AI exam 2 practice exam
Re-organize figure A, B, C below into correct order to represent Dimensionality Reduction:
C (3d), A(2d), B(1d)
what is computer vision?
CNN based algorithms focusing on understanding images/videos by performing tasks such as object classification, detection, and segmentation
Hidden layer stacking order
Convolution layer Activation layer Pooling layer Fully connected layer Final activation layer
True or False: In reinforcement learning, agent lives in a static environment and thus learning to take actions and find the best policy without analyzing the state of its environment
False, it is constantly analyzing the state of its environment
reinforcement learning part 4
Update policy to maximize a reward
logistic regression definition
a classification learning algorithm based on the concept of probability, typically by using the sigmoid function
activation function definition
a function used to decide whether the output exceeds a certain threshold and gives on or off as an output
linear regression definition
a regression learning algorithm that uses a linear combination of features (x) and parameters ( w and b) to build a prediction model
"If a patient is prescribed steroid cream, then the patient is 70% likely of buying antihistamines." The consequent and antecedent in the above sentence (in order) are a) Antihistamines, steroid cream b) Steroid cream, patient c) Steroid cream, antihistamines d) Antihistamines, patient
a) antihistamines, steroid cream
Which of the following terminology is not an element of causal inference? a) Document classification b) Features in each patient c) Observational data d) Treatment assignment
a) document classification
Consider the following sentence as an input in NLP: "I like listening to pop music". How many possible outputs are there if Trigram language model is used? a) Four b) Three c) Five d) Two
a) four
The following functions are suitable as activation functions, except: a) Linear regression b) Sigmoid function c) Tanh function d) ReLU function
a) linear regression
Natural Language Processing is a subfield of AI and an intersection of the following fields: a) Linguistics and computer science b) Neuroscience and sociology c) Neuroscience and computer science d) Sociology and linguistics
a) linguistics and computer science
The following statement is a potential factor contributing the difficulty of NLP progress, except: a) Natural language is not part of computer vision b) Natural language sentences are infinite c) Natural language is ambiguous d) Natural language has complex representations
a) natural language is not a part of computer vision
The following terminology describes language model as the basis of NLP, except: a) Perceptron b) N-gram c) Word2Vec d) Bag-of-words
a) perceptron
Which of the following layer that does not exist in recurrent neural network? a) Pooling layer b) Hidden layer c) Input layer d) Output layer
a) pooling layer
The following application is suitable for CNN, except: a) Speech recognition b) Computer vision c) Object recognition d) Face recognition
a) speech recognition
Which of the following is not considered a challenge in Federated Learning? a) Systems homogeneity b) Expensive communication c) Statistical heterogeneity d) Privacy concerns
a) systems homogeneity
main challenge of CV in the field of healthcare
access to a large and labeled data set that is currently being mitigated with data augmentation technique
6 elements of RL
agent environment state action policy rewards
reinforcement learning part 1
agent placed in an environment to take an action
Gradient Descent definition
an iterative process to minimize the cost value within a cost function (minimizing errors)
Reinforcement learning part 2
at every action the state of the environment changes
The following terminology/description depicts deep neural networks, except: a) Networks trained by deep learning b) A simple model of biological neuron in an artificial neural network c) Consists of multiple hidden layers and output layers d) Example algorithms include convolutional and recurrent neural networks
b) a simple model of biological neuron in an artificial neural network
Which of the following scenarios describes the challenge of implementing causal inference? a) Factual outcomes are not typically observed b) Counterfactual outcomes are not typically observed c) Observational data from patients lack features in each patient d) Causal inference is a simple machine learning problem
b) counterfactual outcomes are not typically observed
Which of the following layer belongs to classification step in CNN? a) Convolution layer b) Fully connected layer c) Pooling layer d) Input layer
b) fully connected layer
Variables that are multiplied by input signals and give importance to various input signals are called a) Step function b) Weights c) Bias d) Activation function
b) weights
The following Federated Learning terminology describes FL Clients, except: a) Mobile devices b) IoT devices c) Server d) Data silos
c) server
The following MRI data format is suitable for applying machine learning algorithm to predict MCI a) phosphorus imaging b) magnetic resonance spectroscopy c) structured t1-weighted d) diffusion tensor imaging
c) structured t1-weighted
Which of the following is not an element of Reinforcement Learning? a) Agent b) Environment c) Training data d) Rewards
c) training data
disadvantages of deep learning
computationally intensive and expensive considered opaque or black box (difficult to understand the output) requires large data to train
Which of the following statements is false regarding association rule mining? a) An association rule learning has two parts, resembling 'if/then' statements b) Association rules are created from analyzing database based on frequency of 'if/then' statements c) Support (s) metric measures how frequently 'if/then' relationship appears in a database d) Confidence (c) metric measures how frequently 'if/then' relationship appears in a database
d) Confidence (c) metric measures how frequently 'if/then' relationship appears in a database
The process to reduce the size or dimension of dataset while maintaining variation within the data as much as possible is a definition of: a) Features b) Curse of Dimensionality c) instances or Observations d) Dimensionality Reduction
d) dimensionality reduction
The following layer can be found within the stacking hidden layers in CNN, except: a) Pooling layer b) Convolution layer c) Fully connected layer d) Input layer
d) input layer
Which of the following describes cross-silo federated learning? a) Multiple patients b) Multiple computers c) Multiple mobile devices d) Multiple institutions
d) multiple institutions
