I need help in completing my task

Sharda_dss11_im_06.doc

Chapter 6: Deep Learning and Cognitive Computing

Learning Objectives for Chapter 6

1. Learn what deep learning is and how it is changing the world of computing

2. Know the placement of deep learning within the broad family of artificial intelligence (AI) learning methods

3. Understand how traditional “shallow” artificial neural networks (ANN) work

4. Become familiar with the development and learning processes of ANN

5. Develop an understanding of the methods to shed light into the ANN black box

6. Know the underlying concept and methods for deep neural networks

7. Become familiar with different types of deep learning methods

8. Understand how convolutional neural networks (CNN) work

9. Learn how recurrent neural networks (RNN) and long short-memory networks (LSTM) work

10. Become familiar with the computer frameworks for implementing deep learning

11. Know the foundational details about cognitive computing

12. Learn how IBM Watson works and what types of application it can be used for

CHAPTER OVERVIEW

Artificial intelligence (AI) is making a re-entrance into the world of commuting and in our lives, this time far stronger and much more promising than before. This unprecedented re-emergence and the new level of expectations can largely be attributed to deep learning and cognitive computing. These two latest buzzwords define the leading edge of AI and machine learning today. Evolving out of the traditional artificial neural networks (ANN), deep learning is changing the very foundation of how machine learning works. Thanks to large collections of data and improved computational resources, deep learning is making a profound impact on how computers can discover complex patterns using the self-extracted features from the data (as opposed to a data scientist providing the feature vector to the learning algorithm). Cognitive computing— first popularized by IBM Watson and its success against the best human players in the game show Jeopardy!—makes it possible to deal with a new class of problems, the type of problems that are thought to be solvable only by human ingenuity and creativity, ones that are characterized by ambiguity and uncertainty. This chapter covers the concepts, methods, and application of these two cutting-edge AI technology trends.

CHAPTER OUTLINE

6.1 Opening Vignette: Fighting Fraud with Deep Learning and Artificial Intelligence

6.2 Introduction to Deep Learning

6.3 Basics of “Shallow” Neural Networks

6.4 Process of Developing Neural Network–Based Systems

6.5 Illuminating the Black Box of ANN

6.6 Deep Neural Networks 343 6.7 Convolutional Neural Networks

6.8 Recurrent Networks and Long Short-Term Memory Networks

6.9 Computer Frameworks for Implementation of Deep Learning

6.10 Cognitive Computing

ANSWERS TO END OF SECTION REVIEW QUESTIONS( ( ( ( ( (

Section 6.1 Review Questions

1. What is fraud in banking?

Fraud in banking is when a customer attempts to access an account that they do not have permission to access for the retrieval of funds or the mining of information.

2. What are the types of fraud that banking firms are facing today?

Student research and experience will vary. Some common types of fraud include stolen credit cards and fraudulent access to online accounts.

3. What do you think are the implications of fraud on banks and on their customers?

The implications of fraud are huge for banks and their customers because of the large monetary value involved and the importance of trust in the banking relationship.

4. Compare the old and new methods for identifying and mitigating fraud.

Old methods of fraud did not catch a significant number of the illegal activities and also generated a large number of false positives. New methods are significantly more effective and can balance the need for detection against excessive false positives.

5. Why do you think deep learning methods provided better prediction accuracy?

The system is able to provide better prediction accuracy because it is able to access and utilize a larger amount of data quickly and in real time, but also because the detection framework can evolve over time to become more accurate and recognize new types of fraud.

6. Discuss the trade-off between false positive and false negative (type 1 and type 2 errors) within the context of predicting fraudulent activities.

False negatives indicate that a fraud was able to be committed, whereas a false positive will create a customer service issue and inconvenience an account holder. Any system must try to balance between these two to ensure that fraud is not occurring, but at the same time the level of customer inconvenience is not so much that it would create a larger business issue.

Section 6.2 Review Questions

1. What is deep learning? What can deep learning do?

Deep learning, the newest branch of AI and machine-learning, attempts mimic the thought process of humans using mathematical algorithms to learn from data.

2. Compared to traditional machine learning, what is the most prominent difference of deep learning?

The performance of traditional machine-learning algorithms such as decision trees, support vector machines, logistic regression, and neural networks relies heavily on the representation of the data. What deep learning has added to the classic machine-learning methods is in fact the ability to automatically acquire the knowledge required to accomplish such informal tasks and consequently extract some advanced features that contribute to the superior system performance.

3. List and briefly explain different learning methods in AI.

These details are nicely summarized in figure 6.4. In knowledge-based systems, manually created representations provide the framework for evaluating information. In classic machine learning, manually created features provide the foundation for mapping of those features and then output. In generic representational learning features are created automatically and then mapping is performed from those features. In deep learning features are also created automatically but move beyond a single stage process to create more advanced features first before completing mapping functions.

4. What is representation learning, and how does it relate to deep learning?

Representation learning techniques entail one type of machine learning (which is also a part of AI) in which the emphasis is on learning and discovering features by the system in addition to discovering the mapping from those features to the output/target. Deep learning is a branch of representation learning.

Section 6.3 Review Questions

1. How does a single artificial neuron (i.e., PE) work?

This process is aided by referring to figure 6.5. Inputs enter the neuron and encounter the adjustable weight and bias. A calculation weight function is applied to the input as is a net input function. The output of the net input function then goes through another function called the transfer function for conversion and the production of actual output.

