Machine Learning Flashcards
For artificial intelligence, “human-in-the-loop” refers to which of the following?
Select one:
a.
Humans must be present during the design of artificial intelligence technology.
b.
Humans must be responsible for monitoring the performance of artificial intelligence tools after deployment to ensure that they continue to work as expected.
c.
Humans must be accountable when artificial intelligence algorithms do not perform as expected.
d.
Humans must maintain autonomy when using artificial intelligence technology so that malfunctions can be overridden.
Humans must maintain autonomy when using artificial intelligence technology so that malfunctions can be overridden
“human-in-the-loop” refers to a human being in the decision-making loop of artificial intelligence tools
Which of the following refers to machine learning where some of the training data are manually labeled while the rest of the data are either not labeled or auto-labeled according to an algorithm?
Select one:
a.
Transfer learning
b.
Semi-supervised learning
c.
Fully supervised learning
d.
Fully unsupervised learning
e.
Weakly supervised learning
correct answer is semi-supervised learning.
Fully supervised learning uses training data that have all been manually labeled to the same or similar extent, while weakly supervised learning uses a subset of training data with detailed labels and the remainder with fewer or less detailed labels. Fully unsupervised learning trains on training data that is unlabeled, and semi-unsupervised learning is not a term that is commonly or ever used. Transfer learning is where most of the training data comes from a different but related domain followed by a smaller subset of training data that is from the intended domain of use.
Which of the following best describes reinforcement learning?
Select one:
a.
Unsupervised learning in which the algorithm trains to minimize loss function using the stochastic gradient descent to reinforce the separation of clusters.
Incorrect.
b.
Supervised learning in which all of the labels on the data have been reinforced by a dimensionality reduction algorithm.
c.
Coarse training performed on a related data set with subsequent reinforcement of that training with a more finely tuned data set.
d.
Algorithm-based learning of a policy which causes an agent to execute an action which may result in a reinforcing reward or penalty.
Algorithm-based learning of a policy which causes an agent to execute an action which may result in a reinforcing reward or penalty. Reinforcement learning uses training cycles in which an algorithm (the agent) perceives the state of the environment, takes a specific action with alters the environment, then receives a reward or penalty based on whether it was the correct action to take. It is used to train game-playing models as well as to forecast financial planning.
Which of the following is an example of an expert system?
Select one:
a.
CADUCEUS
b.
Market Basket Analysis
c.
Artificial Neural Networks
d.
K-Nearest Neighbor
CADUCEUS. CADUCEUS is an expert system based on rules programmed into a knowledgebase.
Market Basket Analysis, Artificial Neural Networks and K-Nearest Neighbor are all examples of machine learning and are not expert systems.
Which of the following is a current challenge to the development, deployment and use of artificial intelligence (AI) in healthcare?
Select one:
a.
There are many hurdles to deploying AI tools in a patient care environment where there may be population differences.
Correct!
b.
Computational infrastructure for AI technology has many robust security mitigations which may delay its performance.
c.
There are many legal requirements, and the billing structure is overly complex for AI tools.
d.
Health care has many data scientists, so organizations will not be able to keep up with the testing and deployment of too many AI algorithms.
There are many hurdles to deploying AI tools in a patient care environment where there may be population differences.
Which of the following terms relates to the observation value for an item used in a machine learning data set?
Select one:
a.
Instance
b.
Hyperparameter
c.
Label
d.
Outlier
e.
Feature
Labels refer to the value for the feature (attribute, vector) of a specific instance (item or sample) in a data set. A hyperparameter is a parameter that is manually set by the developer prior to training. An outlier is an instance which is significantly different from the rest of the instances in the population.
According to the National Institutes of Standards and Technology (NIST), the explainability of artificial intelligence is based on which of the following principles?
Select one:
a.
Explanations of AI deliver evidence for user benefits and societal acceptance.
b.
Explaining artificial intelligence to developers in a meaningful way.
c.
Explanations are accurate about what the tool does but does not describe its limitations.
d.
Information is published about the artificial intelligence model prior to deployment.
Explanations of AI deliver evidence for user benefits and societal acceptance. NIST requires that explanations are accurate and include limitations of its performance and the conditions under which it can operate.
Which of the following is a fundamental difference between machine learning and traditional computer programs?
Select one:
a.
Software engineers write analysis rules in a programming language for traditional programs while a computer writes programming language in machine learning.
b.
Machine learning models analyze data using specified rules while traditional programs are a lot more flexible and fluid.
c.
Traditional computer programs analyze output data while machine learning is only able to discern patterns in input data.
d.
In order to troubleshoot, a traditional computer program may have its code reviewed whereas a machine learning algorithm may be fed more data.
order to troubleshoot, a traditional computer program may have its code reviewed whereas a machine learning algorithm may be fed more data. The rules that a machine learning model uses for analyzing data may be indecipherable, even to the model developer.
Therefore, if an end-user or developer is not sure why a particular output was generated by the model, they may decide to feed more data to the model to try to decipher the model’s actions. Contrast this with traditional programming where the programming code may be reviewed to determine if a programming error is present.
Machine learning models are trained and develop their own mathematical functions or rules to analyze new data, but these rules are often unknown to the end-user or developer. Traditional programs, by contrast, are much more explicit.
Machine learning models do not write their own programming code during training. Instead, they send data through functions and adjust parameters to either maximize the advantage or minimize the error (or loss function).
Traditional computer programs also do not typically analyze output data whereas this is common in supervised machine learning models when they are being trained.
Which of the following best characterizes the differences between expert systems and machine learning?
Select one:
a.
Expert systems use algorithms to learn repetitive data patterns while machine learning does not.
b.
Machine learning allows users to derive knowledge from knowledgebases and a rules engine.
c.
Machine learning requires more computational resources than expert systems, but they can analyze more data.
d.
Machine learning is able to handle large sets of data, but expert systems are better at handling data in complex settings.
C. correct answer is Machine learning requires more computational resources than expert systems, but they can analyze more data.
Expert systems are human knowledgebases programmed with explicit If-Then Statements and a rules engine and do not learn from the data unlike machine learning algorithms.
Machine learning is better able to handle both large amounts of data as well as complex data compared to expert systems, but they also require more computational resources.
Tesler’s theorem
to end-users, artificial intelligence is “whatever hasn’t been done yet” and is a cognitive bias which perceives software in common use as not being artificial intelligence, even when it is
availability heuristic
cognitive bias where people base their prediction of an outcome on the vividness and emotional impact of the outcome rather than its actual probability
Which of the following is a general characteristic of machine learning?
Select one:
a.
Produces an explicit mathematical relationship between inputs and outputs
b.
Is able to produce a clear reason for its output
c.
Needs to know the characteristics and distribution of data prior to analyzing it
Incorrect.
d.
Handles a large number of complex input variables
Handles a large number of complex input variables.
Which of the following is the most common reason why artificial intelligence algorithms have failed after deployment secondary to exacerbated racial, ethnic, socioeconomic, gender or other forms of bias?
Select one:
a.
The artificial intelligence algorithm was not monitored after it was deployed.
b.
The data used to train the model did not include underrepresented populations.
c.
The data used to train the model included outcomes which were influenced by human bias.
d.
Humans were not part of the artificial intelligence decision making process (i.e., there was no “human-in-the-loop”).
data used to train the model included outcomes which were influenced by human bias.
