Soft Computing Lecture 21 Review of using of

Soft Computing Lecture 21 Review of using of NN for solving of real tasks 07. 12. 2005

Autopilot airplane • Lo. FLYTE (Low-Observable Flight Test Experiment) is developed for NASA and USA Air Force by Accurate Automation Corp. , Chattanooga, TN. • Speed is 4 -5 M • NN is learning by pilot (storing associations between situation and action of pilot) and after that may control without pilot 07. 12. 2005 2

Project TNA • System for searching of plastic explosive in baggage in airports is developed by SAIC (Science Application International Corporation) • NN analyzes of spectrum of baggage after irradiation of it by slow neutrons • Recognition of explosive with probability 97%, speed is 10 units per minute 07. 12. 2005 3

Using of NN on financial markets • Citibank uses NN from 1990. In 1992 yield was 25% which is large more then most of brokers • Chemical Bank uses NN (from company Neural Data) for previous processing of transactions in currency exchanges in 23 countries for detection of shady bargains • Fidelity of Boston uses NN for control of portfolio with volume 3 billion USA, Deere & Co – 100 million USA, LBS Capital – 400 million USA • Proceedings of one seminar “AI in Wall Street” includes 6 large volumes. 07. 12. 2005 4

Recognition of stolen credit cards • Developer is HNC Software Corp. , now Fair Isaac Corporation http: //www. fairisaac. com • Software Falcon based on NN • This company controls more then 220 million accounts • NN is learning to recognize unusual behavior of clients with credit card 07. 12. 2005 5

Active reclaim in Internet • Developer is Aptex Software Inc. • Software Select. Cast based on NN • NN is learning by interesting of users and offers to client such reclaim which may be interesting for him 07. 12. 2005 6

Control of mobile robots • LSTM for robots • Planning of path and navigation • Recognition of objects 07. 12. 2005 7

Camera-robot coordination is function approximation • The system we focus on in this section is a work floor observed by a fixed cameras and a robot arm. The visual system must identify the target as well as determine the visual position of the end-effector. 07. 12. 2005 8

Camera-robot coordination is function approximation (2) 07. 12. 2005 9

Camera-robot coordination is function approximation (3). Two approach to use neural networks: • Usage of feed-forward networks – Indirect learning – General learning – Specialized learning • Usage of topology conserving maps 07. 12. 2005 10

Camera-robot coordination is function approximation (4). feed-forward networks Indirect learning system for robotics. In each cycle the network is used in two different places: first in the forward step then for feeding back the error 07. 12. 2005 11

Camera-robot coordination is function approximation (5). feed-forward networks (2) 07. 12. 2005 12

Sensor based control 07. 12. 2005 13

The structure of the network for the autonomous land vehicle 07. 12. 2005 14

Drama 07. 12. 2005 15

Diagnosis of cancer • Yulei Jiang, assistant professor of radiology at the University of Chicago • System that uses a perceptron neural network to analyze eight input nodes, converting the output node into a robability measure, which Jiang calls likelihood of malignancy. • The system is trained on a set of cases, using the leaveone-out method, which is a cross-validation technique for estimating generalization error based on resampling. A net is trained a number of times, each time leaving out one of the subsets. The omitted subset computes whatever error criterion is of interest. The system initially learns from digitized screen mammograms. • The eight input nodes represent features of calcifications, areas in breast tissue where tiny calcium deposits build up and might indicate the presence of cancer. 07. 12. 2005 16

07. 12. 2005 17

07. 12. 2005 18