The "independent components" of natural scenes are edge filters.

By Anthony J. Bell, Terrence J. Sejnowski. 1997

How researchers see this paper

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

In Neuroscience

Citations in the corpus

Pylearn2: a machine learning research library (2013, methodology): “Additionally, there are many kinds of preprocessing, such as PCA [38], ZCA [4], various kinds of local contrast normalization [46], as well as helper functions to set up the entire preprocessing pipeline from some well-known successful and documented systems [8, 45].”
Learning to See by Moving (2015, background): “This hypothesis inspired research in fully unsupervised learning from visual imagery and led to popular algorithms such as ICA [5, 20], autoencoders [24, 9], sparse coding [32], restricted boltzmann machines [27, 37] and others [6].”
Density Modeling of Images using a Generalized Normalization Transformation (2015, background): “…properties at the output is central to theories of efﬁcient sensory coding in neurobiology (Barlow, 1961; Ruderman, 1994; Rieke et al., 1995; Bell & Sejnowski, 1997; Schwartz & Simoncelli, 2001), and also lends itself naturally to the design of cascaded representations such as deep neural…”
Independent component analysis: algorithms and applications (2000, methodology): “A better method is the recently introduced Wavelet Shrinkage method (Donoho, Johnstone & Picard, 1995) in which a transform based on wavelets is used, or methods based on median ﬁltering (Gonzalez & Wintz, 1987).”
Learning Intermediate-Level Representations of Form and Motion from Natural Movies (2012, background): “…many of the observed response properties of neurons in V1 may be accounted for in terms of a sparse coding model of images (Olshausen & Field, 1997; Bell & Sejnowski, 1997): I(x,t) = ∑ i ui(t)Ai(x) + n(x,t), (3.1) where I(x,t) is the image intensity as a function of space (x ∈ R2) and time (t ∈…”
A tutorial on the free-energy framework for modelling perception and learning (2017, methodology): “Existing models of feature extraction (Bell & Sejnowski, 1997; Bogacz, Brown, & Giraud-Carrier, 2001; Olshausen & Field, 1995) and predictive coding (Rao & Ballard, 1999) have been shown to be able to find features efficiently and reproduce the receptive fields of neurons in the primary visual…”
Dynamic visual attention: searching for coding length increments (2008, unknown): “With standard meth ods [2], we learn a set of basis functions that yields a sparse representation of natural image patche s.”
Sparse coding based visual tracking: Review and experimental comparison (2013, methodology): “Zhang et al. [21] proposed an appearance model based on the response distribution of basis functions learned using independent component analysis (ICA) [37].”
Learning Feature Representations with K-Means (2012, methodology): “This metric is easy to compute by first whitening the input data Z with ZCA whitening [3], then computing the pairwise similarities between all of the features:”
A critique of pure learning and what artificial neural networks can learn from animal brains (2019, methodology): “Indeed, there are several unsupervised algorithms which generate representations reminiscent of those found in the visual system 16 – 18 .”
Natural image denoising: Optimality and inherent bounds (2011, background): “Furthermore, numerous research attempts have been devoted to the learning of natural image priors [27, 2, 21, 26, 18, 12].”
Unsupervised Discovery of Nonlinear Structure Using Contrastive Backpropagation (2006, background): “structure in natural images (Olshausen and Field, 1996; Bell and Sejnowski, 1997 ).”
Emergence of Phase- and Shift-Invariant Features by Decomposition of Natural Images into Independent Feature Subspaces (2000, methodology): “One of the stimulusparameterswaschangedatatimetoseehowtheresponsechanges, whiletheotherparameterswereheldconstantattheoptimalvalues.”
Continuous latent variable models for dimensionality reduction and sequential data reconstruction (2001, methodology): “1 Other measures of sparseness Kurtosis, the fourth-order cumulant of a distribution, has been proposed and used by Field (1994) and others (Olshausen and Field, 1996, 1997; Bell and Sejnowski, 1997; Vinje and Gallant, 2000) as a measure of sparseness in the context of neural codes, based on the exp”
Nonlinear Extraction of Independent Components of Natural Images Using Radial Gaussianization (2009, methodology): “…Mallat & Zhang, 1993; Olshausen & Field, 1997; Chen, Donoho, & Saunders, 1998; Lewicki & Sejnowski, 2000; Donoho & Elad, 2003). the expected value of a higher-order contrast function to avoid the difﬁ-culties associated with entropy estimation (Comon, 1994; Bell & Sejnowski, 1997; Cardoso, 1999).”