Artificial Neural Networks Overview

Artificial Neural Networks (ANNs)

• ANNs are simplified models of the brain
• A neuron is the basic processing unit
• Neurons transform inputs to outputs
• ANNs consist of many neurons performing desired functions
• Connected with weighted connections
• ANNs learn and generalize on their own.
• ANNs are used in various applications

Learning Objectives

• Describe ANNs
• Understand ANN business applications
• Understand how ANNs work conceptually
• Understand IBM Watson
• Explain advantages and disadvantages of ANNs

What is an ANN?

• Extremely simplified model of the brain
• Neuron is the basic processing unit of the network
• Input data is transformed into output data optimally
• Composed of many performing neurons
• Many weighted connections between neurons
• Learn and perform tasks on their own
• Strong ability to learn and generalize

The Perceptron

• A simplified neuron model with inputs, weights, and an output
• Weights multiply inputs
• Inputs are summed, and a function (activation function) is applied to determine the final output

Design Principles

• Forward Propagation: Input signals flow forward through the network, layer by layer, to produce an output
• Backward Propagation: Error signals are backpropagated through the network to adjust the weights to improve the output.

How ANNs Work

• Output is a function of the weighted sum of inputs plus a bias
• Activation function is applied to the input
• Common activation function is the sigmoid function (smooth, continuous, monotonically increasing)
• Weights are very important in determining the function
• Weights adjusted by a method called backpropagation
• NN trains by adjusting weights based on feedback from previous decisions

Architecting a Neural Network

• Various architectures possible, each with flexibility
• Feedforward, multi-layered perceptron with back-propagation algorithms are the most popular

Case Study: IBM Watson

• AI platform for businesses using machine learning
• Sophisticated machine learning technique (Deep Learning)
• Improves quality and accuracy of results
• Handles unstructured data (video, audio, images)
• Examples include comparing damaged car images, natural language understanding for different industries, and three level models (basic, industry specific, company specific)

Business Applications of ANNs

• Used with numerous data and complex objective functions expected to improve over time
• Classification
• Pattern recognition, feature extraction
• Approving financial loan applications
• Noise reduction (ex. airline pilot voice and engine noise)
• Prediction and Extrapolation (ex. stock price predictions)

ANN Training to Learn

• Accuracy of ANN predictions improves over a long period of time with large training data sets
• Can become more accurate than human experts
• Deployment in real-time situations is a possibility

Developing an ANN

• Gather training and test data
• Choose network architecture (e.g., Feedforward)
• Select an algorithm (e.g., Multi-layer Perception)
• Set parameters
• Train with training data
• Validate with validation data
• Freeze weights/parameters
• Test with test data
• Deploy if accuracy is good

Advantages of ANNs

• Little restriction on use
• Deal with non-linear relationships
• Learn from examples without significant programming
• Handle various problem types (classification, clustering)
• Tolerant to data quality issues without strict assumptions
• Handle numerical and categorical variables
• Faster than other techniques
• Provide better than statistical counterparts (once trained sufficiently)

Disadvantages of ANNs

• "Black box" solutions; difficult to understand how results are reached
• Optimal design still an art; requires expertise and testing
• Can be difficult to handle large numbers of variables
• Requires large datasets for effective training

Dive into the world of Artificial Neural Networks (ANNs) with this quiz. Explore the basic components such as neurons, their functions, and how they learn and generalize from data. Understand the practical applications of ANNs in business and the advantages and disadvantages of implementing these models.