sfp@stc06.ornl.gov (SPELT P F) (11/10/90)
From: Phil Spelt, Oak Ridge National Laboratory Oak Ridge, TN 37831-6364 USA About a month ago, I posted in this BB a request for feedback from people using Artificial Neural Networks (ANNs) in the area of Human Factors research. I promised, at that time, to post a summary of the replies. Below are excerpts from messages from those who responded to that posting. The overall result is that there will be a paper session at the 2nd Neural Network Workshop at Auburn University, 11-13 FEbruary 1991. There will also be a roundtable discussion about the uses of ANNs in Human Factors research. A summary of all this will be published in the _Bulletin_ of the Human Factors Society sometime in mid 1991, and the papers wil appear in the Proceedings of the Workshop. My thanks to all who responded to the original posting. Phil Spelt ----------=========>>>>>Summaries: From: NAME: Ray Eberts <eberts%ntthif.ntt.jp@relay.cs.net@UMCGATE@OAX> To: sfp%stc06.ctd.ornl.gov@relay.cs.net@UMCGATE@OAX I am responding to your newsgroup announcement about human factors and neural nets. I have been working on several applications (I tried to kill a previous message I sent you, I hope it got killed, because I don't wish to divulge all my research at this time until I can get some of it out). Let me only say, that I am working on using neural nets for user interface design, modeling, and as simulations to investigate training procedures. I have recently completed a survey of neural net research and human factors for the Handbook on Industrial Engineering chapter on human Information Processing. There is very little research in the journals at this point. About the only things available are Schneider and Detweiler in Human Factors on training for dual task skills and the Sejnowski and Rosenberg work on massed vs. distributed training. Some work has also been done on knowledge extraction which could be a human factors domain. If my other message got through, please do not send out that information to anybody else at this time. I can be reached at the following address: Ray Eberts School of Industrial Engineering Purdue University West Lafayette, IN 47907 I am currently on sabbatical in Japan until December at which time I will go off to USC for a semester. My e-mail address in Japan is: eberts%ntthif.ntt.jp@relay.cs.net At Purdue, my e-mail address is: eberts@rainbow.ecn.purdue.edu ------------------------------------------------------------------ From: NAME: Steve Bryan <sxb@inel.gov@UMCGATE@OAX> I have not yet considered the use of neural networks in human factors work, but am intrigued with the possibilities. Please email a copy of the responses that you get, if it would not be too much trouble. [I asked Steve what kind of work he is doing -- NN's or HF. PFS] I am doing neural networks research, but am always on the lookout for interesting applications. I have applied neural networks to control problems, and used them as pattern classifiers as parts of larger projects. Human factors is a pretty broad area for neural network application, but I have seen no publications in neural networks as applied to human factors per se. Did you have computer interfaces that learn individual preferences in mind, or another application? -Steve -- ____________________________________________________________________________ |Steven R. Bryan, Idaho National Engineering Lab. INTERNET: sxb@INEL.GOV | |N7MPY POB 1625 M.S. 1206 Phone: (208) 526-0338 | |_________________Idaho_Falls,_Id._83415___________FAX:___(208)_526-1419_____| ------------------------------------------------------------------------------ From: jfmcgrew@srv.pacbell.com@UMCGATE@OAX I work for the Advanced System Group at Pacific Bell. We build expert systems for inhouse use. I am one of the two human factors specticialists in the group of 10. I do the user interface design and knowledge engineering. I have found that the determining of the structure and content of the users tast and knowledge engineering to have extensive overlap. Because of this I am very interested in the subject of human factors and neural nets. We at present do not use neural nets in our human factors work, although two areas are presently bumping against the technology. These areas are the use of graph theory and matricies to determine the underlying structure of a users knowledge and tasks, and mathimatical group theory to model the users view of objects (this work is in very early stages of develop ment). I would like to know the outcome of your survey and to be kept on any mailing list related to the subject that you develope. Thanks! John F. McGrew ---------------------------------------------------------------------------- From: NAME: Yong