AI0-001 AI Concepts and Foundations Practice Test 2 — 15 Questions

Question 1

Refer to the exhibit. A deep learning model is being trained. Based on the training log, which problem is most evident?

Accepted Answer

Overfitting. The training log shows that the training loss continues to decrease while the validation loss increases after a certain epoch, which is a classic sign of overfitting. The model is memorizing the training data rather than learning generalizable patterns, leading to poor performance on unseen data.

Answer

Vanishing gradients

Answer

Underfitting

Answer

Data leakage

Question 2

Refer to the exhibit. A data scientist defines a model configuration in JSON. Which component is missing from the configuration for a complete machine learning pipeline?

Accepted Answer

Data preprocessing steps. A complete machine learning pipeline must include data preprocessing steps to transform raw data into a format suitable for model training. The JSON configuration defines the model type, evaluation metrics, and training hyperparameters, but omits any specification for data cleaning, normalization, feature encoding, or splitting, which are essential for reproducibility and model performance.

Answer

Training hyperparameters

Answer

Model type

Answer

Evaluation metrics

Question 3

A hospital uses an AI system to prioritize patient triage based on vital signs and medical history. During a trial, the system consistently assigns lower urgency to elderly patients with chronic conditions, even when their symptoms suggest high risk. Which approach best addresses this bias?

Accepted Answer

Audit the training data for representation of elderly patients and retrain with balanced data. The bias originates from the training data underrepresenting elderly patients with chronic conditions, causing the model to learn skewed urgency patterns. Auditing the data for representation and retraining with balanced data directly addresses the root cause by ensuring the model learns from a fair distribution of cases, which is a standard bias mitigation technique in AI systems.

Answer

Use a different dataset from a similar hospital without checking demographics

Answer

Manually increase the weight of age-related features in the model

Answer

Replace the neural network with a decision tree to simplify decision logic

Question 4

An e-commerce company deploys a recommendation system using collaborative filtering. After launch, the system shows high accuracy for popular items but fails to recommend niche products to users who would likely buy them. Which technique should the team implement to improve recommendations for long-tail items?

Accepted Answer

Switch to a hybrid filtering approach that incorporates item metadata. Collaborative filtering relies on user-item interactions, which are sparse for niche products (the long tail). A hybrid filtering approach that incorporates item metadata (e.g., category, description, attributes) can bridge the gap by using content-based signals to recommend niche items even when interaction data is limited. This directly addresses the cold-start and sparsity problems for long-tail items.

Answer

Apply matrix factorization with higher latent factors

Answer

Increase the weight of popular items in the recommendation score

Answer

Collect more user interaction data over time

Question 5

A data scientist is training a binary classification model to detect fraudulent transactions. The dataset has 99% legitimate transactions and 1% fraudulent. The model achieves 99% accuracy but fails to catch most fraud. Which metric should the team prioritize to evaluate model performance?

Accepted Answer

Recall. Recall (sensitivity) measures the proportion of actual positive cases (fraud) correctly identified. With 99% accuracy but failing to catch most fraud, the model is biased toward the majority class (legitimate transactions), so recall is the critical metric to ensure fraud detection improves.

Answer

F1 score

Answer

Precision

Answer

Accuracy

Question 6

A startup is building a chatbot to handle customer inquiries. They want the chatbot to understand context and provide accurate responses without requiring extensive labeled data. Which AI approach is most suitable?

Accepted Answer

Transfer learning with a pre-trained transformer model. Transfer learning with a pre-trained transformer model (e.g., BERT, GPT) is the most suitable approach because it allows the chatbot to understand context and generate accurate responses using knowledge learned from vast general-domain text, requiring only minimal fine-tuning on the startup's specific customer inquiry data. This eliminates the need for extensive labeled datasets, as the model already captures nuanced language patterns and contextual relationships through its self-attention mechanism.

Answer

Reinforcement learning from human feedback

Answer

Rule-based natural language processing

Answer

Convolutional neural networks (CNNs)

Question 7

Which TWO of the following are key characteristics of unsupervised learning?

Accepted Answer

It uses data without labeled responses. Unsupervised learning algorithms, such as k-means clustering or hierarchical clustering, operate exclusively on input data that has no labeled responses. The model must infer the underlying structure directly from the features without any ground-truth outputs to guide it, which is the defining characteristic of unsupervised learning.

Answer

It predicts a target variable based on input features

Answer

It requires a reward signal to learn optimal actions

Answer

It typically requires a separate validation set for tuning

Question 8

A company uses the above policy to control AI model access. A data scientist tries to run inference with model "llama-3-70b" at 150 requests in 30 minutes. What will happen?

Accepted Answer

The first 100 requests are allowed; the remaining 50 are denied. The policy allows up to 100 requests per 30 minutes for models in the allowed list, and 'llama-3-70b' is in that list. The rate limit is applied per 30-minute window, not per minute, so the first 100 requests are allowed, and the remaining 50 exceed the limit and are denied.

