Natural Image Statistics - A Probabilistic Approach to Early Computational Vision

By Aapo Hyvärinen, J. Hurri, Patrik O. Hoyer. 2009

687 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Molecular Biology, Signal Processing

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Molecular Biology

In Signal Processing

Citations in the corpus

NICE: Non-linear Independent Components Estimation (2014, positive): “A standard logistic distribution is used as prior for MNIST and TFD, as its negative log-likelihood function correspond to the smooth L 1 penalty (Hyv¨arinen et al., 2009) x (cid:55)→ log cosh( x ) .”
Generating images with recurrent adversarial networks (2016, neutral): “The early motivation for generating images was mainly as a diagnostic tool, based on the belief that a good generative model can count as evidence for the degree of “understanding” that a model has of the visual world (see, example, [6], [7], or [16] and references in these).”
A note on the evaluation of generative models (2015, neutral): “Examples include visualizations of model samples, interpretations of model parameters (Hyv¨arinen et al., 2009), Parzen window estimates of the model’s log-likelihood (Breuleux et al., 2009), and evaluations of model performance in surrogate tasks such as denoising or missing value imputation.”
Representation Learning: A Review and New Perspectives (2012, neutral): “Instead, we refer the reader to [101], [103].”
Building high-level features using large scale unsupervised learning (2011, positive): “y types of pooling such as mean pooling, max pooling [17] and square-squareroot pooling [15] but in this work we will use the last type of pooling because it allows the learning of invariant features [13, 15]. For learned invariances when the same algorithm is trained on natural images, see http://c”
A Survey of Inductive Biases for Factorial Representation-Learning (2016, neutral): “4, which shows samples from a dataset of nature images from [23].”
Learning to Relate Images (2011, neutral): “This operation can be shown to encode translations independently of content [7], [37] (cf. Section 3).”
A Hebbian/Anti-Hebbian Neural Network for Linear Subspace Learning: A Derivation from Multidimensional Scaling of Streaming Data (2015, positive): “Early sensory processing reduces the dimensionality of streamed inputs (Hyv¨arinen et al., 2009) as evidenced by a high ratio of input to output nerve ﬁber counts (Shep-herd, 2003).”
Scaling Up Spike-and-Slab Models for Unsupervised Feature Learning (2013, positive): “As described in (Hyvärinen et al., 2009), a phenomenon called “end stopping” similar to explaining away has been observed in V1 simple cells.”
Central Similarity Quantization for Efficient Image and Video Retrieval (2019, neutral): “is deﬁned as L Q = P N 6= j (jjj2h i 1j 1jj), where 1 2RKis an all-one vector. As L Qis a non-smooth function which makes it difﬁcult to calculate its derivative, we adopt the smooth function logcosh [10] to replace it. So jxjˇlogcoshx. Then the quantization loss L Qbecomes L Q= XN i XK k=1 (logcosh”
Learning Feature Representations with K-Means (2012, positive): “To remedy this situation, one should use whitening (also called “sphering”) to rescale the input data to remove these correlations [22].”
Deep Representations for Iris, Face, and Fingerprint Spoofing Detection (2014, neutral): “In contrast to LBP and LPQ approaches, BSIF learns a filter set by using statistics of natural images [75].”
Natural image denoising: Optimality and inherent bounds (2011, neutral): “Furthermore, numerous research attempts have been devoted to the learning of natural image priors [27, 2, 21, 26, 18, 12].”
A review of unsupervised feature learning and deep learning for time-series modeling (2014, neutral): “[73] presents an unsupervised spatio-temporal feature learning method using an extension of Independent Subspace Analysis (ISA) [60].”