Machine Learning Fundamentals Assessment — Questions and Answers

Question 1: What is a dendrogram used for in hierarchical clustering?

To show the silhouette score for each cluster
To select the optimal epsilon for DBSCAN
To plot the decision boundary between clusters
To visualize the merging sequence and distances between clusters (Correct answer)

Correct answer: To visualize the merging sequence and distances between clusters

A dendrogram is a tree diagram that records the sequence and distance at which clusters are merged, helping select the number of clusters.

Question 2: Which metric is most appropriate when false negatives are far more costly than false positives, such as in cancer screening?

Precision
Specificity
Recall (Correct answer)
Accuracy

Correct answer: Recall

Recall (sensitivity) measures how many actual positives are correctly identified, making it critical when missing a positive (false negative) has severe consequences.

Question 3: Which of the following correctly describes 'robust scaling' (RobustScaler)?

Applies a log transform before min-max scaling
Scales features using the mean and standard deviation
Clips feature values to a fixed range
Scales features using the median and interquartile range, making it resistant to outliers (Correct answer)

Correct answer: Scales features using the median and interquartile range, making it resistant to outliers

RobustScaler subtracts the median and divides by the IQR, so outliers have minimal influence on the scaling parameters.

Question 4: In elastic net regression, which two penalties are combined?

L0 and L1
L1 and L∞
L1 and L2 (Correct answer)
L2 and L3

Correct answer: L1 and L2

Elastic net combines L1 (Lasso) and L2 (Ridge) penalties, balancing sparsity and coefficient shrinkage.

Question 5: What does 'instruction tuning' (or 'instruction fine-tuning') accomplish for large language models?

It converts the model to a smaller distilled version
It applies reinforcement learning from human feedback
It trains the model to follow natural language task instructions across diverse tasks (Correct answer)
It compresses model weights for faster inference

Correct answer: It trains the model to follow natural language task instructions across diverse tasks

Instruction tuning fine-tunes an LLM on (instruction, output) pairs across many tasks, dramatically improving its ability to generalize to new task descriptions.

Question 6: When creating polynomial features from two features x1 and x2 of degree 2, which interaction term is included?

x1 + x2
x1 - x2
x1 * x2 (Correct answer)
x1 / x2

Correct answer: x1 * x2

Degree-2 polynomial features include squared terms (x1², x2²) and the cross-product interaction term x1*x2.

Question 7: What is a hyperparameter in the context of training a deep neural network?

A configuration value set before training that controls the learning process (Correct answer)
A parameter stored in the model's output layer
A weight learned during backpropagation
The output of the final activation function

Correct answer: A configuration value set before training that controls the learning process

Hyperparameters like learning rate, batch size, and number of layers are chosen before training and are not updated by the optimizer.

Question 8: Which cross-validation approach is most appropriate for time-series data?

Stratified k-fold
Walk-forward (time-series) cross-validation (Correct answer)
LOOCV
Standard k-fold

Correct answer: Walk-forward (time-series) cross-validation

Walk-forward validation respects temporal order by always training on past data and testing on future data, preventing look-ahead bias.

Question 9: Which phenomenon occurs when a classifier performs well on training data but fails to generalize to unseen examples?

Concept Drift
Overfitting (Correct answer)
Covariate Shift
Underfitting

Correct answer: Overfitting

Overfitting means the model has memorized training patterns including noise, resulting in high training accuracy but poor generalization to new data.

Question 10: What is the vanishing gradient problem?

The learning rate becomes too large for gradient descent
Gradients become NaN during backpropagation
Weight updates cause gradients to oscillate
Gradients shrink exponentially as they propagate through many layers (Correct answer)

Correct answer: Gradients shrink exponentially as they propagate through many layers

In deep networks, gradients can become exponentially small during backpropagation, making it difficult to train early layers.

Question 11: In Machine Learning certification, what does redundancy in system design primarily provide?

Increased complexity
Fault tolerance and high availability (Correct answer)
Simplified maintenance
Lower initial cost

Correct answer: Fault tolerance and high availability

Redundancy provides fault tolerance by ensuring that if one component fails, backup components maintain system availability.

