Soft Mixer Assignment in a Hierarchical Generative Model of Natural Scene Statistics

Schwartz, Odelia; Sejnowski, Terrence J.; Dayan, Peter

doi:10.1162/neco.2006.18.11.2680

Cited by 30 publications

(38 citation statements)

References 55 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This is not surprising as the structure is dictated by natural images. However, a novel class of second-layer functions emerges in our model that is not seen in the published work on the density components model (Karklin & Lewicki, 2003a, 2003b or related models (Schwartz et al, 2006). This class of functions represents elongated, collinear edge structure in the image domain (see Figure 5b).…”

Section: Learned Amplitudementioning

confidence: 87%

“…Models that learn form-selective invariances and focus on performance evaluation of object recognition tasks (Wallis & Rolls, 1997;LeCun et al, 2004;Serre et al, 2007) often have the specific invariance built in to the model structure, and the higher-order features that emerge beyond these built-in invariances have not been explored. The model we propose here bears similarities to the density components model of Karklin and Lewicki (2005) and to the hierarchical GSM model of Schwartz, Sejnowski, and Dayan (2006), which learn higher-order structure in images by modeling the dependencies in scale among oriented filter responses. Our model differs in that form-selective invariances are learned from the temporally persistent structure contained in natural movies as opposed to static image patches.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Learning Intermediate-Level Representations of Form and Motion from Natural Movies

Cadieu

Olshausen

2012

Neural Computation

View full text Add to dashboard Cite

We present a model of intermediate-level visual representation that is based on learning invariances from movies of the natural environment.The model is composed of two stages of processing: an early feature representation layer and a second layer in which invariances are explicitly represented. Invariances are learned as the result of factoring apart the temporally stable and dynamic components embedded in the early feature representation. The structure contained in these components is made explicit in the activities of second-layer units that capture invariances in both form and motion. When trained on natural movies, the first layer produces a factorization, or separation, of image content into a temporally persistent part representing local edge structure and a dynamic part representing local motion structure, consistent with known response properties in early visual cortex (area V1). This factorization linearizes statistical dependencies among the first-layer units, making them learnable by the second layer. The second-layer units are split into two populations according to the factorization in the first layer. The form-selective units receive their input from the temporally persistent part (local edge structure) and after training result in a diverse set of higher-order shape features consisting of extended contours, multiscale edges, textures, and texture boundaries. The motion-selective units receive their input from the dynamic part (local motion structure) and after training result in a representation of image translation over different spatial scales and directions, in addition to more complex deformations. These representations provide a rich description of dynamic natural images and testable hypotheses regarding intermediate-level representation in visual cortex.

show abstract

Section: Learned Amplitudementioning

confidence: 87%

Section: Introductionmentioning

confidence: 99%

Learning Intermediate-Level Representations of Form and Motion from Natural Movies

Cadieu

Olshausen

2012

Neural Computation

View full text Add to dashboard Cite

show abstract

“…This argument, together with effective nonlinear SFA models of the visual system (Berkes & Wiskott, 2005;Franzius et al, 2007), indicates that sensory systems are tailored to extract (relevant) predictive information. For further research, we suggest comparing temporally local predictive coding and slow feature analysis to generative hierarchical models for learning nonlinear statistical regularities (Karklin & Lewicki, 2005;Schwartz, Sejnowski, & Dayan, 2006).…”

Section: Discussionmentioning

confidence: 99%

Predictive Coding and the Slowness Principle: An Information-Theoretic Approach

Creutzig

Sprekeler

2008

Neural Computation

View full text Add to dashboard Cite

Understanding the guiding principles of sensory coding strategies is a main goal in computational neuroscience. Among others, the principles of predictive coding and slowness appear to capture aspects of sensory processing. Predictive coding postulates that sensory systems are adapted to the structure of their input signals such that information about future inputs is encoded. Slow feature analysis (SFA) is a method for extracting slowly varying components from quickly varying input signals, thereby learning temporally invariant features. Here, we use the information bottleneck method to state an information-theoretic objective function for temporally local predictive coding. We then show that the linear case of SFA can be interpreted as a variant of predictive coding that maximizes the mutual information between the current output of the system and the input signal in the next time step. This demonstrates that the slowness principle and predictive coding are intimately related.

show abstract

“…Compared to other computational models accounting for extraclassical receptive field properties, our model complies with physiology better than predictive coding -model which assumes separate units detecting error and predicting incoming input (Rao and Ballard 1999;Spratling 2010), and gaussian scale mixture models, which have no obvious direct relationship to physiology (Schwartz et al 2006) even though recently Schwartz et al (2009) viewed their model as commensurate with divisive normalization (Heeger 1992).…”

Section: Physiology and Theory Of Contextual Modulationmentioning

confidence: 90%

Local non-linear interactions in the visual cortex may reflect global decorrelation

Vanni

Rosenström

2010

J Comput Neurosci

View full text Add to dashboard Cite

The classical receptive field in the primary visual cortex have been successfully explained by sparse activation of relatively independent units, whose tuning properties reflect the statistical dependencies in the natural environment. Robust surround modulation, emerging from stimulation beyond the classical receptive field, has been associated with increase of lifetime sparseness in the V1, but the system-wide modulation of response strength have currently no theoretical explanation. We measured fMRI responses from human visual cortex and quantified the contextual modulation with a decorrelation coefficient (d), derived from a subtractive normalization model. All active cortical areas demonstrated local non-linear summation of responses, which were in line with hypothesis of global decorrelation of voxels responses. In addition, we found sensitivity to surrounding stimulus structure across the ventral stream, and large-scale sensitivity to the number of simultaneous objects. Response sparseness across voxel population increased consistently with larger stimuli. These data suggest that contextual modulation for a stimulus event reflect optimization of the code and perhaps increase in energy efficiency throughout the ventral stream hierarchy. Our model provides a novel prediction that average suppression of response amplitude for simultaneous stimuli across the cortical network is a monotonic function of similarity of response strengths in the network when the stimuli are presented alone.

show abstract

Soft Mixer Assignment in a Hierarchical Generative Model of Natural Scene Statistics

Cited by 30 publications

References 55 publications

Learning Intermediate-Level Representations of Form and Motion from Natural Movies

Learning Intermediate-Level Representations of Form and Motion from Natural Movies

Predictive Coding and the Slowness Principle: An Information-Theoretic Approach

Local non-linear interactions in the visual cortex may reflect global decorrelation

Contact Info

Product

Resources

About