AnalogML

A project looking to implement a fully connected feed-forward artificial neural network completely in analog hardware. Part of Columbia ELENE3390, Electrical Engineering Senior Capstone Projects

Best pre-finetuning model -- Train Loss: 0.7688, Val Loss: 0.2113, Val Accuracy: 100.00% Best finetuning model -- Train Loss: 0.4129, Val Loss: 0.1293, Val Accuracy: 100.00%

Modeling the Network in the Digital Domain

A 10x10 matrix created by converting a bitmap of the letter A to a binary matrix:

[[0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0]
 [0 0 1 1 1 1 0 0 0 0]
 [0 0 0 1 1 1 0 0 0 0]
 [0 0 0 1 0 1 0 0 0 0]
 [0 0 1 1 1 1 1 0 0 0]
 [0 0 1 1 0 1 1 0 0 0]
 [0 0 1 1 0 1 1 1 0 0]
 [0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0]]

And scaled to a 20x20 matrix:

[[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0]
 [0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 1 1 1 1 1 1 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 1 1 1 1 1 1 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 1 1 0 0 1 1 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 1 1 0 0 1 1 0 0 0 0 0 0 0 0]
 [0 0 0 0 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0]
 [0 0 0 0 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0]
 [0 0 0 0 1 1 1 1 0 0 1 1 1 1 0 0 0 0 0 0]
 [0 0 0 0 1 1 1 1 0 0 1 1 1 1 0 0 0 0 0 0]
 [0 0 0 0 1 1 1 1 0 0 1 1 1 1 1 1 0 0 0 0]
 [0 0 0 0 1 1 1 1 0 0 1 1 1 1 1 1 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]]

10x10 Matrix, 100 Epochs, Batch Size 64

Hidden Layers	Training Loss	Validation Loss	Validation Accuracy
[2]	3.1931	3.3016	0.00%
[2,2]	3.2458	3.2458	0.00%
[4]	3.0239	2.8984	50.00%
[4,4]	3.1435	3.1451	0.00%
[8]	2.9434	2.8809	33.33%
[8,8]	3.1306	2.9969	33.33%
[16]	2.4503	2.3143	66.67%
[16,16]	2.5845	2.495	66.67%
[32]	2.0512	1.8625	100.00%
[32,32]	1.6095	1.5628	66.67%
[128,64]	0.1866	0.1601	100.00%