DeSTIN (Deep Spatio-Temporal Inference Network), a compositional deep learning architecture, developed by the research group of  Itamar Arel, Principal Investigator for the Machine Intelligence Lab (MIL) & Networking Research Group and Assistant Professor in the Department of Electrical and Computer Engineering at The University of Tennessee at Knoxville,TN, USA.
Email: Этот адрес электронной почты защищён от спам-ботов. У вас должен быть включен JavaScript для просмотра.
Website of Deep Learning Architectures

Когнитивная архитектура DeSTIN разрабатывается под руководством Итамара Ареля (Itamar Arel), руководителя Лаборатории машинного интеллекта (the Machine Intelligence Lab & Networking Research Group) Отделения электротехники и вычислительной техники (Department of Electrical and Computer Engineering) университета штата Теннесси в Ноксвилле (University of Tennessee at Knoxville), Теннесси, США.

Tom Karnowski
Machine Intelligence Lab

Late in 2010 a working group was formed to turn the research code for DeSTIN, originally implemented and developed by Tom Karnowski as part of his PhD, into a quality open source project.

Общие сведения

A Deep SpatioTemporal Inference Network (DeSTIN) - a scalable deep learning architecture that relies on a combination of unsupervised learning and Bayesian inference. Dynamic pattern learning forms an inherent way of capturing complex spatiotemporal dependencies. Simulation results demonstrate the core capabilities of the proposed framework, particularly in the context of high-dimensional signal classification.

Deep Machine Learning (DML) architectures have recently emerged as promising biologically-inspired frameworks for effectively modeling complex signals. The efficacy of machine learning methods has a strong dependence on the features they are applied to. Currently, these features are mostly chosen and created using labor-intensive feature engineering.

The Deep SpatioTemporal Inference Network (DeSTIN) architecture (first introduced in [1]) comprises multiple instantiations of an identical cortical circuit, or node.
Each node is a parameterized model which learns by means of an unsupervised learning process. These nodes populate all layers of the hierarchy where every node attempts to capture the salient spatiotemporal regularities exhibited by patterns it is presented with. The nodes at the lowest layer of the hierarchy receive as input raw sensory data (e.g. pixels of an image) and continuously construct a belief state that attempts to characterize sequences observed. The inputs to all layers other than the first are the belief states of nodes at their corresponding lower layers.
The beliefs formed across the hierarchy are then used as rich features provided to a classifier or regression learner that can be trained using supervised learning.

This work was partially supported by the Intelligence Advanced Research Projects Activity (IARPA) via Army Research Office (ARO) agreement number W911NF-12-1-0017, and by NSF grant #CCF-1218492. The U.S.

Selected Publications
[1] I. Arel, D. Rose, R. Coop, "DeSTIN: A Scalable Deep Learning Architecture with Application to High-Dimensional Robust Pattern Recognition," in Proc. AAAI 2009 Fall Symposium on Biologically Inspired Cognitive Architectures (BICA), November, 2009 [pdf]
[2] I. Arel, D. Rose, T. Karnowski, "A Deep Learning Architecture Comprising Homogeneous Cortical Circuits for Scalable Spatiotemporal Pattern Inference," in Proc. NIPS 2009 Workshop on Deep Learning for Speech Recognition and Related Applications , December, 2009 [pdf]
[3] I. Arel, D. Rose, T. Karnowski, "Deep Machine Learning - A New Frontier in Artificial Intelligence Research," IEEE Computational Intelligence Magazine, Vol. 14, pp. 12-18, November, 2010 [pdf]
[4] S. Young, J. Liu, I. Arel, J. Holleman, "On the Impact of Approximate Computation in an Analog DeSTIN Architecture," IEEE Transactions on Neural Networks and Learning Networks, December, 2013 [pdf]

Itamar Arel presenting work on DeSTIN with function approximation at BICA 2011. Link to video