One shot learning of simple visual concepts

By Brenden M. Lake, Ruslan Salakhutdinov, Jason Gross, Josh Tenenbaum. 2011

904 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

Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (2017, methodology): “For Omniglot, we used strided convolutions instead of max-pooling.”
Matching Networks for One Shot Learning (2016, methodology): “Omniglot [14] consists of 1623 characters from 50 different alphabets.”
Massively Multilingual Neural Machine Translation in the Wild: Findings and Challenges (2019, background): “Other problems closely related to multitask learning include zero or few-shot learning (Lake et al., 2011; Romera-Paredes and Torr, 2015; Vinyals et al., 2016; Pan and Yang, 2010), meta-learning and life-long learning (Thrun and Mitchell, 1995; Silver et al., 2013; Chen and Liu, 2016; Parisi et…”
Zero-Shot Learning Through Cross-Modal Transfer (2013, background): “One-Shot Learning One-shot learning [19, 20] seeks to learn a visual object class by using very few training examples.”
ORBIT: A Real-World Few-Shot Dataset for Teachable Object Recognition (2021, background): “Impressive gains have been achieved on Omniglot and miniImageNet [49, 16, 4, 33], however results are now largely saturated and highly depend on the selected feature embedding.”
Meta-Learning Probabilistic Inference for Prediction (2018, methodology): “Specifically, we consider the Omniglot (Lake et al., 2011) and miniImageNet (Ravi and Larochelle, 2017) datasets which are C-way classification tasks with kc examples per class.”
Advances and Open Problems in Federated Learning (2019, background): “• The evaluation of MAML algorithms for supervised tasks is largely focused on synthetic image classification problems [252, 331] in which infinite artificial tasks can be constructed by subsampling from classes of images.”
A Closer Look at Few-shot Classification (2019, methodology): “For character recognition, we use the Omniglot dataset Lake et al. (2011) As shown in Table A3, in both Omniglot and Omniglot → EMNIST settings, meta-learning methods outperform baseline and baseline++ in 1-shot.”
Meta-Learning with Latent Embedding Optimization (2018, background): “Few-shot learning tasks challenge models to learn a new concept or behaviour with very few examples or limited experience (Fei-Fei et al., 2006; Lake et al., 2011).”
Meta-Learning for Semi-Supervised Few-Shot Classification (2018, background): “Omniglot (Lake et al., 2011) is a dataset of 1,623 handwritten characters from 50 alphabets.”
Differentially Private Meta-Learning (2019, methodology): “For few-shot image classiﬁcation, we use the Omniglot (Lake et al., 2011) and Mini-ImageNet (Ravi and Larochelle, 2017) datasets, both with 5 -shot- 5 -way test tasks.”
LEAF: A Benchmark for Federated Settings (2018, background): “However, popular meta-learning benchmarks such as Omniglot [15, 7, 33, 31] and miniImageNet [29, 7, 33, 31] focus on k -shot learning (i.e., all tasks have the same number of samples, each class has the same number of samples in each task, etc.) and thus fail to capture the real-world challenges…”
Meta-SGD: Learning to Learn Quickly for Few Shot Learning (2017, background): “Few-shot learning can be addressed with generative models [7, 8, 15].”