Artificial Neural Networks Chapter 8 PDF

Artificial Neural Networks Chapter 8 Learning Objectives Describe Artificial Neural Networks (ANNs) Business applications of ANNs Understand how ANNs work at a conceptual level What is IBM Watson? Advantages and Disadvantages of ANNs What is a Artificial Neural Network (ANN)?...

Artificial Neural Networks Chapter 8 Learning Objectives Describe Artificial Neural Networks (ANNs) Business applications of ANNs Understand how ANNs work at a conceptual level What is IBM Watson? Advantages and Disadvantages of ANNs What is a Artificial Neural Network (ANN)? An extremely simplified model of the brain Neuron is the basic processing unit of the network. Transforms inputs into outputs to the best of its ability They are composed of many neurons that perform the desired functions A large number of weighted connections between the elements They figure out how to perform their function on their own Their ability to learn and generalize is really strong The Perceptron https://towardsdatascience.com/deep-learn ing-for-nlp-anns-rnns-and-lstms-explained-9 Design Principles of an ANN Forward Propagation Backward Propagation How does it Work The output of the neuron is a function of the weighted sum of the inputs plus a bias Activation function is applied to the input which is the hidden layer also called the black box Most common activation function is the sigmoid function Smooth, continuous, and monotonically increasing Weights are very important in determining the function Weights are adjusted by a method called backpropagation Train the neural network with the data over and over again with many training cases NN continues to learn by adjusting the weights based on the feedback about its previous decisions Architecting a Neural Network There are many ways to architect the functioning of an ANN using fairly simple and open rules with a tremendous amount of flexibility at each stage. The most popular architecture is a Feed-forward, multi- layered perceptron with back-propagation learning algorithm. There are multiple layers of PEs in the system and the output of neurons are fed forward to the PEs in the next layers; and the feedback on the prediction is fed back into the neural network for learning to occur. Case Study: IBM Watson- Analytics in Medicine It is an AI Platform for Businesses Machine learning a subset of AI that uses computer to analyze data and make intelligent decisions based on what they learn Watson uses sophisticated Machine Learning technique called Deep Learning. It layers algorithms to create ANN that continuously learns on the job, constantly improving quality and accuracy of the results Watson easily deals with unstructured data (video, audio and image files) Insurance company compare the images of a damaged car with the undamaged car Natural language understanding capabilities IBM Watson understands the language of your industry using transfer learning Three layer mode (basic + industry specific + company specific) https://www.youtube.com/watch?v=ZPXCF5e1_HI https://natural-language-understanding-demo.ng.bluemix.net/ http://fortune.com/2016/11/02/ibm-watson-cancer-demo-brainstorm-health/ Business Applications of ANN Used when there is a lot of data availability and the objective function is complex and the model is expected to improve over time Classification (classify information) Pattern recognition, feature extraction, image matching Approving a financial loan application Noise Reduction Recognize patterns in the inputs and produce noiseless outputs Airline pilot voice and engine noise example Prediction (predict outcomes) Extrapolation based on historical data Stock price prediction Artificial Neural Network: Training to Learn Depending upon the nature of the problem and the availability of good training data, at some point the neural network will learn enough and begin to match the predictive accuracy of a human expert. The predictions of ANN, trained over a long period of time with a large number of training data, have begun to decisively become more accurate than human experts. At that point ANN can begin to be seriously considered for deployment in real situations in real time. Developing an ANN Gather data. Divide into training data and test data. Select the network architecture, such as Feedforward network. Select the algorithm, such as Multi-layer Perception. Set various parameters. Train the ANN with training data. Validate the model with validation data. Freeze the weights and other parameters. Test the trained network with test data. Deploy the ANN when it achieves good predictive accuracy. Advantages of using ANN’s ANNs impose very little restrictions on their use. ANNs can deal with highly nonlinear relationships on their own, without much work from the user or analyst. There is no need to program neural networks, as they learn from examples. They get better with use, without much programming effort. They can handle a variety of problem types, including classification, clustering, associations, etc. ANN are tolerant of data quality issues and they do not restrict the data to follow strict normality and/or independence assumptions. They can handle both numerical and categorical variables. ANNs are much faster than other techniques. They usually provide better results (prediction and/or clustering) compared to statistical counterparts, once they have been trained enough. Disadvantages of using ANN’s They are deemed to be black-box solutions, sometimes lacking how the solution is arrived. Thus they are difficult to communicate about, except through the strength of their results. Optimal design of ANN is still an art: it requires expertise and extensive experimentation. It can be difficult to handle a large number of variables. It takes large data sets to train an ANN