Learning Feature Representations with K-Means

By Adam Coates, Andrew Y. Ng. 2012

710 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

Deep Clustering for Unsupervised Learning of Visual Features (2018, background): “Coates and Ng [10] also use k-means to pre-train convnets, but learn each layer sequentially in a bottom-up fashion, while we do it in an end-to-end fashion.”
Unsupervised Learning by Predicting Noise (2017, methodology): “Coates & Ng (2012) uses k -means to pre-train con-vnets, by learning each layer sequentially in a bottom-up fashion.”
Differentially Private Learning Needs Better Features (or Much More Data) (2020, background): “Yet, without privacy, higher accuracy is achievable with linear models and non-learned “handcrafted” features, e.g., (Coates & Ng, 2012; Oyallon & Mallat, 2015).”
Weakly-Supervised Semantic Segmentation via Sub-Category Exploration (2020, background): “To reduce the domain knowledge requirement, Coates and Ng [7] validate that feature-learning systems with K-means can be a scalable unsupervised learning module that can train a model of the unlabeled data for extracting meaningful features.”
Unsupervised Continual Learning Via Pseudo Labels (2021, background): “Clustering is one type of unsupervised learning methods that has been extensively studied in computer vision problems [7, 54, 53, 11, 55], which requires little domain knowledge from unlabeled data compared with self-supervised learning [19].”
Ernest: Efficient Performance Prediction for Large-Scale Advanced Analytics (2016, background): “performing floating point operations like matrix-vector multiplication or convolutions [32], and thus are sensitive to the number of cores and memory bandwidth available.”
Clustering with Bregman Divergences: an Asymptotic Analysis (2016, methodology): “This situation also arises in k-means feature-based methods which have seen significant success in computer vision, e.g., [6].”
Interactive Machine Learning via a GPU-accelerated Toolkit (2017, methodology): “Though the algorithm is simple, it can ﬁnd very useful feature representations if used properly [9].”
Meta-Learning Update Rules for Unsupervised Representation Learning (2018, background): “Coates and Ng (2012) showed that k-means can be used for feature learning.”
Machine learning for Internet of Things data analysis: A survey (2017, background): “1 [76] [77] [78] Density-Based Spatial Clustering of Applications with Noise Clustering 5.”
Neural Networks and Deep Learning (2018, neutral): “The work in [76] clusters on small image patches with a k-means algorithm in order to generate features.”
Calibrated Structured Prediction (2015, methodology): “We train a linear SVM on features derived from k-means clustering and that produce high accuracies (79%) on this dataset [16].”
KeystoneML: Optimizing Pipelines for Large-Scale Advanced Analytics (2016, methodology): “Specifically, TensorFlow implements a model similar to the one presented in [33], while in KeystoneML we implement a version of the model similar to [16].”