Deck 6: Deep Learning and Cognitive Computing

Question 1

In the opening vignette, banks had begun to use deep learning to provide more accurate recognition of fraud.

Answer 1

True

Question 2

Deep learning is a separate branch of science from AI.

Answer 2

False

Question 3

Structured, human-mediated machine-learning approaches have been working fine for rather abstract and formal tasks.

Answer 3

True

Question 4

The initial idea for deep learning began in 2013.

Question 5

In artificial neurons weight terms are adjustable but biased terms are fixed.

Question 6

Deep neural networks are referred to as "deep" because they are primarily used to evaluate significant philosophical issues.

Question 7

The growth in the use of deep neural networks has been spurred by advancements in GPU hardware technology.

Question 8

Feedforward models only allow a single feedback connection.

Question 9

CNNs are only applicable to image data sets.

Question 10

ANN are capable of solving complex problems and provide detailed explanations of their capabilities.

Question 11

Convolution layers are often followed by a pooling layer.

Question 12

Deep MLP and convolutional networks are specialized for processing a sequential grid of values.

Question 13

RNN basically models a dynamic system.

Question 14

In Caffe, everything is done using text files instead of code.

Question 15

Theano is a Java library developed by the Deep Learning Group at the University of Montreal.

Question 16

In the opening vignette, the results did not include:

A) elimination of fraud completely at 15% of branches.
B) reduction in false positives by 60%.
C) increase in true positives by 50%.
D) resources focused on actual cases of fraud.

Question 17

What factors are not allowing banks to use more advanced AI based fraud detection systems?

A) evolving new learning systems enabled by AI
B) hardware and processors becoming more powerful
C) specialized algorithms becoming available
D) easy access to development capital

Question 18

Which type of learning system is represented by the following process: input - manually created representation - output?

A) knowledge-based systems
B) classic machine learning
C) generic representation learning
D) deep learning

Question 19

Which type of learning system is represented by the following process: input - simple features - more advanced features - mapping from features - output?

A) knowledge-based systems
B) classic machine learning.
C) generic representation learning
D) deep learning

Question 20

Which type of a and a transfer function is described: operation increases in a line based on condition using an operation such as a=n.

A) hard limit
B) linear
C) log sigmoid
D) positive linear

Question 21

Which type of a and a transfer function is described: operation increases in a line based on condition, but only after a minimum is met using an operation such as a=n if n>0, or a=0 if n<0.

A) hard limit
B) linear
C) log sigmoid
D) positive linear

Question 22

Sensitivity analysis is a method for extracting the what among the inputs and the outputs of a trained neural network model?

A) underlying constraints
B) cause-and-effect relationships
C) bias
D) detailed quantitative predictive models

Question 23

Which of the following does not describe a deep feedforward network?

A) most general type of deep network
B) many layers of neurons
C) network elements are significantly different than ANN models
D) tensors are handled as input

Question 24

Which type of neural network is characterized by allowing feedback connections?

A) deep feedforward network
B) feedforward network
C) recurrent neural network
D) none of these

Question 25

Why is pretraining a deep MLP network appropriate?

A) increases the chances of a global optimum being reached
B) is necessary always necessary for this type of network
C) increases duration before global optimum is reached
D) none of these

Question 26

What concept do CNN's use to avoid large numbers of weight parameters being defined and trained?

A) convolutional theory
B) parameter sharing
C) filtering
D) all of these

Question 27

Real applications of deep learning in general and CNN's in particular highly depend on the availability of:

A) large, unsorted data sets.
B) small, unsorted data sets.
C) large, annotated data sets.
D) small, annotated data sets.

Question 28

Who published a weekly supervised training image recognition deep learning project in May 2018?

A) IBM
B) Google
C) Adobe
D) Facebook

Question 29

LSTM networks are typically not used for what type of data?

A) handwriting recognition
B) speech recognition
C) image recognition
D) machine translation

Question 30

Which deep learning framework supports TPUs?

A) Torch
B) Caffe
C) TensorFlow
D) Theano

Question 31

Which deep learning framework adds support to the LuaJIT library?

A) Torch
B) Caffe
C) TensorFlow
D) Theano

Question 32

Which deep learning framework is particularly known to be an efficient framework for image processing due to incredible speed in processing image files?

A) Torch
B) Caffe
C) TensorFlow
D) Theano

Question 33

Machine learning is a subset of ________.

Question 34

Large ________ have greatly compensated for the lack of intimate knowledge and reduced the level of skill needed for implementing deep neural networks.

Question 35

As it is being practiced today, ________ seems to be nothing but an extension of neural networks with the idea that deep learning is able to deal with more complicated tasks with a higher level of sophistication.

Question 36

Typically, a neuron has more than ________ input.

Question 37

Deep belief networks (DBNs) is a(n) ________ pretraining deep learning methods.

Question 38

Generally, a ________ order of layers has to be held between the input and the output layers in the MLP-type network architecture.

Question 39

A convolution layer is followed by a pooling layer, also know them as a ________ layer.

Question 40

________ has driven many deep learning-based text mining projects in a wide range of contexts.

Question 41

The ________ is caused by very small derivatives of sigmoid functions in some regions of the images.

Question 42

________ 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.

Question 43

Human thinking and understanding to a great extent relies on ________.

Question 44

________ refers to a subfield of AI that employs computer programs to translate speech or text from one language to another.

Question 45

________ is an open-source neural network library that functions as a high-level application programming interface (API).

Question 46

To program GPUs, the ________ language is utilized.

Question 47

Cognitive computing is often used interchangeably with ________, the umbrella term used for technologies that rely on data and scientific methods/computations to make (or help/support in making) decisions.

Question 48

What are the key attributes of cognitive computing systems?

Question 49

Describe the framework of DeepQA used by IBM's Watson.

Question 50

Create a diagram that describes how deep learning is aligned with representation learning, machine learning and artificial intelligence.

Question 51

Describe the nine steps in the development process for an ANN application.

Question 52

What are the five steps in the backpropagation learning algorithm?

Question 53

Define the term sensitivity analysis as it relates to ANNs.

Question 54

Describe an MLP deep network. Why are these labeled as feedforward networks?

Question 55

Select and discuss a computer framework for implementing deep learning. Make sure to discuss its origin hardware targets and popular implementations.

Question 56

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