One-shot learning by inverting a compositional causal process

By Brenden M. Lake, Ruslan Salakhutdinov, Josh Tenenbaum. 2013

272 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

Few-shot Autoregressive Density Estimation: Towards Learning to Learn Distributions (2017, methodology): “Contemporary machine learning systems are still far behind humans in their ability to rapidly learn new visual concepts from only a few examples (Lake et al., 2013). This setting, called few-shot learning, has been studied using deep neural networks and many other approaches in the context of discri”
BIVA: A Very Deep Hierarchy of Latent Variables for Generative Modeling (2019, neutral): “For dynamically binarized MNIST and OMNIGLOT, BIVA achieves similar improvements with L1e3 = 78.41 (state-of-the-art) and L1e3 = 91.34 respectively, see Tables 10 and 11 in Appendix G. Semi-supervised learning.”
Importance Weighted Autoencoders (2015, methodology): “We evaluated the models on two benchmark datasets: MNIST, a dataset of images of handwritten digits (LeCun et al., 1998), and Omniglot, a dataset of handwritten characters in a variety of world alphabets (Lake et al., 2013).”
Ladder Variational Autoencoders (2016, methodology): “To test our models we use the standard benchmark datasets MNIST, OMNIGLOT [10] and NORB [11].”
Meta Networks (2017, methodology): “8 Deep Boltzmann Machines (Lake et al., 2013) - 62.”
Generalization in anti-causal learning (2018, background): “Lake et al. (2013) learn a causal model of image generation to outperform the state-of-the-art in one-shot classiﬁcation.”
The More You Know: Using Knowledge Graphs for Image Classification (2016, background): “There have been a few works in this area recently such as [2] which inverts the process that creates the input and [37] which learns new categories by finding common patches between classes.”
Low-Shot Learning from Imaginary Data (2018, background): “Often, these generative models have to be hand-designed for the domain, such as strokes [18, 17] or parts [38] for handwritten characters.”
TADAM: Task dependent adaptive metric for improved few-shot learning (2018, background): “[13], Edwards and Storkey [4], Lacoste et al.”
Few-Shot Image Recognition by Predicting Parameters from Activations (2017, methodology): “A method called Hierarchical Bayesian Program Learning [26] was later proposed to speciﬁcally approach the one-shot problem on character recognition by a generative model.”
Neural Voice Cloning with a Few Samples (2018, methodology): “For example, hierarchical Bayesian models are used to exploit compositionality and causality for few-shot generation of characters [Lake et al., 2013, 2015] and words in speech [Lake et al., 2014].”
Alleviating catastrophic forgetting using context-dependent gating and synaptic stabilization (2018, methodology): “task. This process of using past knowledge to rapidly solve new tasks is referred to as transfer learning, and several groups have recently proposed how ANNs can implement this form of rapid learning [22, 23, 24]. Although the XdG method we have proposed in this study likely does not support transfe”
Discrete Variational Autoencoders (2016, methodology): “We train the resulting discrete VAEs on the MNIST [42], OMNIGLOT7 [43], and Caltech-101 Silhouettes datasets [44].”
Towards a Visual Turing Challenge (2014, methodology): “In this paper we argue for a Visual Turing Test - an open domain task of question-answering based on real-world images that resemblances the famous Turing Test [27, 28] and deviates from other attempts [29, 30, 31] - and discuss challenges together with tools to benchmark different models on such ta”
Few-Shot Object Detection via Feature Reweighting (2018, background): “[21] proposes a Hierarchical Bayesian one-shot learning system that exploits compositionality and causality.”
Iterative Amortized Inference (2018, methodology): “For MNIST and Omniglot, we used a generative model architecture with 2 hidden layers, each with 512 units, a latent space of size 64, and a symmetric iterative inference model.”
Recent Advances in Open Set Recognition: A Survey (2018, background): “To meet this challenge, several related research topics have been explored including lifelong learning [1], [2], transfer learning [3], [4], [5], domain adaptation [6], [7], zero-shot [8], [9], [10], one-shot (few-shot) [11], [12], [13], [14], [15], [16], [17], [18], [19], [20] recognition/learning,”