Assignment 2

Well finished 20.1... took me all night... well at least Ive begun.

 
  ,= ,-_-. =.
 ((_/)o o(\_))
  `-'(. .)`-'
      \_/

http://ilanpillemer.com
Entia non sunt multiplicanda praeter necessitatem

I've done 1/2 of 20.1, but I am struggling with the generation of the Fig 20.1 (b) graphs.

I just can't figure out how you would go from equation 20.2 to Fig 20.1 (b)?

Does anyone know how the first 5 or so points are calculated in Fig 20.1 (b)?

This question is driving me crazy!

I answered you on other thread. I have left out the MAP question and done some of the others. Not finished yet though.

 
  ,= ,-_-. =.
 ((_/)o o(\_))
  `-'(. .)`-'
      \_/

http://ilanpillemer.com
Entia non sunt multiplicanda praeter necessitatem

Thanks for that! I have been stubbornly fighting this question and will now have to zip through all the rest of the assignment

Well I just spent all night on Question 20.18.

I am using Java and finally got the feed forward working now at 12:20 am.
I now need to add in the learning (backward propagation... thats for Thursday night.)

 
  ,= ,-_-. =.
 ((_/)o o(\_))
  `-'(. .)`-'
      \_/

http://ilanpillemer.com
Entia non sunt multiplicanda praeter necessitatem

awrghhhhh,..

I finally got my NN with one hidden layer learning XOR. There is very little chance I will complete either of the Chap 21 questions at this rate.

sigh..

Well at least I got my neural network learning... at midnight it felt hopeless..

Quote
sample run
[ilan@localhost ff]$ java Network
Training Neural Network....
Training: XOR
Final Weights
Hidden Node: 3: 1.0
Hidden Node: 4: 1.0
Weight of Input13: 0.09999999999999981
Weight of Input14: -0.20000000000000026
Weight of Input23: -0.10000000000000014
Weight of Input24: 0.09999999999999987
Epochs taken to learn the network: 13
Neural Network trained...
Calling NEURAL-NETWORK-OUTPUT with examples....
input 1: 1.0 input 2: 1.0 NN result: 0.0
input 1: 0.0 input 2: 1.0 NN result: 1.0
input 1: 1.0 input 2: 0.0 NN result: 1.0
input 1: 0.0 input 2: 0.0 NN result: 0.0
[ilan@localhost ff]$

 
  ,= ,-_-. =.
 ((_/)o o(\_))
  `-'(. .)`-'
      \_/

http://ilanpillemer.com
Entia non sunt multiplicanda praeter necessitatem

I am just about to give up on this subject. We have to do all 3 assignments and I am battling with the questions from chapter 21 and I very much doubt my answers for chapter 20 is any good.My approach for 18 was to show the basic structure and leave it at that, hopefully get at least some points. Wish I at least had the results from assignment 1 so I know if I should even bother completing this assignment

Well I found assignment 1 real easy, except for the last question. This stuff had been covered already in COS351.

I really struggled with this assignment, and I left out the chapter 21 questions altogether. I also don't think many of my answers for the other questions were very good. But I think I struggled enough with it; that I will be able to learn from the model answer when it eventually arrives.

Assignment 3 doesn't look that hard; its just a little reading and you do a summary. I must admit I also thought of dropping this one... but I find the subject itself very fascinating... and I am pinning my hopes on the model answer being well written.

 
  ,= ,-_-. =.
 ((_/)o o(\_))
  `-'(. .)`-'
      \_/

http://ilanpillemer.com
Entia non sunt multiplicanda praeter necessitatem

This subject and COS451 was the only reasons I decided to do honors because I am fascinated by AI. I spend a lot of time on the answers from chapter 20 too so my hopes are on the model answers. I think my biggest problem is that the changed the syllabus on COS351 since I did it and now this builds on stuff I have never done. I am not to sure what we need to do to get into the exam, it only says you have to hand in 3 assignments but I am not to sure that that would be it. I have now started reading the text book from chapter 1 to try and see if it wont make more sense...

Just hope we get some marks back before the next assignment

I also am fascinated by AI.. I think its definitely going to be my research direction into masters and beyond.

Im not too stressed about when the marks come back as no assignment seems to depend on the material in a different assignment.

As long as its back in time for my exam studying.

 
  ,= ,-_-. =.
 ((_/)o o(\_))
  `-'(. .)`-'
      \_/

http://ilanpillemer.com
Entia non sunt multiplicanda praeter necessitatem

Quote

I think my biggest problem is that the changed the syllabus on COS351 since I did it and now this builds on stuff I have never done.

Quote

I have now started reading the text book from chapter 1 to try and see if it wont make more sense...

I think I am also going to need to do the same.

I also think the textbook is crap though(At least the later parts can't comment on the earleir part yet), that stuff for question 20.1 for example is definitely not explained in an adequate way IMHO,

--
"Knowledge has much better uses than self-pity and superiority"

From the first couple of chapters it actually doesn't seem to bad. I think it might be a problem of jumping into the text book towards the end instead of working though it systematically. I do however still think this is not the ideal book for a correspondace degree as in my opinion they don't do a very good job of explaining the work. I found the explanations on neural networks on the internet to be much better than the text book so as I am working through it I am researching on the web for everything I dont understand from the text book. But this is taking so much time, and I still got another 5 subjects to work on. Strangely though I do not seem to be having nearly as much trouble with the other ones except for some of them getting results back takes about 2 months as well