New Features of Release 4.1

Users already familiar with SNNS and its usage may be interested in the differences between the versions 4.0 and 4.1. New users of SNNS may skip this section and proceed with the next chapter.

New Features of Release 4.1:

1. New batch execution language batchman

2. Integration of the genetic algorithm tool Knete. Knete is available as seperate tar file on our ftp-server at the same location as SNNS. For all enquiries about Knete please contact enzo-request@ira.uka.de.

3. Improved window handling. On re-opening of a SNNS window it gets put on top, instead of issuing an error message.

4. new learning algorithm 'scaled conjugated gradient' (SCG)

5. The parameter lists in the control panel adapt to the number of paramters required, i.e. all widgets present need to be filled out.

6. WinSNNS, a MS-Windows front-end to SNNS batch execution on unix workstations. WinSNNS is NOT part of our official SNNS distribution! It can be found at http://www.lans.ece.utexas.edu/winsnns.html.
   Please contact ghosh@pine.ece.utexas.edu for more information.

7. optional parameter `Teacher-Forcing' for Jordan networks introduced.

8. validation now possible for BPTT networks

9. Extensive debugging:
   1. Seg-fault when trying to train a network without output units

   2. Seg-fault when pressing the info button without loaded patterns

   3. wrong feedback loops in associative memory nets deleted

   4. wrong activation functions in associative memory when constructed with bignet

   5. wrong initialization function in kohonen networks when constructed with bignet

   6. Wrong error computation in Hebbian learning

   7. Wrong error computation of analyze function band

   8. Wrong MSE computation for validation

   9. Better font size adaption of label widgets

   10. Pruning now allows validation with 0 cycles

   11. GRAPH panel could sometimes not be closed

   12. Wrong scanning of update parameters in snnsbat

   13. Error in the manual: Missing parameter of BackpropWeightDecay