2. List and briefly describe the most commonly used ANN activation functions.

Most commonly used AAN and activation functions are:

· Hard limit - operation as a baseline but increases to a set point if the condition is met

· Linear - operation increases as condition increases

· Log-Sigmoid - operation increases on a log scale as condition increases

· Positive linear - operation has a baseline but then increases linearly after a condition is met

It is very helpful to refer to table 6.1

3. What is MLP, and how does it work?

MLP is multi- layer perception and represents neurons with multiple inputs as seen in figure 6.6. In this type of neuron each input would have its own adjustable weight.

4. Explain the function of weights in ANN.

Weights define the impact that a given input has on a neuron in the next layer. As such, they embody what the network has learned so far. As a network learns, its weights are adjusted.

5. Describe the summation and activation functions in MLP-type ANN architecture.

The summation function determines the total input to a neuron by calculating the weighted sum of its individual input values. Its output is input to the transformation (transfer) function. The transformation (or transfer) function determines the output of a neuron from its input (i.e., output of the summation function).

Many possible transformation functions exist. Most are sigmoid functions. (A sigmoid function is a non-decreasing function whose value is in the range 0–1.) A simple transformation function produces an output of 0 for inputs up to a threshold value, an output of 1 from that value up.

Section 6.4 Review Questions

1. List the nine steps in conducting a neural network project.

These are shown in the flowchart of Figure 6.9. After testing (step 8), it is possible to return to a previous step:

Step 1:Collect data

Step 2:Separate (data) into training and test sets

Step 3: Define a network structure

Step 4: Select a learning algorithm

Step 5:Set parameters and values

Step 6:Initialize weights and start training

Step 7:Stop training, freeze the network weights

Step 8:Test the trained network

Step 9:Implementation: use the network with new cases

2. What are some of the design parameters for developing a neural network?

Some design parameters to consider include the type of network to employ, the number of nodes (input, hidden, and output) and layers, the types of transformation functions within each neuron, the original weight settings, and the acceptable delta (error) level.

3. Draw and briefly explain the three-step process of learning in ANN.

In supervised learning, the learning process is inductive; that is, connection weights are derived from existing cases. The usual process of learning involves three tasks:

1. Compute temporary outputs.

2. Compare outputs with desired targets.

3. Adjust the weights and repeat the process.

Student drawings will closely resemble figure 6.12.

4. How does backpropagation learning work?

Backpropagation (short for back-error propagation) is the most widely used supervised learning algorithm in neural computing. Backpropagation involves a feedforward network with one or more hidden layers. Backpropagation involves supervised learning because predetermined correct answers for each training pattern are given and compared to the results given by the backpropagation algorithm. The difference between the correct answer and the output (known as the delta value) is calculated, and weights of the output nodes are adjusted accordingly. Weight adjustment also proceeds backward to the hidden layers; hence the term “backpropagation.” This process repeats, continuously testing against training patterns, until the delta value is sufficiently low.

5. What is overfitting in ANN learning? How does it happen, and how can it be mitigated?

Overfitting happens when the trained model is highly fitted to the training data set but performs poorly with regard to external data sets. A large group of strategies known as regularization strategies is designed to prevent models from overfitting by making changes or defining constraints for the model parameters or the performance function.

6. Describe the different types of neural network software available today.

Many commercial ANN software products function like software shells. They provide a set of standard architectures, learning algorithms, and parameters, along with the ability to manipulate the data. Some development tools can support up to several dozen network paradigms and learning algorithms. Most of the leading data mining tools (e.g., SAS Enterprise Miner, IBM SPSS Modeler, Statistica Data Miner) include neural network learning algorithms. Some specialized neural network products include California Scientific (BrainMaker), NeuralWare, NeuroDimension Inc., Ward Systems Group (Neuroshell), and Megaputer. Others are implemented as spreadsheet add-ins. In addition, there are class libraries and APIs for languages such as Java and C++. Mathematical applications such as MATLAB also include neural network algorithms.

Section 6.5 Review Questions

1. What is the so-called black-box syndrome?

ANN are typically thought of as black boxes, capable of solving complex problems but lacking the explanation of its capabilities. This phenomenon is commonly referred to as the “black-box” syndrome.

2. Why is it important to be able to explain an ANN’s model structure?

It is important to be able to explain a model’s “inner being”; such an explanation offers assurance that the network has been properly trained and will behave as desired once deployed in a business intelligence environment. Such a need to “look under the hood” might be attributable to a relatively small training set (as a result of the high cost of data acquisition) or a very high liability in case of a system error.

3. How does sensitivity analysis work in ANN?

Sensitivity analysis is a method for extracting the cause-and-effect relationships among the inputs and the outputs of a trained neural network model. In the process of performing sensitivity analysis, the trained neural network’s learning capability is disabled so that the network weights are not affected.

4. Search the Internet to find other methods to explain ANN methods. Report the result

Open-ended answer, depending on what a student finds.

Section 6.6 Review Questions

1. What is meant by “deep” in deep neural networks? Compare deep neural networks to shallow neural networks.

Deep neural networks have a large number of hidden layers in addition to a large number of neurons in each layer. This allows for a much more complex processing activity than was available in a shallow neural network and is supported by improvements in hardware technology.

2. What is GPU? How does it relate to deep neural networks?

A GPU is a graphical processing unit that is able to be used for AI purposes and allows neural networks with over 1 million neurons to function. They are significantly faster than traditional CPUs for this purpose.

3. How does a feedforward multilayer perceptron-type deep network work?

These networks are simply large-scale neural networks that can contain many layers of neurons and handle tensors as their input. These models are called feedforward because the flow of information that goes through them is always forwarding and no feedback connections (i.e., connections in which outputs of a model are fed back to itself) are allowed.

