Introducing Neural NetworksIntroducing Neural Networks[Note: if there are questions, suggestions or discoveries about this lab that will be of general use to the class, I will add them to this page - so feel free to check out this page at the course web site in case there are helpful tips.]
This page updated 10/20/97, 10:00am
Answer to a question about how to run the exercises in the Word Superiority Effect part of the lab.
Click here to load BrainWave simulator from the Linguistics Department server: you can run the BrainWave simulator from any location with internet access (e.g. all university computer sites), provided you are using a Java-capable browser such as Netscape 3 or Internet Explorer 3. Note: if BrainWave loads quickly but you find nothing appearing under the Networks menu, try restarting your browser and reloading BrainWave. Note: if you're using Netscape on a Mac, I strongly suggest using version 3 (Navigator) rather than version 4 (Communicator) ... I have had much more success with version 3.
Click here to open the BrainWave manuals in color (from Australia, but loads fairly quickly).
Neural networks, parallel distributed processing (PDP), associative memory, interactive constraint-based processing, statistical learning ... these approaches have become extremely popular in recent psycholinguistic research, and they are all based upon a view of mental representation and computation which is rather different from the symbolic representations and computation of "classical" linguistics and cognitive science. Moreover, they rely upon ideas which are highly intuitive at a very general level, but often forbiddingly technical in their implementation. Therefore, the goal of this first lab exercise is to give you the opportunity to get some hands-on experience with building and modifying "toy" neural networks, so that you will be better able to understand and evaluate models using neural networks. Fortunately, a group at the University of Queensland (Australia) has developed BrainWave, an easy-to-use neural network simulator with accompanying step-by-step tutorials that guide you through some of the leading concepts of neural nets.
Note: neural nets can seem quite dauntingly technical at first. The goal of this lab exercise is most definitely not to drown you in mathematics or programming technicalia. The goal is to make you more familiar with the basic ideas behind neural network models in cognitive science. Although there is some heavy duty calculus in the manual for the lab, you will not need to digest this in order to do the exercises. [If you really want to understand this material though, for future reference, then it might be worth a try.] Don't panic. If you are having trouble, contact a classmate or contact me for help (email is the best way to reach me: colin@udel.edu). To head off anxiety, you should start work early on the lab, to ensure that you can get the simulator running and that you are able to work through the basic instructions.
Most of the credit for this lab will be given for evidence that you have made a good-faith attempt to work through and answer all the assigned exercises. Although you should hand in an individual lab-report, group work is both acceptable and encouraged.
The BrainWave simulator is written in Java, an operating-system independent programming language which is widely used in programs that run in World Wide Web pages. Therefore you should be able to run it on MacOS, Windows or Unix computers, whichever is your preference. If you want to run the simulator in a computer lab or on a computer at home, you can download the program from the BrainWave homepage in Australia (about 400k to download), or you can get it from me on a floppy disk. In addition, I will install the files in a folder called BrainWave 1.1 on the Macintosh computers in the Linguistics Department Lab in #42 East Delaware Ave, room #102. You can borrow a key to this room from Jane Creswell in the Linguistics Department office.
You can run BrainWave in a web browser that supports Java. In Netscape 3.0 run the command Open File (in Browser, if using Netscape Gold) from the File menu, and select the file BrainWave.html in the folder BrainWave 1.1 in the dialog box that appears. This will launch, after a short delay, the BrainWave simulator in a new Java window. The same commands will work to launch BrainWave in Internet Explorer 3.0 or later. [Note: if you are using Internet Explorer (at least on a Mac) then you will need to (i) open the Preferences dialog from the toolbar, and open the Java options under Web Browser; in the "Java Virtual Machine" menu, make sure that the option "Microsoft JIT VM" is not selected - you should choose either "Apple VM" or "Microsoft VM" ... why - I don't know! (ii) you may be surprised to find that the BrainWave menus do not appear in the Java window - for some reason they appear instead among the menus at the top of your screen!]
Note: you should not expect to be able to save any networks that you create if you are running BrainWave from a web browser on a Mac or Windows computer. Nor will you need to.
Now that you're mentally prepared and you have BrainWave up and running, you can start working through the examples in the manual. You can use what I handed out in class, but it is actually better to access the pages from the BrainWave homepage, because there all of the diagrams and screenshots are in color, which makes them easier to understand.
You will work through most of the non-technical examples in the following chapters:
The first two chapters will introduce you to how the BrainWave simulator works, and then walk you through a couple of very simple network models. The second two chapters deal with two problems that are more directly relevant to problems in psycholinguistics: perception from degraded input, and interaction between levels of representation.
You should work through the manual, doing the exercises listed, and writing-up short answers to the exercises in boldface. Mathematical detail not required in the answers: just show that you have got the hang of what's going on. Technical sections of the manual (marked by a * in the table of contents) can be skimmed or skipped).
Your lab-report is due on Thursday September 11th, but you should start working through the examples much earlier, in order to ensure that you can get it up and running. Remember: if you're having trouble, contact me or ask a classmate.