For more info: Learning Deep Architectures for AI
This seminar was presented by Ash. The summary of this seminar is as follows.
***Peeking Into The Seminar***
Learning Deep Architectures for AI
When people begin learning, they become aware of the most simple and complex concepts are learned step by step. When this occurs, the information is processed in the brain through interactions between a number of neurons which can be described as a myriad of layers. Using this concept, engineers can proceed sequentially in the treatment of their work. The deep learning technique whichis a the same processing method used in the human brain. In contrast to prior trends using shallow methods, deep learning techniques learn more complex concepts through each layer by performing the formation of deep structures like in the brain. As an example, the simplest layer just recognizes an image’s pixels, the next layer recognizes only edges, part of the face (object), and the whole face(object) in that order. As this process works automatically, it is possible to process things quickly without a human’s manual intervention. To realize this technique, deep learning structures should function as follows
It should be able to learn complex concepts
It should be able to learn more complex techniques with minimal human input
It should be able to learn a very large set of information
It should be able to learn from mostly unlabeled data
It should function well during unsupervised learning
Using borrowed neural network structures with higher-level concept RBM, it offers a more accurate, faster, and more convenient technique than the original neural network.