Course Address: https://class.coursera.org/ntumltwo-002
1. What are the motivations of neural networks (nnet)?
A single perceptron (Perceptron) model is simple, limited in capability and only linearly segmented. It is easy to implement logic and, or, non, and convex sets by combining the perceptual machine model, but it is not possible to achieve the XOR operation and the ability is limited. Multi-level perceptual machine (perceptrons) model, not only can realize the XOR, the function is more powerful.
2. The output of neural network can be any kind of classification, regression and logistic regression.
3. Activation function (conversion function) of neurons in neural network, Tanh (s) introduced in courseware
4. The final neural network model, the structure of the neural network, input, output, connection rights, thresholds, conversion functions.
5. How to find the connection right wij.
6.BP algorithm steps:
7. Optimization of neural networks. When there are multiple hidden layers, ein is basically not a convex function (Non-convex), the BP algorithm is likely to obtain the local optimal solution, rather than the global optimal solution. The following three experience practices can help solve the problem to some extent:
- The initial point selection weights of GD/SGD are larger than
- Regularization (regularization)
- Don't iterate too many times
Machine learning Techniques-neural Network (nnet)