Question 12: Which technique allows a CNN trained on ImageNet to be adapted for a smaller domain-specific dataset with limited data?

Ensemble learning
Generative pre-training
Transfer learning (Correct answer)
Knowledge distillation

Correct answer: Transfer learning

Transfer learning reuses weights from a model pre-trained on a large dataset, fine-tuning them on the target task to achieve good performance with limited data.

Question 13: What distinguishes a variational autoencoder (VAE) from a standard autoencoder?

VAE learns a probabilistic latent space and can generate new samples by sampling from it (Correct answer)
VAE requires labeled data; standard AE is unsupervised
VAE uses a convolutional encoder; standard AE uses fully connected layers
VAE uses binary cross-entropy; standard AE uses MSE only

Correct answer: VAE learns a probabilistic latent space and can generate new samples by sampling from it

VAE encodes inputs as distributions (mean and variance) over a latent space, enabling smooth interpolation and generation of new data by sampling.

Question 14: In professional documentation for Machine Learning, what is the best practice for organizing information?

Logical structure with clear headings and progression (Correct answer)
Alphabetical order always
Random order of ideas
Longest sections first

Correct answer: Logical structure with clear headings and progression

Logical structure with clear headings helps readers find information quickly and follow the progression of ideas effectively.

Question 15: What is the exploding gradient problem and a common mitigation technique?

Gradients becoming zero; solved by ReLU activation
Loss increasing unboundedly; fixed by dropout
Gradients growing exponentially; mitigated by gradient clipping (Correct answer)
Weights diverging due to large batches; fixed by smaller learning rates

Correct answer: Gradients growing exponentially; mitigated by gradient clipping

Exploding gradients occur when backpropagated values grow exponentially; gradient clipping caps gradient norms to keep updates stable.

Question 16: Which documentation is essential when working with feature engineering in Machine Learning?

General descriptions without specifics
Detailed technical specifications and as-built diagrams (Correct answer)
Only verbal notes
Marketing materials

Correct answer: Detailed technical specifications and as-built diagrams

Detailed technical specifications and as-built diagrams provide the accurate reference information needed for maintenance and troubleshooting.

Question 17: Why is positional encoding added to token embeddings in a transformer?

To separate encoder and decoder embeddings
To normalize the embeddings before attention
To inject information about the order of tokens since transformers have no inherent sequential structure (Correct answer)
To reduce the embedding dimensionality

Correct answer: To inject information about the order of tokens since transformers have no inherent sequential structure

Transformers process all tokens in parallel and have no notion of order, so positional encodings provide sequence position information.

Question 18: Which metric would you use to evaluate a ranking model (e.g., search engine results)?

Cohen's Kappa
RMSE
Accuracy
Normalized Discounted Cumulative Gain (NDCG) (Correct answer)

Correct answer: Normalized Discounted Cumulative Gain (NDCG)

NDCG evaluates the quality of ranked results by considering both relevance and position, discounting lower-ranked items.

Question 19: What is the key advantage of using Mean Absolute Error (MAE) over Mean Squared Error (MSE)?

MAE is always smaller than MSE
MAE is less sensitive to outliers (Correct answer)
MAE penalizes large errors more heavily
MAE is differentiable everywhere

Correct answer: MAE is less sensitive to outliers

MAE uses absolute differences rather than squared differences, making it more robust to outliers compared to MSE.

Question 20: Which decoding strategy introduces randomness by sampling from the top-p cumulative probability mass rather than the full vocabulary?

Beam search
Nucleus (top-p) sampling (Correct answer)
Greedy decoding
Temperature=0 sampling

Correct answer: Nucleus (top-p) sampling

Nucleus sampling restricts the sampling pool to the smallest set of tokens whose cumulative probability exceeds p, balancing diversity and coherence.

Question 21: In Machine Learning practice, what is the primary benefit of using multiple assessment methods?

