A New Meta-Baseline for Few-Shot Learning

By Yinbo Chen, Xiaolong Wang, Zhuang Liu, Huijuan Xu, Trevor Darrell. 2020

224 citations recorded by the source corpus.

Subfields: Representation & Self-Supervised Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Representation & Self-Supervised Learning

Citations in the corpus

Rethinking Few-Shot Image Classification: a Good Embedding Is All You Need? (2020, methodology): “earning is more generally applicable to few-shot learning); and we use a ﬁxed neural network for feature extraction, rather than ﬁne-tuning it on the meta-testing set. The ﬁndings in concurrent works [6,20] are inline with our simple baseline. Our model learns representations by training a neural ne”
ORBIT: A Real-World Few-Shot Dataset for Teachable Object Recognition (2021, methodology): “The model can be optimized to either i) directly recognize a set of objects [46, 5] or ii) learn how to recognize a set of objects (i.”
CrossTransformers: spatially-aware few-shot transfer (2020, methodology): “We incorporate two improvements from Meta-Baseline [16], which at test time is similar to Prototypical Nets (though it isn’t trained as an episodic learner).”
Are We Overfitting to Experimental Setups in Recognition (2020, methodology): “*Equal contribution Recent works have raised concerns about the progress in few-shot and continual learning by presenting simple baselines that are comparable to state of the art in each field [32, 6, 48, 44].”
Universal Representation Learning from Multiple Domains for Few-shot Classification (2021, methodology): “Baseline++ [6] only updates a parametric classifier with cosine distance, while Meta-Baseline [7] fine-tunes entire network with a nearestcentroid cosine similarity and a scale parameter.”
Characterizing Generalization under Out-Of-Distribution Shifts in Deep Metric Learning (2021, background): “As a remedy few-shot learning [60, 67, 14, 48, 32, 6, 66] assumes few samples of the test distribution to to be available during training, i.”
FedBABU: Towards Enhanced Representation for Federated Image Classification (2021, background): “In addition, Yang et al. (2020) and Chen et al. (2020b) utilized transfer learning to enhance local models by transferring knowledge between related clients.”
Orthogonal Projection Loss (2021, unknown): “Adaptation of supervised pre-trained models to the episodic evaluation setting of FSL tasks is explored in [61, 4].”
Improving Task Adaptation for Cross-domain Few-shot Learning (2021, methodology): “Meta-Baseline [6] and Baseline [9] ﬁne-tune the whole model during meta-test, while Baseline++ [5] only updates a parametric classiﬁer with the cosine distance, RFS [41] learns a linear classiﬁer with Logistic regression and MetaOptNet [21] optimizes a linear support vector machine (SVM).”
Tip-Adapter: Training-free Adaption of CLIP for Few-shot Classification (2022, methodology): “Acquiring a cache model from fewshot training data has been explored by many previous methods, including Matching Network [59], Prototypical Networks [53], MAML [15], Relation Network [55] and others [12, 7, 57, 6].”
A Comprehensive Survey of Few-shot Learning: Evolution, Applications, Challenges, and Opportunities (2022, methodology): “Especially for few-shot image classification, as a pre-training model [71], [72], it needs to use an external large-scale label dataset to extract prior knowledge from similar task.”
Learning a Universal Template for Few-shot Dataset Generalization (2021, methodology): “Following recent work (Chen et al., 2019; 2020; Dvornik et al., 2020), we treat each of those readout heads as a cosine classiﬁer , i.e. a layer without a bias, parameterized only by a weight matrix, where the activations that are the inputs to the layer, as well as the rows of that matrix are…”
A Universal Representation Transformer Layer for Few-Shot Image Classification (2020, background): “[3], the training episodes have 50% probability coming from the ImageNet data source.”