کاملLearning to Compare: Relation Network for Few-Shot Learning
We present a conceptually simple, flexible, and general framework for few-shot learning, where a classifier must learn to recognise new classes given only few examples from each. Our method, called the Relation Network (RN), is trained end-to-end from scratch. During meta-learning, it learns to learn a deep distance metric to compare a small number of images within episodes, each of which is de...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Proceedings of the ... AAAI Conference on Artificial Intelligence
سال: 2022
ISSN: ['2159-5399', '2374-3468']
DOI: https://doi.org/10.1609/aaai.v36i2.20076