 mandlm
|
5e32724b1a
|
Layers average their output sums
|
2015-11-15 16:08:49 +01:00 |
|
mandlm
|
8d01edb7a1
|
Removed naked double * calls
|
2015-10-31 14:59:10 +01:00 |
|
mandlm
|
650b4be9fc
|
Added forward-feeding of double-ptrs
|
2015-10-27 15:33:10 +01:00 |
|
mandlm
|
81d1f54c98
|
Started loader-class for 8x8 pixel digit training data
|
2015-10-26 22:05:50 +01:00 |
|
mandlm
|
37b5153d6a
|
Cleaned up code to work with MinGW/GCC Kit
|
2015-10-26 20:34:26 +01:00 |
|
mandlm
|
b899c6f55e
|
Replaced a few int with size_t to fix 64 bit warnings.
|
2015-10-26 07:33:45 +01:00 |
|
mandlm
|
99ef63e019
|
Added simple (de-)serialization of (trained) nets
|
2015-10-25 17:40:22 +01:00 |
|
mandlm
|
6943fc0116
|
Threaded learning and signaling in the Qt UI
|
2015-10-24 18:03:07 +02:00 |
|
mandlm
|
4eb232b1e9
|
Working net, calculates the mean value of two inputs
|
2015-10-23 22:16:12 +02:00 |
|
mandlm
|
6ed30e56c4
|
Finished the max-value net (2/3/1 neurons) with 10k learning iterations. No good.
|
2015-10-22 22:09:35 +02:00 |
|
mandlm
|
f22e4221a1
|
Fixed feed-forward algo
|
2015-10-22 16:02:27 +02:00 |
|
mandlm
|
3d30346f2d
|
Implemented dynamic learning
|
2015-10-18 22:05:18 +02:00 |
|
mandlm
|
6ef1f9657c
|
Backprop seems to be working, yay
|
2015-10-18 21:20:37 +02:00 |
|
mandlm
|
a79abb5db1
|
First implementation of weight updates. Very slow rate of change in the output value.
|
2015-10-17 22:05:27 +02:00 |
|
mandlm
|
370451c2e6
|
Calculation of hidden neuron gradients (partial)
|
2015-10-16 22:59:04 +02:00 |
|
mandlm
|
ea454b58c6
|
Added 64-bit configuration and support
|
2015-10-15 22:37:13 +02:00 |
|
mandlm
|
7ba16e9e9d
|
Renamed a few things, started working on back-propagation
|
2015-10-15 19:18:26 +02:00 |
|
mandlm
|
2f556d1b92
|
Added a (hacky) bias neuron
|
2015-03-24 13:45:38 +01:00 |
|
mandlm
|
e3a804242c
|
Split up to different source files, entry-point for back propagation
|
2015-03-23 21:58:30 +01:00 |