Answer

All requests are allowed because the model is in the allowed list

Answer

All requests are denied because the rate limit is per minute and 150 exceeds the limit

Answer

All requests are denied because the second rule blocks all models

Question 9

A manufacturing company uses a computer vision AI to inspect products on an assembly line for defects. The AI model was trained on images from a single camera angle under bright, uniform lighting. Recently, the company moved the inspection station to a different part of the factory where lighting is dimmer and varies due to nearby windows. The model now misclassifies many non-defective products as defective, causing false alarms and production delays. The team has limited labeled data from the new environment. Which action should the team take to restore inspection accuracy while minimizing downtime?

Accepted Answer

Apply domain adaptation techniques using a small set of labeled images from the new environment. Domain adaptation techniques allow a model trained on a source domain (bright, uniform lighting) to generalize to a target domain (dim, variable lighting) using only a small set of labeled images from the new environment. This approach minimizes downtime because it avoids the need for large-scale data collection or retraining from scratch, and it directly addresses the distribution shift that causes false positives.

Answer

Increase the defect classification threshold to reduce false positives

Answer

Revert to the previous lighting setup by reinstalling bright, uniform lights

Answer

Retrain the model from scratch using a large dataset of images from the new environment

Question 10

A financial institution uses a machine learning model to approve loan applications. The model was trained on historical data that inadvertently encoded a bias against applicants from certain zip codes, leading to discriminatory lending practices. A recent audit reveals that the model's decisions are unfair, and regulators require the bank to remediate the bias without significantly reducing overall approval accuracy. The data science team has access to the training data, the model, and a set of fairness metrics. They also have a small, unbiased validation set. Which course of action should the team take to satisfy regulatory requirements?

Accepted Answer

Implement adversarial debiasing using the unbiased validation set to enforce fairness constraints. Adversarial debiasing directly addresses the bias encoded in the model by training a predictor and an adversary simultaneously. The adversary tries to predict the protected attribute (e.g., zip code) from the model's predictions, while the predictor is penalized for allowing such inference, enforcing fairness constraints. Using the unbiased validation set ensures the debiasing process is guided by ground truth labels that are free from historical bias, allowing the model to retain high accuracy while reducing discrimination.

Answer

Remove the zip code feature from the model and retrain

Answer

Increase the weight of samples from disadvantaged zip codes in the training data

Answer

Retrain the model using only the unbiased validation set

Question 11

A retail company wants to build a model to predict customer churn based on purchase history and demographics. The dataset includes categorical features like region and gender, and numerical features like total spend. What is the best initial step before training the model?

Accepted Answer

One-hot encode categorical variables and normalize numerical variables. One-hot encoding categorical variables and normalizing numerical variables is standard preprocessing to convert categorical data into numeric format and scale features, which many algorithms require for optimal performance.

Answer

Train a deep neural network directly on raw data

Answer

Remove all categorical features to simplify the model

Answer

Perform principal component analysis (PCA) on all features

Question 12

An AI system is being designed to automatically detect fraudulent transactions in real-time. The system must have low latency and high precision to minimize false alarms. Which algorithm is most appropriate?

Accepted Answer

Random forest. Random forest is the most appropriate algorithm because it handles high-dimensional transaction data, provides feature importance for interpretability, and achieves high precision with low latency through ensemble decision trees. Its parallelizable structure allows real-time scoring, and it naturally balances precision and recall without the computational overhead of deep learning.

Answer

Logistic regression

Answer

Convolutional neural network

Answer

Deep reinforcement learning

Question 13

A data scientist trains a deep neural network for image classification. The training loss decreases but validation loss starts increasing after 50 epochs. What should the data scientist do to improve generalization?

Accepted Answer

Apply dropout and early stopping. The increasing validation loss while training loss decreases is a classic sign of overfitting. Dropout randomly deactivates neurons during training, which prevents co-adaptation and forces the network to learn more robust features. Early stopping halts training when validation performance stops improving, directly addressing the overfitting by selecting the model with the best generalization before it degrades.

Answer

Decrease batch size

Answer

Add more hidden layers

Answer

Increase learning rate

Question 14

A company wants to deploy a chatbot that uses natural language understanding (NLU) to answer customer queries. Which AI technique is most suitable for understanding the intent of user input?

Accepted Answer

Sequence-to-sequence model with attention. Sequence-to-sequence models with attention are specifically designed to handle variable-length input sequences (like user queries) and map them to output sequences (like intent labels or responses). The attention mechanism allows the model to focus on the most relevant parts of the input when determining intent, which is critical for understanding nuanced or long user queries in NLU tasks.

Answer

K-means clustering

Answer

Linear regression

Answer

Decision tree

Question 15

An AI model is trained to predict loan default. The training data contains 95% non-default and 5% default. Which metric is most appropriate to evaluate model performance given the imbalanced dataset?

Accepted Answer

F1-score. The F1-score is the harmonic mean of precision and recall, making it robust to class imbalance. In this dataset with 95% non-default and 5% default, accuracy would be misleadingly high (95%) even if the model never predicts default, while F1-score penalizes poor recall of the minority class.

Answer

Mean squared error

Answer

Accuracy

Answer

R-squared