4. Comment on the impact of random weights in developing deep MLP.

Optimization of the performance (loss) function in many real applications of deep MLPs is a challenging issue. The problem is that applying the common gradient-based training algorithms with random initialization of weights and biases that is very efficient for finding the optimal set of parameters in shallow neural networks most of the time could lead to getting stuck in the locally optimal solutions rather than catching the global optimum values for the parameters. As the depth of network increases, chances of reaching a global optimum using random initializations with the gradient-based algorithms decrease.

5. Which strategy is better: more hidden layers versus more neurons?

Whereas theoretically it is still an open research question, practically using more layers in a network seems to be more effective and computationally more efficient than using many neurons in a few layers.

Section 6.7 Review Questions

1. What is CNN?

CNN is a convolutional neural network that includes at least one layer involving a convolutional weight function instead of the general matrix multiplication.

2. For what type of applications can CNN be used?

CNN was initially designed for computer vision applications but is also applicable to non-image data sets as well.

3. What is convolution function in CNN and how does it work?

The convolution function uses the concept of parameter sharing which in addition to computational efficiency allows for weights parameters to be shared thus reducing processing load in systems with a large number of weight parameters.

4. What is pooling in CNN? How does it work?

Generally, a convolution layer is followed by another layer known as the pooling layer. The purpose of a pooling layer is to consolidate elements in the input matrix to produce a smaller output matrix while maintaining the important features. Normally, a pooling function involves an r * c consolidation window that moves around the input matrix and in each move calculates some summary statistics of the elements involved in the consolidation window so that it can be put in the output image.

5. What is ImageNet and how does it relate to deep learning?

ImageNet is an ongoing research project that provides researchers with a large database of images, each linked to a set of synonym words from WordNet. ImageNet is a huge database for developing image processing–type deep networks.

6. What is the significance of AlexNet? Draw and describe its architecture.

AlexNet was one of the first convolutional networks designed for image classification using the ImageNet data set. It was composed of five convolution layers followed by three fully connected layers. One of the contributions of this architecture that made its training remarkably faster and computationally efficient was the use of rectified linear unit (ReLu) transfer functions in the convolution layers instead of the traditional sigmoid functions.

7. What is GoogLeNet? How does it work?

Google Lens uses deep learning artificial neural network algorithms (along with other AI techniques) to deliver information about the images captured by users from their nearby objects. This involves identifying the objects, products, plants, animals, and locations and providing information about them on the Internet.

8. How does CNN process text? What are word embeddings, and how do they work?

CNN processes text using the same approach as is used for processing images. Word in beddings are also known as word vectors where the relative position of words in text is given weight for additional context during processing.

9. What is word2vec, and what does it add to traditional text mining

Word2vec is a two-layer neural network that gets a large text corpus as the input and converts each word in the corpus to a numeric vector of any given size with very interesting features. Although word2vec itself is not a deep learning algorithm, its outputs already have been widely used in many deep learning research and commercial projects as inputs. By providing such a meaningful representation of textual data, in recent years, word2vec has driven many deep learning–based text mining projects in a wide range of contexts (e.g., medical, computer science, social media, marketing), and various types of deep networks have been applied to the word embeddings created by this algorithm to accomplish different objectives.

Section 6.8 Review Questions

1. What is RNN? How does it differ from CNN?

RNN is a recurrent neural network. RNNs are the type of neural networks that have memory and that apply that memory to determine their future outputs It differs from CNN in that it has these defined feedback methods.

2. What is the significance of “context,” “sequence,” and “memory” in RNN?

Our and ends are useful because they are able to include context in their processing. In this circumstance context refers to an understanding of the past sequence of events and the ability to use memory to hold these past events and use them in consideration for decisions.

3. Draw and explain the functioning of a typical recurrent neural network unit.

Student drawings will vary, but will closely resemble figure 6.31 and describe how past inputs are considered before outputs are generated.

4. What is the LSTM network, and how does it differ from RNNs?

LSTM (long short-term memory) networks have been widely used in many sequence modeling applications, including image captioning, handwriting recognition and generation, parsing, speech recognition, and machine translation. LSTM differs from RNN in that it has the ability to process longer term dependencies.

5. List and briefly describe three different types of LSTM applications.

· Image captioning - the ability to label video images using context of previous screens

· Speech recognition - the ability to recognize spoken words accurately utilizing their context within a sentence.

· Machine translation - the ability to translate speech to text from one language to another

6. How do Google’s Neural Machine Translation and Microsoft Skype Translator work

These voice translation services include both speech recognition and machine translation in real time. Users can speak in different languages and their conversation is recognized and then translated for the other listener.

Section 6.9 Review Questions

1. Despite the short tenure of deep learning implementation, why do you think there are several different computing frameworks for it?

Perceptions and opinions will vary will likely focus on the power and utility of these frameworks.

2. Define CPU, NVIDIA, CUDA, and deep learning, and comment on the relationship between them.

CPU - central processing unit

NVIDIA - developer of GPU hardware

CUDA - programming language used on NVIDIA GPU hardware

Deep learning - the newest branch of AI and machine-learning, attempts mimic the thought process of humans using mathematical algorithms to learn from data.

Much of the progress in deep learning is due to the availability of GPU hardware that allows for detailed learning architectures.

3. List and briefly define the characteristics of different deep learning frameworks.

The operation of these libraries mostly relies on a parallel computing platform and application programming interface (API) developed by NVIDIA called Compute Unified Device Architecture (CUDA), which enables software developers to use GPUs made by NVIDIA for general-purpose processing. In fact, each deep learning framework consists of a high-level scripting language (e.g., Python, R, Lua) and a library of deep learning routines usually written in C (for using CPUs) or CUDA (for using GPUs).