Kim <yjkim@milton.u.washington.edu@UMCGATE@OAX> Can I get the reply? Thanks!! yjkim@milton.u.washington.edu ---------------------------------------------------------------------------- From: jones@nprdc.navy.mil@UMCGATE@OAX I currently have a small 6.2 grant to study the feasibility of applying neural net techniques to the analysis of cortical evoked potential data. The general theme of the division that I work in at the Navy Personnel R&D Center in San Diego is the prediction of job performance from EP data collected on cognitive tasks, with the primary application being in selection and classification. Tradition statistical techniques, such as discriminant analysis and multiple regression have been less than satisfactory, presumably because the relationship are complex and nonlinear. In general, psychometric tests are also not very good at predicting job performance in our sample of subjects (Marines). The initial thrust of this project is to predict rifle and pistol marksmanship from data collected from our subjects on several cognitive tasks. I don't know whether or not this would fit in with your planned workshop, but I would be interested in participating or discussing it further if you desire. My postal address is David L. Ryan-Jones, Navy Personnel R&D Center, Code 141, San Diego, CA 92152. David Ryan-Jones ---------------------------------------------------------------------------- From: lr@chmsr.gatech.edu@UMCGATE@OAX I am currently working on a visual perception problem in a dynamic environment using the back-propagation network. It involves simulating human perceptual learning through mimicking similar inputs into the network. I have done work in the neural nets field in the past, but the human factors area is relatively new to me. I hope to have some results available by the end of the fall quarter (i.e. around January). Ling Rothrock (Grad / Ga Tech / Human-Machine Systems) [ Also, the faculty contact at Ga. Tech is: Alex Kirlik: kirlik@chmsr.gatech.edu snail-mail: Center for Human-Machine Systems Research School of Industrial and Systems Engineering Ga. Inst. Tech. Atlanta, GA 30332-0205 PFS ] --------------------------------------------------------------------------- From: janet@minster.york.ac.uk@UMCGATE@OAX I've been using associative memories in particular in HCI for some time - two main (related) applications: automatic user modelling - training the memory on examples of user behaviour and so classifying user types for adaptive interface/intelligent help type applications; and using similar techniques to recognise patterns in protocols in order to aid evaluation. These two applications are reported in the papers listed below and are the subject of my DPhil thesis (recently submitted). I'll happily provide more information if you're interested. User Modelling By Classification: A Neural-Based Approach authors: R Beale, J Finlay, J Austin, M Harrison in: New Developments in Neural Computing editors: J G Taylor & C L T Mannion abstract: A neural-based system is used for a classification task in the field of user-modelling. The properties associated with neural methods, such as learning by example and generalisation are exploited to provide a system that overcomes the limitations of traditional approaches, removing the need for an extensive, explicit knowledge base, and exhiiting the desirable properties of domain indepedence, noise tolerance and resource efficiency." [The other references had no abstracts. I can provide them if needed. PFS] Janet Finlay, Human Computer Interaction Group, Dept. of Computer Science, University of York, Heslington, YORK. YO1 5DD. UK. Tel: [044] (0904) 432765 janet@uk.ac.york.minster JANET connexions janet%minster.york.ac.uk@nsfnet-relay.ac.uk ARPA connexions ..!mcsun!ukc!minster!janet UUCP connexions janet@minster.york.ac.uk eab mail ----------------------------------------------------------------------- From: NAME: Jim R.Chen <jchen@apple.com@UMCGATE@OAX> Two years ago I did a project with the Human Interface Group at Apple Computer on the use of neural network to aid MacIntosh desktop color selection. In that project I used back-propagation to recall input activations in a layered network which simulates an analog associative memory with asymmetric connections. That was reported in Chen, James R., Richard K. Belew, and Gitta B. Salomon, "A Connectionist Network for Color Selection", Proceedings of the International Joint Conference on Neural Networks, Jan. 1990, 467-470. I am currently working on a neural network hypertext interface at the Human Factors Center at IBM Santa Teresa Lab. The project is still at an early stage and I don't have much to report yet. Jim Chen chen@cs.ucsd.edu (408) 463-2469 (day) (408) 741-5348