| Yann LeCun, Silver Professor of Computer Science The Courant Institute of Mathematical Sciences
New York University Сайт Янна ЛеКуна, доктора компьютерных наук Нью-Йоркского университета
Разработки и исследования
Статьи и публикации Кто не читал - всем советую! Исключительные публикации по нейронным сетям Learning Algorithms, Generalization, and Regularization
I proposed a "pruning" (weight elimination) methods for neural networks called Optimal Brain Damage that has become quite popular over the years [LeCun, Denker, Solla, 1990].
Harris Drucker and I worked on a regularization method called double backpropagation that improves the smoothness of the functions learned by multilayer neural nets (and therefore their robustness to small distortions of the inputs) [Drucker, LeCun 1992] [Drucker, LeCun 1991a] [Drucker, LeCun 1991b].
Later, Patrice Simard and I (with J. Denker and B. Victorri) found a regularization method called Tangent Prop that can force neural nets to become invariant (or robust) to a set of chosen transformations of the inputs. [Simard, Victorri, LeCun, Denker 1992] [Simard, LeCun, Denker, Victorri 1992]. A recent review paper on Tangent Prop (and Tangent Distance) is available [Simard, LeCun, Denker, Victorri 1998]. Invariant Recognition: Convolutional Neural Networks
Since my PhD, I had been interested in the problem of invariant visual perception, and how learning methods could help solve it. I had been experimenting with simple locally-connected neural net architectures during my PhD, and showed during my PostDoc local connections and shared weights clearly improved the performance of neural nets on simple shape recognition tasks [LeCun, 1989b] [LeCun, 1989a].
Other Classification Algorithms: Tangent Distance, Boosting, SVM,....
Patrice Simard, John Denker and I worked on a method called Tangent Distance for measuring similarities between shapes while being robust to small distortions and displacements of the input pattern. We applied this method to handwriting recognition [Simard, LeCun, Denker 1993] [Simard, LeCun, Denker 1994]. A recent and rather complete paper on Tangent Prop and Tangent Distance is available [Simard, LeCun, Denker, Victorri 1998].
|