4. What is Keras, and how is it different from the other frameworks?

Keras (https:// keras.io/) is an open-source neural network library written in Python that functions as a high-level application programming interface (API) and is able to run on top of various deep learning frameworks including Theano and TensorFlow. In essence, Keras just by getting the key properties of network building blocks (i.e., type of layers, transfer functions, and optimizers) via an extremely simple syntax automatically generates syntax in one of the deep learning frameworks and runs that framework in the backend.

Section 6.10 Review Questions

1. What is cognitive computing, and how does it differ from other computing paradigms?

Cognitive computing makes a new class of problems computable. It addresses highly complex situations that are characterized by ambiguity and uncertainty. Instead of seeking results as other paradigms may, cognitive computing seeks to augment human capability to find solutions.

2. Draw a diagram and explain the conceptual framework of cognitive computing. Make sure to include inputs, enablers, and expected outcomes in your framework.

Student drawings will differ but will closely resemble figure 6.36.

3. List and briefly define the key attributes of cognitive computing.

· Adaptive: Cognitive systems must be flexible enough to learn as information changes and goals evolve. The systems must be able to digest dynamic data in real time and make adjustments as the data and environment change.

· Interactive: Human-computer interaction (HCI) is a critical component in cognitive systems. Users must be able to interact with cognitive machines and define their needs as those needs change. The technologies must also be able to interact with other processors, devices, and cloud platforms.

· Iterative and stateful: Cognitive computing technologies can also identify problems by asking questions or pulling in additional data if a stated problem is vague or incomplete. The systems do this by maintaining information about similar situations that have previously occurred. •

· Contextual: Understanding context is critical in thought processes, so cognitive systems must understand, identify, and mine contextual data, such as syntax, time, location, domain, requirements, and a specific user’s profile, tasks, or goals. Cognitive systems may draw on multiple sources of information, including structured and unstructured data and visual, auditory, or sensor data.

4. How does cognitive computing differ from ordinary AI techniques?

· Technologies - both technologies utilize machine learning, and LP, neural networks and deep learning: but cognitive computing adds text mining and sentiment analysis

· capabilities offered - while AI seeks to find hidden patterns in data sources, cognitive computing attempts to simulate the human thought process to find solutions

· purpose - AI seeks to automate complex processes whereas cognitive computing seeks to augment human capability

· industries - both have application in most industries, but the focus of cognitive computing is currently in customer service, marketing, healthcare, entertainment and services

5. What are the typical use cases for cognitive analytics?

The typical use cases for cognitive computing include the following:

· Development of smart and adaptive search engines

· Effective use of natural language processing

· Speech recognition • Language translation

· Context-based sentiment analysis

· Face recognition and facial emotion detection

· Risk assessment and mitigation

· Fraud detection and mitigation

· Behavioral assessment and recommendations

6. Explain what the terms cognitive analytics and cognitive search mean.

Cognitive analytics is a term that refers to cognitive computing–branded technology platforms, such as IBM Watson, that specialize in processing and analyzing large, unstructured data sets.

Cognitive search is the new generation search method that uses AI (advanced indexing, NLP, and machine learning) to return results that are much more relevant to users.

7. What is IBM Watson and what is its significance to the world of computing?

IBM Watson is perhaps the smartest computer system built to date. In 2010, an IBM research team developed Watson, an extraordinary computer system—a novel combination of advanced hardware and software—designed to answer questions posed in natural human language.

8. How does Watson work?

Watson uses DeepQA:

· Massive parallelism. Watson needed to exploit massive parallelism in the consideration of multiple interpretations and hypotheses.

· Many experts. Watson needed to be able to integrate, apply, and contextually evaluate a wide range of loosely coupled probabilistic questions and content analytics.

· Pervasive confidence estimation. No component of Watson committed to an answer; all components produced features and associated confidences, scoring different question and content interpretations. An underlying confidence-processing substrate learned how to stack and combine the scores.

· Integration of shallow and deep knowledge. Watson needed to balance the use of strict semantics and shallow semantics, leveraging many loosely formed ontologies.

9. List and briefly explain five use cases for IBM Watson

Selections and descriptions will vary, but five potential use areas described in the chapter include:

· healthcare and medicine

· security

· finance

· retail

· education

· government

· research

ANSWERS TO APPLICATION CASE QUESTIONS FOR DISCUSSION( (

Application Case 6.1: Finding the Next Football Star with Artificial Intelligence

1. What does SciSports do? Look at its Web site for more information.

SciSports is a provider of intelligence and analytics to professional football organizations.

2. How can advanced analytics help football teams?

Analytics can provide a number of benefits by allowing teams to better understand gameplay overall including player and team strengths and weaknesses, competitor analysis, game play analysis and predictive modeling for player selection.

3. What is the role of deep learning in solutions provided by SciSports?

The company uses deep learning in several solutions including learning the best layouts for players in corner kicks him and managing the compilation of 3D imaging by better understanding ball movements.

Application Case 6.2: Gaming Companies Use Data Analytics to Score Points with Players

1. What are the main challenges for gaming companies?

The videogame industry is highly competitive, and gaming companies must ensure that they are providing a product that meets their customers needs. As a part of this marketing companies have determined that understanding and connecting with players can drive sales and loyalty.

2. How can analytics help gaming companies stay competitive?

Analytics allows gaming companies to better understand their players activities and desires. Better understanding in these areas allows player behaviors to be monetized more completely; this includes running more effective marketing campaigns, improving retention and using results to improve future games.

3. What types of data can gaming companies obtain and use for analytics?

Giving companies are able to access large amounts of information on gamers for titles that are played online. This information can include duration of play, time of play, types of games played and the size of a gamers network.

Application Case 6.3: Artificial Intelligence Helps Protect Animals from Extinction

1. What is WildTrack and what does it do?

WildTrack is an artificial intelligence system that is designed to aid humans in the tracking of animals.

