Compositionality, MDL Priors, and Object Recognition

By Elie Bienenstock, Stuart Geman, Daniel Potter. 1996

How researchers see this paper

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

Citations in the corpus

Compositional Convolutional Neural Networks: A Deep Architecture With Innate Robustness to Partial Occlusion (2020, background): “Compositionality is a fundamental aspect of human cognition [2, 28, 9, 3] that is also reflected in the hierarchical compositional structure of the ventral stream in visual cortex [34, 27, 21].”
Learning generative models of mid-level structure in natural images (2012, background): “This will effectively enable the DSN to model structured noise at different levels of complexity in a manner similar to ideas in compositional models (Jin and Geman, 2006; Bienenstock et al., 1996) in which parts can be bound into higher-level structures but can also remain unbound if such a binding”
Hierarchical Matching of Deformable Shapes (2007, methodology): “Our compositional approach is related to the work in [4].”
Generative Modeling of Infinite Occluded Objects for Compositional Scene Representation (2019, background): “By decomposing the raw observations, organized and compact knowledge of visual scenes can be summarized (Bienenstock et al., 1997).”
Image Parsing: Unifying Segmentation, Detection, and Recognition (2003, unknown): “Instead, the parsing subgraphs should be built by grouping small elements in a “compositional” manner [ 5 ].”
Statistical Modeling and Conceptualization of Visual Patterns (2003, background): “For the other levels, qðWIþ1jWiÞ 1⁄4 qðWiþ1j iðWiÞÞ with iðWiÞ being some compatibility functions and metrics [75], [7].”
Compositional Boosting for Computing Hierarchical Image Structures (2007, background): “More specifically, as compositionality and part-sharing [5, 3, 11] are common principles in object modeling and recognition, an effective inference algorithm must explore the compositional structures.”
A generative vision model that trains with high data efficiency and breaks text-based CAPTCHAs (2017, background): “Second, hierarchy provides efficiency through the sharing of features between different objects (42).”