It is required by law in all cases
It provides a more complete picture of student understanding (Correct answer)
It eliminates the need for grading
It fills more class time

Correct answer: It provides a more complete picture of student understanding

Using multiple assessment methods captures different aspects of student understanding and reduces the limitations of any single measure.

Question 22: Which approach best demonstrates mastery of regression in Machine Learning practice?

Following procedures without understanding
Applying principles to novel situations with sound judgment (Correct answer)
Relying entirely on technology
Avoiding complex scenarios

Correct answer: Applying principles to novel situations with sound judgment

True mastery involves understanding underlying principles well enough to apply them to new and unfamiliar situations with professional judgment.

Question 23: What is 'feature hashing' (the hashing trick) primarily used for?

Normalizing numerical features
Reducing high-cardinality categorical features to a fixed-size vector (Correct answer)
Encrypting sensitive feature values
Detecting duplicate rows in a dataset

Correct answer: Reducing high-cardinality categorical features to a fixed-size vector

Feature hashing maps categories to indices in a fixed-size array using a hash function, controlling dimensionality without a full vocabulary.

Question 24: What is the purpose of a skip connection (residual connection) in ResNets?

To skip certain training examples deemed too easy
To allow gradients and inputs to bypass layers, easing training of very deep networks (Correct answer)
To randomly deactivate entire layers during training
To connect the first layer directly to the output layer

Correct answer: To allow gradients and inputs to bypass layers, easing training of very deep networks

Skip connections add the input of a block directly to its output, providing gradient shortcuts that alleviate vanishing gradients in very deep networks.

Question 25: What does the reward signal represent in reinforcement learning?

The probability of selecting a given action
The gradient of the loss function
The accuracy of predictions made by the model
Immediate scalar feedback indicating how good an action was (Correct answer)

Correct answer: Immediate scalar feedback indicating how good an action was

The reward signal provides immediate scalar feedback to the agent, indicating the desirability of the action taken in the current state.

Question 26: In the context of neural networks, what is backpropagation?

The process of feeding input data forward through the network
The algorithm that computes gradients of the loss with respect to each weight by applying the chain rule backward through the network (Correct answer)
The method for selecting the optimal learning rate
A technique for initializing network weights

Correct answer: The algorithm that computes gradients of the loss with respect to each weight by applying the chain rule backward through the network

Backpropagation efficiently computes gradients for all weights by applying the chain rule of calculus backward from the output loss to each layer's parameters.

Question 27: In Machine Learning practice, reliability in assessment refers to:

The cost of the assessment tool
The speed of administration
The popularity of the instrument
The consistency and reproducibility of results (Correct answer)

Correct answer: The consistency and reproducibility of results

Reliability refers to the consistency and reproducibility of assessment results when the test is repeated under similar conditions.

Question 28: Which self-supervised learning technique for computer vision trains a model to maximize agreement between differently augmented views of the same image?

Contrastive learning (e.g., SimCLR) (Correct answer)
Autoencoders
Knowledge distillation
Generative adversarial training

Correct answer: Contrastive learning (e.g., SimCLR)

Contrastive learning methods like SimCLR pull representations of augmented views of the same image together while pushing apart views from different images.

Question 29: What is instance segmentation, and how does it differ from semantic segmentation?

It assigns one label to the whole image, while semantic segmentation labels pixels
It operates only on foreground objects, while semantic segmentation labels everything
It distinguishes individual object instances with separate masks, while semantic segmentation does not differentiate instances (Correct answer)
It labels each pixel with a class, while semantic segmentation uses bounding boxes

Correct answer: It distinguishes individual object instances with separate masks, while semantic segmentation does not differentiate instances

Instance segmentation identifies and masks each individual object separately, whereas semantic segmentation groups all pixels of the same class together without distinguishing instances.

Question 30: What is the best practice for documenting assessment results in Machine Learning practice?

Document all findings objectively and completely (Correct answer)
Record only abnormal findings
Document results from memory at end of day
Use subjective descriptions only

Correct answer: Document all findings objectively and completely

Best practice requires documenting all findings objectively and completely at the time of assessment for accuracy and legal protection.