2. How can advanced analytics help WildTrack?

The system is able to provide detailed information on animals based on their tracks, but is also able to learn and provide more complete and accurate information over time.

3. What are the roles that deep learning plays in this application case?

The ability to better understand patterns in the tracks and become more accurate and detailed is a feature set provided by deep learning.

Application Case 6.4: Sensitivity Analysis Reveals Injury Severity Factors in Traffic Accidents

1. How does sensitivity analysis shed light on the black box (i.e., neural networks)?

Sensitivity analysis techniques provide a clear interpretation of how a neural network does what it does; that is, specifically how (and to what extent) the individual inputs factor into the generation of specific network output. This process extracts the cause-and-effect relationships among the inputs and the outputs of a trained neural network model.

2. Why would someone choose to use a black-box tool such as neural networks over theoretically sound, mostly transparent statistical tools like logistic regression?

ANN are known to be superior in capturing highly nonlinear complex relationships between predictor and target variables, assuming a linear relationship is often an oversimplification of the problem.

3. In this case, how did neural networks and sensitivity analysis help identify injury-severity factors in traffic accidents?

ANN and sensitivity analysis helped estimate the significance of the crash factors on the level of injury severity sustained by the driver. This study was a two-step process. In the first step, the testers developed a series of prediction models (one for each injury severity level) to capture the in-depth relationships between the crash-related factors and a specific level of injury severity. In the second step, they conducted sensitivity analysis on the trained neural network models to identify the prioritized importance of crash-related factors as they relate to different injury severity levels.

The study revealed that the variable seatbelt use was the most important determinant for predicting higher levels of injury severity but it was one of the least significant predictors for lower levels of injury severity. Other interesting findings involved gender (good predictor for low injury severity, but not for high) and age (vice versa).

Application Case 6.5: Georgia DOT Variable Speed Limit Analytics Help Solve Traffic Congestions

1. What was the nature of the problems that GDOT was trying to solve with data science?

The department wanted to evaluate the effectiveness of a proof of concept variable speed limit (VSL) pilot program.

2. What type of data do you think was used for the analytics?

Student opinions will vary but most likely will include minimum speed, maximum speed, average speed, number of vehicles and number of complete stops (traffic jams).

3. What were the data science metrics developed in this pilot project? Can you think of other metrics that can be used in this context?

The two new metrics that were developed as part of this study included an evaluation of frequency and duration of slowdown locations as well as a better understanding of turbulence created by bottlenecks. Student ideas on additional metrics will vary.

Application Case 6.6: From Image Recognition to Face Recognition

1. What are the technical challenges in face recognition?

Face recognition is difficult because the goal is to identify an individual as opposed to a class using moving 3D images.

2. Beyond security and surveillance purposes, where else do you think face recognition can be used?

This is an open ended question student ideas will be different.

3. What are the foreseeable social and cultural problems with developing and using face recognition technology?

It is possible that extensive use of facial recognition can be an issue of privacy, with individuals being easily tracked whenever they are in public.

Application Case 6.7: Deliver Innovation by Understanding Customer Sentiment

1. Why do you think sentiment analysis is gaining overwhelming popularity?

Student opinions will vary, but may focus on the importance of understanding consumer sentiments as it relates to many aspects of life including commerce and politics.

2. How does sentiment analysis work? What does it produce?

Sentiment analysis works by analyzing text and notes available from individuals in some relation to the topic being addressed. Information on products and features gleaned categorized and analyzed. This results in a better understanding of how a product is perceived and how it meets customer needs.

3. In addition to the specific examples in this case, can you think of other businesses and industries that can benefit from sentiment analysis? What is common among the companies that can benefit greatly from sentiment analysis?

Student opinions on other examples will vary, but one commonality may be that companies that deal directly with the public where consumers have a choice in product or service offering may benefit significantly.

Application Case 6.8: IBM Watson Competes against the Best at Jeopardy!

1. In your opinion, what are the most unique features about Watson?

Student opinions and responses will vary.

Watson is a question answering (QA) computer system developed by an IBM Research team and named after IBM’s first president as part of a project called DeepQA. What makes it special is that it is able to compete at the human champion level in real time on the TV quiz show, Jeopardy!; in fact, in 2011, it was able to defeat Ken Jennings, who held the record for the longest winning streak in the game. Like Deep Blue has done with chess, Watson is showing that computer systems are getting quite good at demonstrating human-like intelligence.

2. In what other challenging games would you like to see Watson compete against humans? Why?

Student opinions and selections will vary.

3. What are the similarities and differences between Watson’s and humans’ intelligence?

The DeepQA architecture involves massive parallelism, many experts, pervasive confidence estimation, and integration of the-latest-and-greatest in text analytics, involving both shallow and deep semantic knowledge. As implemented in Watson, DeepQA brings more than 100 different techniques for analyzing natural language, identifying sources, finding and generating hypotheses, finding and scoring evidence, and merging and ranking hypotheses. More important than any particular technique is the combination of overlapping approaches that can bring their strengths to bear and contribute to improvements in accuracy, confidence, and speed.

ANSWERS TO END OF CHAPTER QUESTIONS FOR DISCUSSION( ( (

1. What is deep learning? What can deep learning do that traditional machine-learning methods cannot?

Deep learning, the newest branch of AI and machine-learning, attempts mimic the thought process of humans using mathematical algorithms to learn from data.

2. List and briefly explain different learning paradigms/ methods in AI.

These details are nicely summarized in figure 6.4. In knowledge-based systems, manually created representations provide the framework for evaluating information. In classic machine learning, manually created features provide the foundation for mapping of those features and then output. In generic representational learning features are created automatically and then mapping is performed from those features. In deep learning features are also created automatically but move beyond a single stage process to create more advanced features first before completing mapping functions.

