Emergence of Object-Selective Features in Unsupervised Feature Learning

By Adam Coates, Andrej Karpathy, Andrew Y. Ng. 2012

66 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, 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 Computer Vision and Pattern Recognition

In Representation & Self-Supervised Learning

Citations in the corpus

Predicting Parameters in Deep Learning (2013, neutral): “This has parallels to other work on inducing topology in representations [10] as well as work on learning pooling structures in deep networks [4].”
Distributed learning of deep neural network over multiple agents (2018, methodology): “Compressed representations generated using un-supervised learning and autoencoders can be used directly for classiﬁcation [32].”
Deep Learning of Representations: Looking Forward (2013, background): “One option, explored by Coates et al. (2012) is to use as building blocks for learning features algorithms such as k-means that can be run efficiently over large minibatches (or the whole data) and thus parallelized easily on a cluster (they learned 150,000 features on a cluster with only 30…”
On the importance of single directions for generalization (2018, background): “The selectivity index also identiﬁes units with the same class tuning properties which have been highlighted in the analysis of DNNs (Le et al., 2011, Zeiler & Fergus, 2014, Coates et al., 2012, Zhou et al., 2014, Radford et al., 2017).”
One-Shot Adaptation of Supervised Deep Convolutional Models (2013, background): “Earlier deep learning efforts addressed this challenge by learning layers in an unsupervised fashion using unlabeled data to discover salient mid-level structures [6, 8].”
Learning Semantic Image Representations at a Large Scale (2014, methodology): “We note that related work such as [18] adopt a similar approach by max-pooling the outputs of similar codes to generate next-layer features in a deep fashion.”
Deep learning with COTS HPC systems (2013, background): “Like (Le et al., 2012) and a similar result in (Coates et al., 2012), we have found that some neurons in our network are selective for faces.”
Direct Modeling of Complex Invariances for Visual Object Features (2013, background): “Indeed, the current direction tends to require considerable additional data, such as video datasets (Zou et al., 2012), or in some cases massive amount of unlabeled data (Le et al., 2012; Coates et al., 2012).”
The Visual System's Internal Model of the World (2015, background): “From this perspective, simple cells could be tuned to a great variety of specific features, similar to the rich dictionary of feature detectors learned using K-mean related methods in computer vision [54], [60], [61].”
Blockout: Dynamic Model Selection for Hierarchical Deep Networks (2015, background): “Lower layers tend to represent general, low-level image features like lines and edges while higher layers encode higher-level concepts like object parts or even objects themselves [4].”