Attentive Recurrent Comparators

By Pranav Shyam, Shubham Gupta, Ambedkar Dukkipati. 2017

116 citations recorded by the source corpus.

Rapid learning requires flexible representations to quickly adopt to new evidence. We develop a novel class of models called Attentive Recurrent Comparators (ARCs) that form representations of objects by cycling through them and making observations. Using the representations extracted by ARCs, we develop a way of approximating a \textit{dynamic representation space} and use it for one-shot learning. In the task of one-shot classification on the Omniglot dataset, we achieve the state of the art performance with an error rate of 1.5\%. This represents the first super-human result achieved for this task with a generic model that uses only pixel information.

Subfields: Cognitive Neuroscience, Computer Vision and Pattern Recognition, Model Architectures

How researchers see this paper

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

In Cognitive Neuroscience

In Computer Vision and Pattern Recognition

In Model Architectures

Citations in the corpus

Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (2017, background): “One successful approach for few-shot classiﬁcation is to learn to compare new examples in a learned metric space using e.g. Siamese networks (Koch, 2015) or recurrence with attention mechanisms (Vinyals et al., 2016; Shyam et al., 2017; Snell et al., 2017).”
Auto-Meta: Automated Gradient Based Meta Learner Search (2018, neutral): “The metric-based meta-learning has been known to perform well for few-shot image classification tasks [24, 23].”
Data Augmentation Generative Adversarial Networks (2017, methodology): “…we saw an improvement of 0.5% over the state of the art (96.9% to 97.4%), pushes the matching network performance to the level of the Conv Arc (Shyam et al., 2017) technique which used substantial prior knowledge of the data form (that the data was built from pen strokes) to achieve those…”
TADAM: Task dependent adaptive metric for improved few-shot learning (2018, background): “The attentive recurrent comparators [27] go beyond classical siamese approaches and use a recurrent architecture to learn to perform pairwise comparisons and predict if the compared examples belong to the same class.”
Sample Efficient Adaptive Text-to-Speech (2018, background): “…studies have attempted to address the problem of few-shot learning by using deep neural networks, and they have shown promising results on classiﬁcation tasks in vision (Santoro et al., 2016; Shyam et al., 2017), language (Vinyals et al., 2016) and speech recognition (Abdel-Hamid and Jiang, 2013).”
Few-Shot Learning Through an Information Retrieval Lens (2017, background): “Attentive Recurrent Comparators [16] also perform pairwise comparisons but form the representation of the pair through a sequence of glimpses at the two points that comprise it via a recurrent neural network.”
Generalizing from a Few Examples (2019, methodology): “The two most popular applications are character recognition [14, 36, 37, 65, 70, 96, 113, 114, 119, 121, 130, 138, 146] and image classification [37, 70, 96, 106, 119, 121, 127, 130, 132, 138, 142, 143, 149].”
One-Shot Action Localization by Learning Sequence Matching Network (2018, background): “[43, 36, 39] refine this idea by introducing recurrence and attention mechanisms.”
Few-shot learning with adaptively initialized task optimizer: a practical meta-learning approach (2019, background): “Hariharan and Girshick (2017) and Wang et al. (2018b)) learn a generation prior to augment examples given a single image; Dai et al. (2017) extracts logical derivations from related tasks; Wang et al. (2017a)) and Shyam et al. (2017) utilize the prior to attend images.”
Graph Matching Networks for Learning the Similarity of Graph Structured Objects (2019, result): “Independent of our work, recently (Shyam et al., 2017) proposed a cross-example attention model for visual similarity as an alternative to Siamese networks based on similar motivations and achieved good results.”
Evolutionary architecture search for deep multitask networks (2018, methodology): “use there is clear intuition that knowledge of several alphabets will make learning another one easier. Omniglot has been used in an array of settings: generative modeling [23, 36], one-shot learning [21, 23, 40], and deep MTL [2, 28, 29, 35, 49]. Previous deep MTL approaches used random training/te”
A survey of deep meta-learning (2020, background): “Source: Shyam et al. (2017).”
The Omniglot Challenge: A 3-Year Progress Report (2019, background): “Attentive recurrent comparators (ARCs) use a learned attention mechanism to make repeated targeted comparisons between two images (Shyam et al., 2017), achieving”
Multimodal Model-Agnostic Meta-Learning via Task-Aware Modulation (2019, background): “To classify images with few examples, metric-based meta-learning frameworks have been proposed [16, 48, 42, 41, 43, 29, 3], which strive to learn a metric or distance function that can be used to compare two different samples effectively.”
A Two-Stage Approach to Few-Shot Learning for Image Recognition (2019, background): “Attentive comparators [41] compare patches of images sequentially through an attention mechanism and then arrive at a prediction.”
Multimodal One-shot Learning of Speech and Images (2018, background): “Apart from Siamese models [4], which we focus on here, other metric learning based approaches have been proposed [21–23] which build on advances in attention and memory mechanisms for neural networks.”
Generative Adversarial Residual Pairwise Networks for One Shot Learning (2017, background): “The recent work in [29] proposes a recurrent model that also outputs a pairwise embedding, however their reasoning is towards using attention to selectively infer the next step whereas we are focused on a natural, scalable extension to the CNN for one-shot learning.”