A generative vision model that trains with high data efficiency and breaks text-based CAPTCHAs

By Dileep George, Wolfgang Lehrach, Ken Kansky, Miguel L'azaro-Gredilla, C. Laan, Bhaskara Marthi, Xinghua Lou, Zhaoshi Meng, Hua-Yan Wang, Alexander Lavin and 1 more. 2017

270 citations recorded by the source corpus.

Subfields: Cognitive Neuroscience, Computer Vision and Pattern Recognition, Information Systems, Generative Models

How researchers see this paper

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

In Cognitive Neuroscience

In Computer Vision and Pattern Recognition

In Information Systems

In Generative Models

Citations in the corpus

The Next Decade in AI: Four Steps Towards Robust Artificial Intelligence (2020, background): “See also DeepMind’s MONet (Burgess et al., 2019)and MERLIN (Wayne et al., 2018) as well the expressive generative model in (Gregor et al., 2019)] and. e.g., inverse graphics papers by Vicarious (Kansky et al., 2017; George et al., 2017) and Josh Tenenbaum’s group (Mao et al., 2019; Veerapaneni et al”
Low-Shot Learning from Imaginary Data (2018, background): “One class of approaches builds generative models that can share priors across categories [7, 25, 10].”
Deep Learning: A Critical Appraisal (2018, background): “ are far more efficient in learning complex rules than deep learning systems are (Lake, Salakhutdinov, & Tenenbaum, 2015; Lake, Ullman, Tenenbaum, & Gershman, 2016). (See also related work by George et al (2017), and my own work with Steven Pinker on children’s overregularization errors in c”
Knowledge Graph Completion: A Review (2020, background): “So the above methods are greatly limited in the aspect of knowledge graph completion; Secondly, the essence of knowledge graph data is a kind of semantic network, in which entities and relationships contain rich semantic information [50].”
AOGNets: Deep AND-OR Grammar Networks for Visual Recognition (2017, background): “More recently, recursive cortical networks [11] have been proposed with much more data efficiency in learning which adopts the AND-OR grammar framework [46], showing great potential of grammar like structures in developing general AI systems.”
Learning to Model the Tail (2017, background): “The most related formulation is that of one/few-shot learning [43, 44, 45, 46, 47, 48, 21, 49, 35, 50, 23, 51, 52, 53, 54, 55, 56].”
SketchEmbedNet: Learning Novel Concepts by Imitating Drawings (2020, unknown): “(Snell et al., 2017; Lake et al., 2019) Conv4 Omniglot - - 86.30 - RCN (Supervised) (George et al., 2017; Lake et al., 2019) N/A Omniglot - - 92.70 - VHE (Supervised) (Hewitt et al., 2018; Lake et al., 2019) N/A Omniglot - - 81.30 -”
The Omniglot Challenge: A 3-Year Progress Report (2019, methodology): “…generative models have recently been applied to one-shot classification, including the neural statistician (12% error betweenalphabets; Edwards and Storkey, 2016) and recursive cortical networks (RCNs), a more explicitly compositional architecture (7.3% error within-alphabets; George et al., 2017).”
Faster Attend-Infer-Repeat with Tractable Probabilistic Models (2019, background): “More generally, other structured probabilistic models that do not address scene understanding (Lake et al., 2015; George et al., 2017) may also benefit from tractable components.”
Learning Physical Graph Representations from Visual Scenes (2020, background): “Moreover, models with both local and long-range recurrence are substantially more parameter-efﬁcient than feedforward CNNs on a variety of perceptual grouping tasks, which suggested that multiple forms of recurrence could provide useful inductive biases for scene decomposition [23, 32, 33, 14].”
Small Sample Learning in Big Data Era (2018, background): “Given the same character “A” in a wide variety of appearances shape, RCN (George et al., 2017) can parse “A” with contours and surfaces, scene context and background with lateral connections, and achieve higher accuracy compared with CNN with fewer samples.”
Combining Compositional Models and Deep Networks For Robust Object Classification under Occlusion (2019, methodology): “In contrast to deep models, compositional models have been shown to be robust to partial occlusion [7, 10], even if they have not seen partially occluded objects during training [19, 23].”
Compositional Convolutional Neural Networks: A Deep Architecture With Innate Robustness to Partial Occlusion (2020, methodology): “Such models have been successfully applied for detecting partially occluded object parts [29, 37] and for recognizing 2D patterns under partial occlusion [10, 15].”