A simple model of biological neuron in an artificial neural network is the definition of: a) Activation function b) Convolutional neural networks c) Deep learning d) Perceptron
d) perceptron
Which statement best describes the curse of dimensionality phenomenon? a) The algorithm fails to identify interesting relations between features due to unstructured database b) Failure in learning algorithms due to insufficient number of of feature dimension within datasets c) The algorithm cannot learn effectively due to the number of observations exceeding the total number of features d) The algorithm cannot learn effectively due to the number of features exceeding the total number of observations
d) the algorithm cannot learn effectively due to the number of features exceeding the total number of observations
definition of high dimensional data
data in which the number of features is greater than the number of observations (p>n)
definition of low dimensional data
data in which the number of observations far outnumbers the number of features
linear regression in neural networks
each neuron is a linear combination of inputs (x), weights (w), and biases (b)
true or false: In bag-of-words language model, the order and position of words relative to one another is the outmost important
false
true or false: Neural network is a sub-field of deep learning in the overall artificial intelligence hierarchy
false
true or false: Pre-processing step in NLP such as data cleaning typically takes the least amount of project time
false
true or false: Randomized Control Trials (RCTs) are considered superior compared to Adaptive Clinical Trials due to implementation of causal inference principles
false
true or false: Image recognition is one of the strong suits of RNN application
false, CNN
true or false: Convolutional neural networks only applicable to sequential data and are commonly used in voice recognition software
false, image data
True or False: Association rule learning is only suitable to handle numerical data
false, only categorical
true or false: Federated Learning allows direct raw data communication while distributed machine learning prevents this action
false, opposite
advantages of deep learning
features automatically selected considered robust and can handle variation applicable to diff data types scalable
output layer
final layer where prediction is
gradient descent neural networks
gauge accuracy in each iteration optimize w and b
elements of CNN
image data inputs and outputs independent parameter and weight sharing no recurrent connection
Elements of ANN
inputs text or tabular data inputs and outputs independent no parameter or weight sharing no recurrent connection
two machine learning algorithms that can be applied in Toxicology
kNN support vector machine
NLP impact in healthcare
machine translation speech recognition summarization topic modeling
example of CV in healthcare
medical imaging, flagging certain areas in imaging
A language terminology that focuses on the smallest units that have meanings are called a) Phonetics b) Syntax c) Semantics d) Morphology
morphology
What is federated learning?
multiple entities collaborate in solving a machine learning problem under the coordination of a central server. Raw data is stored locally and not exchanged or transferred.
The following terminology is an alias to artificial neural network, except: a) Feedforward neural networks b) Multilayer perceptrons c) Vanilla neural networks d) Recurrent neural networks
recurrent neural networks
association rule learning definition
rule based machine learning method for discovering interesting relations between variables in large databases
elements of RNN
sequence data outputs are dependent on what comes before them parameter and weight sharing recurrent connection
NLP impact in education
spam detection
NLP impact in business
spelling check personal assistant document classification hand writing recognition
activation function in neural network
sum at each neuron decides if the neuron is turned on or off
two ML algorithms that can be applied to predict MCI diagnosis
support vector machine, support vector regression
reinforcement learning part 3
the agent will analyze the state a take action suitable for that state
The following statements are considered as contributing factors for implementing artificial intelligence and machine learning in the field of toxicology, except:
the database containing chemical properties for investigational drugs is small and limited
goal of association rule learning
to identify strong rules discovered in databases using some measure of interestingness
True or False: A given dataset containing one hundred features with a total of sixty observations is considered high-dimensional data
true
true or false: Activation functions are computed at each node throughout the neural networks
true
true or false: Back-propagation includes adjusting weights to reduce error in perceptron model
true
true or false: Both linear and logistic regression attempt to minimize parameter w and b
true
true or false: Causal inference can be applied to improve individual treatment effects in patients that may be superior to group level treatment approach
true
true or false: Dimensionality reduction is a type of feature extraction technique
true
true or false: Gradient descent is an iterative process to minimize errors that is used in both linear and logistic regression
true
true or false: Perceptron model process implies a supervised learning algorithm that solves classification problem
true
true or false: Phonetics primarily focus on the sound or acoustic patterns in languages
true
true or false: Retrieving information about number of features without knowing the actual size of observations may lead to inaccurate dimensionality assessments
true
true or false: The dataset represented below is considered high-dimensional data 6 features, 3 observations
true
true or false: supervised learning algorithms typically produce higher accuracies when predicting Alzheimer's disease patients than patients with mild cognitive impairment versus healthy controls
true
true or false: tumor cell type and size are some examples of input features in study framework to test a PBPK model in tumor-bearing mice
true
logistic regression in neural networks
used within an activation function to determine whether the output of a neuron is on or off
An ordered list of scalar values in 1-D matrix or array shape and usually representation of attributes/features is a definition of: a) Matrix b) Tensor c) Vector d) Scalar
vector
factual outcomes
what it is actually observed
counterfactual outcomes
what would have happened
input layer
where data enters the network
hidden layer
where data is processed between the input and output layers