Question 31: In topic modeling with Latent Dirichlet Allocation (LDA), what are documents modeled as?

Clusters of similar sentences
Mixtures of topics, where each topic is a distribution over words (Correct answer)
Bag-of-words frequency vectors only
Sequences of named entities

Correct answer: Mixtures of topics, where each topic is a distribution over words

LDA models each document as a mixture of latent topics, and each topic as a probability distribution over vocabulary words.

Question 32: Which evaluation strategy is most suitable when you have very limited data?

Bootstrap with 1000 samples
Leave-One-Out Cross-Validation (LOOCV) (Correct answer)
Hold-out validation
Train-test split 90/10

Correct answer: Leave-One-Out Cross-Validation (LOOCV)

LOOCV uses every sample as a test set exactly once, maximizing training data usage and is ideal for small datasets.

Question 33: What distinguishes a recurrent neural network (RNN) from a feedforward network?

RNNs use different activation functions than feedforward networks
RNNs have connections that feed hidden state back into the next time step (Correct answer)
RNNs have no hidden layers and process data in one pass
RNNs use convolutional filters while feedforward networks do not

Correct answer: RNNs have connections that feed hidden state back into the next time step

RNNs maintain a hidden state passed between time steps, enabling them to model sequential dependencies in time-series or text data.

Question 34: Which technique replaces a categorical feature with the mean of the target variable for each category?

Frequency encoding
Label encoding
Target encoding (Correct answer)
One-hot encoding

Correct answer: Target encoding

Target encoding substitutes each category with the mean target value, which can capture ordinal relationships in high-cardinality features.

Question 35: In Self-Organizing Maps (SOMs), what is the 'neighborhood function' responsible for?

Computing the reconstruction error
Normalizing input features before training
Selecting the number of output neurons
Updating weights of neurons near the winning neuron to preserve topological structure (Correct answer)

Correct answer: Updating weights of neurons near the winning neuron to preserve topological structure

The neighborhood function ensures that neurons near the winning (best-matching) unit are also updated, causing the map to preserve topological relationships of the input data.

Question 36: Which strategy is most effective for promoting student engagement in Machine Learning education?

Memorization-focused activities only
Reading assignments without discussion
Lengthy lectures without interaction
Active learning with meaningful participation (Correct answer)

Correct answer: Active learning with meaningful participation

Active learning strategies that involve meaningful participation increase engagement, retention, and deeper understanding of material.

Question 37: What is the primary advantage of using an ensemble method like Random Forest over a single decision tree?

Simpler model interpretation
Faster training time
Lower variance through averaging multiple trees (Correct answer)
Reduced memory usage

Correct answer: Lower variance through averaging multiple trees

Random Forest reduces variance by averaging predictions from many decorrelated trees built on random subsets of features and data.

Question 38: In logistic regression, what function transforms the linear combination of features into a probability?

Softmax function
ReLU function
Sigmoid function (Correct answer)
Tanh function

Correct answer: Sigmoid function

The sigmoid (logistic) function maps any real-valued number to a probability between 0 and 1.

Question 39: Which of the following is an example of a regression problem in supervised learning?

Classifying handwritten digits 0-9
Predicting the selling price of a house based on its features (Correct answer)
Identifying whether a tumor is malignant or benign
Predicting whether an email is spam or not spam

Correct answer: Predicting the selling price of a house based on its features

Predicting house prices produces a continuous numeric output, making it a regression problem rather than a classification problem.

Question 40: What is the primary goal of a reinforcement learning agent?

To maximize cumulative reward over time (Correct answer)
To label training data accurately
To cluster similar data points together
To minimize classification error on a test set

Correct answer: To maximize cumulative reward over time

An RL agent learns by interacting with an environment to maximize the total cumulative reward it receives over time.

Question 41: What is the value of continuing education in classification for Machine Learning professionals?

It is only needed for recertification
It is primarily a social activity
It replaces workplace experience
It keeps professionals current with evolving standards and practices (Correct answer)

Correct answer: It keeps professionals current with evolving standards and practices

Continuing education ensures professionals stay current with the latest developments, standards, and best practices in their field.