3. What is representation learning, and how does it relate to machine learning and deep learning?

Representation learning techniques entail one type of machine learning (which is also a part of AI) in which the emphasis is on learning and discovering features by the system in addition to discovering the mapping from those features to the output/target. Deep learning is a branch of representation learning.

4. List and briefly describe the most commonly used ANN activation functions.

Most commonly used AAN and activation functions are:

· Hard limit - operation as a baseline but increases to a set point if the condition is met

· Linear - operation increases as condition increases

· Log-Sigmoid - operation increases on a log scale as condition increases

· Positive linear - operation has a baseline but then increases linearly after a condition is met

It is very helpful to refer to table 6.1

5. What is MLP, and how does it work? Explain the function of summation and activation weights in MLP-type ANN.

MLP is multi- layer perception and represents neurons with multiple inputs as seen in figure 6.6. In this type of neuron each input would have its own adjustable weight.

6. List and briefly describe the nine-step process in conducting a neural network project.

These are shown in the flowchart of Figure 6.9. After testing (step 8), it is possible to return to a previous step:

Step 1:Collect data

Step 2:Separate (data) into training and test sets

Step 3: Define a network structure

Step 4: Select a learning algorithm

Step 5:Set parameters and values

Step 6:Initialize weights and start training

Step 7:Stop training, freeze the network weights

Step 8:Test the trained network

Step 9:Implementation: use the network with new cases

7. Draw and briefly explain the three-step process of learning in ANN.

In supervised learning, the learning process is inductive; that is, connection weights are derived from existing cases. The usual process of learning involves three tasks:

1. Compute temporary outputs.

2. Compare outputs with desired targets.

3. Adjust the weights and repeat the process.

Student drawings will closely resemble figure 6.12.

8. How does the backpropagation learning algorithm work?

Backpropagation (short for back-error propagation) is the most widely used supervised learning algorithm in neural computing. Backpropagation involves a feedforward network with one or more hidden layers. Backpropagation involves supervised learning because predetermined correct answers for each training pattern are given and compared to the results given by the backpropagation algorithm. The difference between the correct answer and the output (known as the delta value) is calculated, and weights of the output nodes are adjusted accordingly. Weight adjustment also proceeds backward to the hidden layers; hence the term “backpropagation.” This process repeats, continuously testing against training patterns, until the delta value is sufficiently low.

9. What is overfitting in ANN learning? How does it happen, and how can it be prevented?

Overfitting happens when the trained model is highly fitted to the training data set but performs poorly with regard to external data sets. A large group of strategies known as regularization strategies is designed to prevent models from overfitting by making changes or defining constraints for the model parameters or the performance function.

10. What is the so-called black-box syndrome? Why is it important to be able to explain an ANN’s model structure?

ANN are typically thought of as black boxes, capable of solving complex problems but lacking the explanation of its capabilities. This phenomenon is commonly referred to as the “black-box” syndrome.

11. How does sensitivity analysis work in ANN? Search the Internet to find other methods to explain ANN methods.

Sensitivity analysis is a method for extracting the cause-and-effect relationships among the inputs and the outputs of a trained neural network model. In the process of performing sensitivity analysis, the trained neural network’s learning capability is disabled so that the network weights are not affected.

12. What is meant by “deep” in deep neural networks? Compare deep neural network to shallow neural network.

Deep neural networks have a large number of hidden layers in addition to a large number of neurons in each layer. This allows for a much more complex processing activity than was available in a shallow neural network and is supported by improvements in hardware technology.

13. What is GPU? How does it relate to deep neural networks?

A GPU is a graphical processing unit that is able to be used for AI purposes and allows neural networks with over 1 million neurons to function. They are significantly faster than traditional CPUs for this purpose.

14. How does a feedforward multilayer perceptron–type deep network work?

These networks are simply large-scale neural networks that can contain many layers of neurons and handle tensors as their input. These models are called feedforward because the flow of information that goes through them is always forwarding and no feedback connections (i.e., connections in which outputs of a model are fed back to itself) are allowed.

15. Comment on the impact of random weights in developing deep MLP.

Optimization of the performance (loss) function in many real applications of deep MLPs is a challenging issue. The problem is that applying the common gradient-based training algorithms with random initialization of weights and biases that is very efficient for finding the optimal set of parameters in shallow neural networks most of the time could lead to getting stuck in the locally optimal solutions rather than catching the global optimum values for the parameters. As the depth of network increases, chances of reaching a global optimum using random initializations with the gradient-based algorithms decrease.

16. Which strategy is better: more hidden layers versus more neurons?

Whereas theoretically it is still an open research question, practically using more layers in a network seems to be more effective and computationally more efficient than using many neurons in a few layers.

17. What is CNN?

CNN is a convolutional neural network that includes at least one layer involving a convolutional weight function instead of the general matrix multiplication.

18. For what type of applications can CNN be used?

CNN was initially designed for computer vision applications but is also applicable to non-image data sets as well.

19. What is the convolution function in CNN, and how does it work?

The convolution function uses the concept of parameter sharing which in addition to computational efficiency allows for weights parameters to be shared thus reducing processing load in systems with a large number of weight parameters.

20. What is pooling in CNN? How does it work?

Generally, a convolution layer is followed by another layer known as the pooling layer. The purpose of a pooling layer is to consolidate elements in the input matrix to produce a smaller output matrix while maintaining the important features. Normally, a pooling function involves an r * c consolidation window that moves around the input matrix and in each move calculates some summary statistics of the elements involved in the consolidation window so that it can be put in the output image.

