Dynamic Few-Shot Visual Learning Without Forgetting
By Spyros Gidaris, Nikos Komodakis. 2018
1,219 citations recorded by the source corpus.
The human visual system has the remarkably ability to be able to effortlessly learn novel concepts from only a few examples. Mimicking the same behavior on machine learning vision systems is an interesting and very challenging research problem with many practical advantages on real world vision applications. In this context, the goal of our work is to devise a few-shot visual learning system that during test time it will be able to efficiently learn novel categories from only a few training data while at the same time it will not forget the initial categories on which it was trained (here called base categories). To achieve that goal we propose (a) to extend an object recognition system with an attention based few-shot classification weight generator, and (b) to redesign the classifier of a ConvNet model as the cosine similarity function between feature representations and classification weight vectors. The latter, apart from unifying the recognition of both novel and base categories, it also leads to feature representations that generalize better on "unseen" categories. We extensively evaluate our approach on Mini-ImageNet where we manage to improve the prior state-of-the-art on few-shot recognition (i.e., we achieve 56.20% and 73.00% on the 1-shot and 5-shot settings respectively) while at the same time we do not sacrifice any accuracy on the base categories, which is a characteristic that most prior approaches lack. Finally, we apply our approach on the recently introduced few-shot benchmark of Bharath and Girshick [4] where we also achieve state-of-the-art results.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Meta & Continual Learning
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Bernt Schiele
- Dacheng Tao
- Trevor Darrell
- Yoshua Bengio
- Sergey Levine
- Mohit Bansal
- Han-Jia Ye
- Quoc V. Le
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Dacheng Tao
- Trevor Darrell
- Bernt Schiele
- Ming-Hsuan Yang
- Xiaogang Wang
- Yi Yang
- Andrew Zisserman
In Meta & Continual Learning
Citations in the corpus
- Rethinking Few-Shot Image Classification: a Good Embedding Is All You Need? (2020, neutral): “66 Dynamic Few-shot [14] 64-64-128-128 56.”
- Meta-Learning Probabilistic Inference for Prediction (2018, background): “Approaches that employ pre-training and/or residual networks (Bauer et al., 2017; Qiao et al., 2017; Rusu et al., 2018; Gidaris and Komodakis, 2018; Oreshkin et al., 2018; Garcia and Bruna, 2017; Lacoste et al., 2018) have been excluded so that the quality of the learning algorithm can be assessed s”
- Meta-learning with differentiable closed-form solvers (2018, methodology): “For our comparison, we report the results of methods which train their models from scratch for few-shot classification, omitting very recent work of Qiao et al. (2018) and Gidaris & Komodakis (2018), which instead make use of pre-trained embeddings.”
- Meta-Dataset: A Dataset of Datasets for Learning to Learn from Few Examples (2019, background): “Others relate to Prototypical Networks by learning a data representation as well as a compact representation for a classifier of data under that representation (Bertinetto et al., 2019; Gidaris & Komodakis, 2018; Oreshkin et al., 2018; Gidaris & Komodakis, 2018).”
- A Closer Look at Few-shot Classification (2019, methodology): “Second, by replacing the linear classifier with a distance-based classifier as used in Gidaris & Komodakis (2018); Qi et al. (2018), the baseline method is surprisingly competitive to current state-of-art meta-learning algorithms.”
- Meta-Learning with Latent Embedding Optimization (2018, methodology): “Recent state-of-the-art approaches use the penultimate layer representation [31, 3, 10]; however, we choose the intermediate feature representation in layer 21, given that”
- Diversity With Cooperation: Ensemble Methods for Few-Shot Classification (2019, background): “There is indeed strong evidence [5, 11, 23] that train-1 ing CNNs from scratch using meta-learning performs substantially worse than if CNN features are trained in a standard fashion—that is, by minimizing a classical loss function relying on corpus annotations; on the other hand, learning only the…”
- Margin-Based Few-Shot Class-Incremental Learning with Class-Level Overfitting Mitigation (2022, methodology): “However, as the original FSL task (except for some subspecies task of FSL, e.g., generalized FSL [13]) does not require the recognition of base classes, FSL generally emphasizes more on the novel-class generalization.”
- Continual Learning through Retrieval and Imagination (2022, methodology): “Although co-sine normalization has been adopted in other visual problem (Gidaris et al. 2018; Wei et al. 2019; Guo et al. 2020; Qi et al. 2018; Hou et al. 2019), we demonstrate that it can effectively eliminate the imbalance in the magnitude of class embeddings and facilitate continual learning…”
- Constrained Few-shot Class-incremental Learning (2022, methodology): “For the natural image datasets, we use a Resnet-12 architecture as feature extractor [11,55].”
- Doodle It Yourself: Class Incremental Learning by Drawing a Few Sketches (2022, methodology): “Average classification accuracy of DIY-FSCIL framework using our self-designed baselines and adopted SOTA FSCIL [17] (not specifically designed for cross-modal).”
- Overcoming Catastrophic Forgetting in Incremental Few-Shot Learning by Finding Flat Minima (2021, methodology): “Current few-shot methods mainly include optimization-based methods [12, 23, 32, 39, 45, 46, 55] and metric-based methods [13, 19, 44, 49, 52, 58, 57, 53].”
- Subspace Regularizers for Few-Shot Class Incremental Learning (2021, methodology): “This version, which we call single-session FSCIL, has been extensively studied by previous work (Qi et al., 2018; Gidaris & Komodakis, 2018; Ren et al., 2019; Yoon et al., 2020).”
- Generalized and Incremental Few-Shot Learning by Explicit Learning and Calibration without Forgetting (2021, methodology): “Episodic training is a common way to evaluate few-shot learning methods, further we detail the difference from the standard training protocol [3, 11, 18].”
- Few-Shot and Continual Learning with Attentive Independent Mechanisms (2021, methodology): “Our classifier is adopted directly from [20, 15] having φ = Gθ(Z̄).”
- Learngene: From Open-World to Your Learning Task (2021, background): “Although meta continual learning (Yao et al. 2020; Gidaris and Komodakis 2018; Jerfel et al. 2019; Javed and White 2019; Gupta, Yadav, and Paull 2020) can solve the sequential task problem by recording the efficiency task buffer, it still cannot prevent data-interaction.”
- Self-Promoted Prototype Refinement for Few-Shot Class-Incremental Learning (2021, background): “To address this issue, [8] proposes a dynamic few-shot recognition system with an attention based few-shot classification weight generator.”
- Few-Shot Class-Incremental Learning via Relation Knowledge Distillation (2021, background): “Recently, some FSL works (Gidaris and Komodakis 2018; Ren et al. 2019) attempt to learn a unified model capable of recognizing both the base and novel classes while those are not able to keep learning new tasks because they rely on the base class for sampling meta-learning tasks.”