Practice building different deep neural net models starting with simple problems
April 29th update: -Created a simple NeuralNet framework using the Numpy library:
- Enables creation of deep neural networks
- User can specify the depth of the network and the width of each layer
- XOR_test uses NeuralNet to attempt to learn the XOR function
- Convergence is observed for 2 bit and 3 bit XOR functions
-To see the framework in action:
- put generate_data.py, NeuralNet.py, and XOR_Test.py in the same directory
- run the XOR_Test module, the TrainingError vs Iterations graph will appear
-Known problems:
- The network does not always converge the where you want!!
- The result of convergence is highly sensitive to the initial weights
- Oscillation in the training error is observed for some learning rates