21. What is ImageNet, and how does it relate to deep learning?

ImageNet is an ongoing research project that provides researchers with a large database of images, each linked to a set of synonym words from WordNet. ImageNet is a huge database for developing image processing–type deep networks.

22. What is the significance of AlexNet? Draw and describe its architecture.

AlexNet was one of the first convolutional networks designed for image classification using the ImageNet data set. It was composed of five convolution layers followed by three fully connected layers. One of the contributions of this architecture that made its training remarkably faster and computationally efficient was the use of rectified linear unit (ReLu) transfer functions in the convolution layers instead of the traditional sigmoid functions.

23. What is GoogLeNet? How does it work?

Google Lens uses deep learning artificial neural network algorithms (along with other AI techniques) to deliver information about the images captured by users from their nearby objects. This involves identifying the objects, products, plants, animals, and locations and providing information about them on the Internet.

24. How does CNN process text? What is word embeddings, and how does it work?

CNN processes text using the same approach as is used for processing images. Word in beddings are also known as word vectors where the relative position of words in text is given weight for additional context during processing.

25. What is word2vec, and what does it add to the traditional text mining?

Word2vec is a two-layer neural network that gets a large text corpus as the input and converts each word in the corpus to a numeric vector of any given size with very interesting features. Although word2vec itself is not a deep learning algorithm, its outputs already have been widely used in many deep learning research and commercial projects as inputs. By providing such a meaningful representation of textual data, in recent years, word2vec has driven many deep learning–based text mining projects in a wide range of contexts (e.g., medical, computer science, social media, marketing), and various types of deep networks have been applied to the word embeddings created by this algorithm to accomplish different objectives.

26. What is RNN? How does it differ from CNN?

RNN is a recurrent neural network. RNNs are the type of neural networks that have memory and that apply that memory to determine their future outputs. It differs from CNN in that it has these defined feedback methods.

27. What is the significance of context, sequence, and memory in RNN?

Our and ends are useful because they are able to include context in their processing. In this circumstance context refers to an understanding of the past sequence of events and the ability to use memory to hold these past events and use them in consideration for decisions.

28. Draw and explain the functioning of a typical recurrent neural network unit.

Student drawings will vary, but will closely resemble figure 6.31 and describe how past inputs are considered before outputs are generated.

29. What is LSTM network, and how does it differ from RNNs?

LSTM (long short-term memory) networks have been widely used in many sequence modeling applications, including image captioning, handwriting recognition and generation, parsing, speech recognition, and machine translation. LSTM differs from RNN in that it has the ability to process longer term dependencies.

30. List and briefly describe three different types of LSTM applications.

· Image captioning - the ability to label video images using context of previous screens

· Speech recognition - the ability to recognize spoken words accurately utilizing their context within a sentence.

· Machine translation - the ability to translate speech to text from one language to another

31. How do Google’s Neural Machine Translation and Microsoft Skype Translator work?

These voice translation services include both speech recognition and machine translation in real time. Users can speak in different languages and their conversation is recognized and then translated for the other listener.

32. Despite its short tenure, why do you think deep learning implementation has several different computing frameworks?

Perceptions and opinions will vary will likely focus on the power and utility of these frameworks.

33. Define and comment on the relationship between CPU, NVIDIA, CUDA, and deep learning.

CPU - central processing unit

NVIDIA - developer of GPU hardware

CUDA - programming language used on NVIDIA GPU hardware

Deep learning - the newest branch of AI and machine-learning, attempts mimic the thought process of humans using mathematical algorithms to learn from data.

Much of the progress in deep learning is due to the availability of GPU hardware that allows for detailed learning architectures.

34. List and briefly define the characteristics of different deep learning frameworks.

The operation of these libraries mostly relies on a parallel computing platform and application programming interface (API) developed by NVIDIA called Compute Unified Device Architecture (CUDA), which enables software developers to use GPUs made by NVIDIA for general-purpose processing. In fact, each deep learning framework consists of a high-level scripting language (e.g., Python, R, Lua) and a library of deep learning routines usually written in C (for using CPUs) or CUDA (for using GPUs).

35. What is Keras, and how does it differ from other frameworks?

Keras (https:// keras.io/) is an open-source neural network library written in Python that functions as a high-level application programming interface (API) and is able to run on top of various deep learning frameworks including Theano and TensorFlow. In essence, Keras just by getting the key properties of network building blocks (i.e., type of layers, transfer functions, and optimizers) via an extremely simple syntax automatically generates syntax in one of the deep learning frameworks and runs that framework in the backend.

36. What is cognitive computing and how does it differ from other computing paradigms?

Cognitive computing makes a new class of problems computable. It addresses highly complex situations that are characterized by ambiguity and uncertainty. Instead of seeking results as other paradigms may, cognitive computing seeks to augment human capability to find solutions.

37. Draw a diagram and explain the conceptual framework of cognitive computing. Make sure to include inputs, enablers, and expected outcomes in your framework.

Student drawings will differ but will closely resemble figure 6.36.

38. List and briefly define the key attributes of cognitive computing.

· Adaptive: Cognitive systems must be flexible enough to learn as information changes and goals evolve. The systems must be able to digest dynamic data in real time and make adjustments as the data and environment change.

· Interactive: Human-computer interaction (HCI) is a critical component in cognitive systems. Users must be able to interact with cognitive machines and define their needs as those needs change. The technologies must also be able to interact with other processors, devices, and cloud platforms.

· Iterative and stateful: Cognitive computing technologies can also identify problems by asking questions or pulling in additional data if a stated problem is vague or incomplete. The systems do this by maintaining information about similar situations that have previously occurred. •

· Contextual: Understanding context is critical in thought processes, so cognitive systems must understand, identify, and mine contextual data, such as syntax, time, location, domain, requirements, and a specific user’s profile, tasks, or goals. Cognitive systems may draw on multiple sources of information, including structured and unstructured data and visual, auditory, or sensor data.