Question 42: What is the value of continuing education in computer vision for Machine Learning professionals?

It keeps professionals current with evolving standards and practices (Correct answer)
It is primarily a social activity
It is only needed for recertification
It replaces workplace experience

Correct answer: It keeps professionals current with evolving standards and practices

Continuing education ensures professionals stay current with the latest developments, standards, and best practices in their field.

Question 43: What is 'data leakage' in a supervised learning pipeline?

Overfitting due to too many training epochs
Training data being stored insecurely
Information from the test set inadvertently influencing model training (Correct answer)
Features being removed from the dataset

Correct answer: Information from the test set inadvertently influencing model training

Data leakage occurs when information outside the training set (e.g., from the test set) is used during training, causing artificially inflated performance metrics.

Question 44: Which technique is used to select the optimal regularization parameter in Ridge or Lasso regression?

Maximum likelihood estimation
Cross-validation (Correct answer)
Bayesian inference
Bootstrapping

Correct answer: Cross-validation

Cross-validation evaluates model performance across different λ values to select the one that minimizes validation error.

Question 45: When using one-hot encoding for categorical features, what problem can arise with high-cardinality variables?

Dimensionality explosion with many sparse binary columns (Correct answer)
The model cannot converge
Values become negative
Labels become continuous

Correct answer: Dimensionality explosion with many sparse binary columns

High-cardinality categorical variables create an extremely large and sparse feature matrix when one-hot encoded, increasing memory and computation costs.

Question 46: What is the role of the 'epsilon' parameter in DBSCAN?

It sets the minimum number of clusters
It defines the maximum radius of a neighborhood around a point (Correct answer)
It specifies the maximum number of iterations
It controls the learning rate

Correct answer: It defines the maximum radius of a neighborhood around a point

Epsilon (ε) defines the radius within which DBSCAN searches for neighboring points to determine core points and cluster membership.

Question 47: What is the best practice for maintaining neural networks performance over time?

Outsource all maintenance
Wait for failures before acting
Implement scheduled preventive maintenance (Correct answer)
Upgrade all equipment annually

Correct answer: Implement scheduled preventive maintenance

Scheduled preventive maintenance catches potential issues before they cause failures, maintaining reliability and extending equipment life.

Question 48: Which loss function is most appropriate for a multi-class classification neural network?

Categorical cross-entropy (Correct answer)
Mean squared error
Hinge loss
Binary cross-entropy

Correct answer: Categorical cross-entropy

Categorical cross-entropy measures the dissimilarity between predicted probability distributions and one-hot encoded true labels in multi-class problems.

Question 49: In gradient boosting, how are successive trees trained?

Independently on random subsets of features
On randomly selected subsets of training samples
On the residual errors of the previous ensemble (Correct answer)
Using the same training data without modification

Correct answer: On the residual errors of the previous ensemble

Each new tree in gradient boosting is trained to predict the residual errors (pseudo-residuals) left by the current ensemble, iteratively improving predictions.

Question 50: What is differentiated instruction in the context of Machine Learning?

Separating students by ability permanently
Teaching the same content the same way to all students
Adjusting teaching methods to meet diverse learner needs (Correct answer)
Using only standardized materials

Correct answer: Adjusting teaching methods to meet diverse learner needs

Differentiated instruction involves tailoring teaching approaches, content, and assessment to accommodate the diverse needs and abilities of learners.

Question 51: What is the purpose of cross-validation in classifier evaluation?

To get a more reliable estimate of model performance by training and testing on multiple data splits (Correct answer)
To increase training data size through augmentation
To automatically tune hyperparameters using gradient descent
To reduce the number of features needed for classification

Correct answer: To get a more reliable estimate of model performance by training and testing on multiple data splits

Cross-validation repeatedly splits data into train/test folds, averaging performance across folds to produce a less biased and more stable generalization estimate.

Question 52: In principal component regression (PCR), what is regressed on the response variable?

