CAIS USAII Practice Questions

Question #6 (Topic: demo questions)

In the context of large language models, what is the significance of zero-shot learning?

A.

It enables the model to generate output without any input prompts

B.

It reduces the number of parameters needed for new tasks

C.

It improves the model's performance on tasks with limited data

D.

It allows the model to perform tasks without being explicitly trained on them

Correct Answer: D

Explanation:

Zero-shot learning allows large language models to perform tasks without being explicitly trained on
them. This capability is a result of the broad generalization power these models acquire during pretraining,
enabling them to adapt to new tasks based on their learned knowledge.

Question #7 (Topic: demo questions)

What is a key advantage of using cross-validation over a single train-test split for model evaluation?

A.

It significantly reduces thecomputational cost compared to single train-test split.

B.

It guarantees a perfectly balanced dataset for training and testing.

C.

It provides a more robust estimate of model performance by averaging results over multiple folds.

D.

It eliminates the need for hyperparameter tuning during evaluation.

Correct Answer: C

Explanation:

Cross-validation provides a more robust estimate of model performance by averaging results across
multiple folds. This approach reduces the variance in the performance estimate and ensures that the
model is evaluated on different subsets of the data, making it more reliable than a single train-test split.

Question #8 (Topic: demo questions)

Which optimization technique adapts learning rates individually for each parameter?

A.

Adadelta

B.

SGD

C.

Adam

D.

Nesterov

Correct Answer: C

Explanation:

Adam (Adaptive Moment Estimation) computes adaptive learning rates for each parameter by
considering both the first and second moments of the gradient, enhancing convergence speed and
performance.

Question #9 (Topic: demo questions)

In image classification, which technique can be used to prevent the model from memorizing the training
data and ensure it generalizes well to new data?

A.

Increasing Dataset Size

B.

Applying Advanced Optimization Algorithms

C.

Using Larger Models

D.

Regularization Techniques such as Dropout and L2 Regularization

Correct Answer: D

Explanation:

Regularization techniques like Dropout and L2 Regularization help prevent the model from memorizing
the training data (overfitting) and ensure it generalizes better to new, unseen data by adding constraints
to the learning process.

Question #10 (Topic: demo questions)

Which method is used to find the optimal weights in logistic regression?

A.

Maximum likelihood estimation is applied to find weights.

B.

The least squares methodis used for best fit.

C.

Gradient descent is applied to minimize the loss function.

D.

Stochastic gradient descent is used to avoid overfitting.

Next Question

Correct Answer: A

Explanation:

Logistic regression uses Maximum Likelihood Estimation (MLE) to determine the optimal weights. MLE
seeks to maximize the likelihood that the observed data fits the model, effectively finding the best
parameter estimates for classification tasks.

USAII CAIS - Certified Artificial Intelligence Scientist (CAIS) Certification Exam

In the context of large language models, what is the significance of zero-shot learning?

It enables the model to generate output without any input prompts

It reduces the number of parameters needed for new tasks

It improves the model's performance on tasks with limited data

It allows the model to perform tasks without being explicitly trained on them

Correct Answer: D

Zero-shot learning allows large language models to perform tasks without being explicitly trained onthem. This capability is a result of the broad generalization power these models acquire during pretraining,enabling them to adapt to new tasks based on their learned knowledge.

What is a key advantage of using cross-validation over a single train-test split for model evaluation?

It significantly reduces thecomputational cost compared to single train-test split.

It guarantees a perfectly balanced dataset for training and testing.

It provides a more robust estimate of model performance by averaging results over multiple folds.

It eliminates the need for hyperparameter tuning during evaluation.

Correct Answer: C

Which optimization technique adapts learning rates individually for each parameter?

Adadelta

SGD

Adam

Nesterov

Correct Answer: C

Adam (Adaptive Moment Estimation) computes adaptive learning rates for each parameter byconsidering both the first and second moments of the gradient, enhancing convergence speed andperformance.

In image classification, which technique can be used to prevent the model from memorizing the trainingdata and ensure it generalizes well to new data?

Increasing Dataset Size

Applying Advanced Optimization Algorithms

Using Larger Models

Regularization Techniques such as Dropout and L2 Regularization

Correct Answer: D

Regularization techniques like Dropout and L2 Regularization help prevent the model from memorizingthe training data (overfitting) and ensure it generalizes better to new, unseen data by adding constraintsto the learning process.

Which method is used to find the optimal weights in logistic regression?

Maximum likelihood estimation is applied to find weights.

The least squares methodis used for best fit.

Gradient descent is applied to minimize the loss function.

Stochastic gradient descent is used to avoid overfitting.

Correct Answer: A

Logistic regression uses Maximum Likelihood Estimation (MLE) to determine the optimal weights. MLEseeks to maximize the likelihood that the observed data fits the model, effectively finding the bestparameter estimates for classification tasks.

Zero-shot learning allows large language models to perform tasks without being explicitly trained on
them. This capability is a result of the broad generalization power these models acquire during pretraining,
enabling them to adapt to new tasks based on their learned knowledge.

Adam (Adaptive Moment Estimation) computes adaptive learning rates for each parameter by
considering both the first and second moments of the gradient, enhancing convergence speed and
performance.

In image classification, which technique can be used to prevent the model from memorizing the training
data and ensure it generalizes well to new data?

Regularization techniques like Dropout and L2 Regularization help prevent the model from memorizing
the training data (overfitting) and ensure it generalizes better to new, unseen data by adding constraints
to the learning process.

Logistic regression uses Maximum Likelihood Estimation (MLE) to determine the optimal weights. MLE
seeks to maximize the likelihood that the observed data fits the model, effectively finding the best
parameter estimates for classification tasks.