39. How does cognitive computing differ from ordinary AI techniques?

· Technologies - both technologies utilize machine learning, and LP, neural networks and deep learning: but cognitive computing adds text mining and sentiment analysis

· capabilities offered - while AI seeks to find hidden patterns in data sources, cognitive computing attempts to simulate the human thought process to find solutions

· purpose - AI seeks to automate complex processes whereas cognitive computing seeks to augment human capability

· industries - both have application in most industries, but the focus of cognitive computing is currently in customer service, marketing, healthcare, entertainment and services

40. What are the typical use cases for cognitive analytics?

The typical use cases for cognitive computing include the following:

· Development of smart and adaptive search engines

· Effective use of natural language processing

· Speech recognition • Language translation

· Context-based sentiment analysis

· Face recognition and facial emotion detection

· Risk assessment and mitigation

· Fraud detection and mitigation

· Behavioral assessment and recommendations

41. What is cognitive analytics? What is cognitive search?

Cognitive analytics is a term that refers to cognitive computing–branded technology platforms, such as IBM Watson, that specialize in processing and analyzing large, unstructured data sets.

Cognitive search is the new generation search method that uses AI (advanced indexing, NLP, and machine learning) to return results that are much more relevant to users.

42. What is IBM Watson, and what is its significance to the world of computing?

IBM Watson is perhaps the smartest computer system built to date. In 2010, an IBM research team developed Watson, an extraordinary computer system—a novel combination of advanced hardware and software—designed to answer questions posed in natural human language.

43. How does IBM Watson work?

Watson uses DeepQA:

· Massive parallelism. Watson needed to exploit massive parallelism in the consideration of multiple interpretations and hypotheses.

· Many experts. Watson needed to be able to integrate, apply, and contextually evaluate a wide range of loosely coupled probabilistic questions and content analytics.

· Pervasive confidence estimation. No component of Watson committed to an answer; all components produced features and associated confidences, scoring different question and content interpretations. An underlying confidence-processing substrate learned how to stack and combine the scores.

· Integration of shallow and deep knowledge. Watson needed to balance the use of strict semantics and shallow semantics, leveraging many loosely formed ontologies.

44. List and briefly explain five use cases for IBM Watson.

Selections and descriptions will vary, but five potential use areas described in the chapter include:

· healthcare and medicine

· security

· finance

· retail

· education

· government

· research

ANSWERS TO END OF CHAPTER EXercises( (

Teradata University Network (TUN) and Other Hands-On and Internet Exercises

1. Go to the Teradata University Network Web site (teradatauniversitynetwork.com). Search for teaching and learning materials (e.g., articles, application cases, white papers, videos, exercises) on deep learning, cognitive computing, and IBM Watson. Read the material you have found. If needed, also conduct a search on the Web to enhance your findings. Write a report on your findings.

Student research and reports will vary.

2. Deep learning is relatively new to the world of analytics. Its application cases and success stories are just starting to emerge in the Web. Conduct a comprehensive search on your school’s digital library resources to identify at least five journal articles where interesting deep learning applications are described. Write a report on your findings.

Student reports will vary.

3. Most of the applications of deep learning today are developed using R- and/or Python-based open-source computing resources. Identify those resources (frameworks such as Torch, Caffe, TensorFlow, Theano, Keras) available for building deep learning models and applications. Compare and contrast their capabilities and limitations. Based on your findings and understanding of these resources, if you were to develop a deep learning application, which one would you choose to employ? Explain and justify/defend your choice.

Student selections and analysis will vary.

4. Cognitive computing has become a popular term to define and characterize the extent of the ability of machines/ computers to show “intelligent” behavior. Thanks to IBM

Watson and its success on Jeopardy!, cognitive computing and cognitive analytics are now part of many realworld intelligent systems. In this exercise, identify at least three application cases where cognitive computing was used to solve complex real-world problems. Summarize your findings in a professionally organized report.

Student research, analysis and reports will differ.

5. Download KNIME analytics platform, one of the most popular free/open-source software tools from knime. org. Identify the deep learning examples (where Keras is used to build some exemplary prediction/classification models) in its example folder. Study the models in detail. Understand what it does and how exactly it does it. Then, using a different but similar data set, build and test your own deep learning prediction model. Report your findings and experiences in a written document.

Student reports will vary.

6. Search for articles related to “cognitive search.” Identify at least five pieces of written material (a combination of journal articles, white papers, blog posts, application cases, etc.). Read and summarize your findings. Explain your understanding of cognitive search and how it differs from regular search methods.

Student selection of articles and thus reports will differ.

7. Go to Teradata.com. Search and find application case studies and white papers on deep learning and/or cognitive computing. Write a report to summarize your findings, and comment on the capabilities and limitations (based on your understanding) of these technologies.

Students will select and write on different cases.

8. Go to SAS.com. Search and find application case studies and white papers on deep learning and/or cognitive computing. Write a report to summarize your findings, and comment on the capabilities and limitations (based on your understanding) of these technologies.

Students will select and write on different cases.

9. Go to IBM.com. Search and find application case studies and white papers on deep learning and/or cognitive computing. Write a report to summarize your findings, and comment on the capabilities and limitations (based on your understanding) of these technologies.

Students will select and write on different cases.

10. Go to TIBCO.com or some other advanced analytics company Web site. Search and find application case studies and white papers on deep learning and/or cognitive computing. Write a report to summarize your findings, and comment on the capabilities and limitations (based on your understanding) of these technologies.

Student research, analysis and reports will differ.