The original correlated predictors
Principal components derived from predictors (Correct answer)
Standardized predictor ranks
Residuals from a first-stage regression

Correct answer: Principal components derived from predictors

PCR uses uncorrelated principal components of the predictors as inputs, addressing multicollinearity by projecting into a lower-dimensional space.

Question 53: In DBSCAN, what term describes a point that does not belong to any cluster?

Noise point (Correct answer)
Border point
Orphan point
Core point

Correct answer: Noise point

Noise points (also called outliers) in DBSCAN are points that are neither core points nor reachable from any core point.

Question 54: What is the most important element of effective professional communication in Machine Learning?

Writing lengthy documents
Using complex vocabulary
Clarity and audience-appropriate language (Correct answer)
Avoiding all technical terms

Correct answer: Clarity and audience-appropriate language

Effective communication requires clarity and language appropriate for the audience to ensure the message is understood as intended.

Question 55: A leverage point in regression refers to an observation that:

Causes the R² to decrease
Has a large residual
Introduces multicollinearity
Has an unusual predictor value that strongly influences the fitted line (Correct answer)

Correct answer: Has an unusual predictor value that strongly influences the fitted line

Leverage measures how far an observation's predictor values are from the mean; high-leverage points can disproportionately influence the regression fit.

Question 56: What is the primary advantage of DBSCAN over K-Means clustering?

It can find arbitrarily shaped clusters and identify outliers (Correct answer)
It always produces globally optimal clusters
It requires fewer hyperparameters
It is faster on large datasets

Correct answer: It can find arbitrarily shaped clusters and identify outliers

DBSCAN identifies clusters of arbitrary shape based on density and automatically marks sparse points as noise/outliers.

Question 57: What is the fundamental principle behind feature engineering in the Machine Learning domain?

Cost minimization at all costs
Following a single vendor solution
Balancing performance, reliability, and efficiency (Correct answer)
Using the newest technology exclusively

Correct answer: Balancing performance, reliability, and efficiency

Effective technical design requires balancing performance requirements with reliability needs and operational efficiency.

Question 58: What does the term 'dead neuron' refer to when using ReLU activations?

A neuron that outputs the same value regardless of input
A neuron that always outputs zero because its weights push its input permanently negative (Correct answer)
A neuron removed by the dropout process
A neuron whose weights have not been initialized

Correct answer: A neuron that always outputs zero because its weights push its input permanently negative

A dead ReLU neuron receives only negative inputs, always outputting zero with zero gradient, meaning it can never recover during training.

Question 59: What is the epsilon-greedy strategy used for in reinforcement learning?

To determine an optimal adaptive learning rate for policy updates
To balance exploration and exploitation by taking random actions with probability epsilon (Correct answer)
To normalize reward values between epsilon and 1 for stable training
To set the convergence threshold for stopping training

Correct answer: To balance exploration and exploitation by taking random actions with probability epsilon

The epsilon-greedy strategy selects a random action with probability epsilon (exploration) and the greedy best-known action with probability 1-epsilon (exploitation), balancing discovery and performance.

Question 60: What is 'Cohen's Kappa' used to measure?

Agreement between classifier predictions and actual labels, adjusted for chance (Correct answer)
Variance explained by a regression model
Distance between cluster centroids
Correlation between continuous variables

Correct answer: Agreement between classifier predictions and actual labels, adjusted for chance

Cohen's Kappa measures inter-rater agreement adjusted for the agreement expected by chance, providing a more reliable accuracy measure.

Machine Learning Fundamentals Assessment

The Machine Learning Fundamentals Assessment covers supervised and unsupervised learning, neural networks, deep learning, feature engineering, model evaluation, regression, classification, clustering, natural language processing, and computer vision techniques used in modern ML applications.

Questions

90 min

Time Limit

70%

To Pass

Checking for saved progress...

Exam Rules

You can skip questions and return to them later
Flag questions for review before submitting
No feedback shown until you submit the entire exam
Unanswered questions count as wrong — answer everything
10 pretest questions are mixed in and don't affect your score
Timer auto-submits when time runs out
Your progress is auto-saved every 30 seconds