AI & ML
A student says, “Every AI system learns from data.” Is that statement correct? Explain briefly.
No. Some AI systems use explicitly programmed rules rather than learning from data. Machine learning is one approach within AI.
An instructor groups students according to similar study habits. No performance labels or predefined group names are provided. Which learning approach fits this task?
Unsupervised learning, because the goal is to discover groups without predefined labels.
Two students measure the same object several times. Their measurements are very close to one another but consistently above the object’s true length. Are their measurements precise, accurate, both, or neither?
They are precise but not accurate: the measurements are consistent but not close to the true value.
A university combines thousands of individual dining transactions to calculate the total number of purchases made each day. Which preprocessing technique is this?
Aggregation.
A recommendation system estimates how many minutes a user will watch a video. Is this classification or regression?
Regression, because the output is a numerical value.
A game-playing agent makes a move, receives a reward or penalty, and adjusts its future decisions. Which learning approach is being used?
Reinforcement learning.
A survey dataset records one participant’s age as 21, another participant’s age as “twenty-one,” and another’s as “21 years.” What data-quality problem should be addressed?
Inconsistent values or formats. The ages should be converted into a consistent format before analysis.
A researcher has one million records but selects 10,000 to explore patterns before analyzing the full dataset. Which preprocessing technique is being used, and what is one concern?
Sampling. The sample may not represent the full dataset, especially if certain groups or patterns are underrepresented.
A model predicts whether a passenger survived the Titanic using ticket class, age, and fare. Identify the input features and the target.
The features are ticket class, age, and fare. The target is whether the passenger survived.
A dataset contains photos labeled with the type of clothing shown. A team removes the labels and groups the photos based on visual similarities. Is the task supervised or unsupervised? Explain.
Unsupervised learning, because the team is grouping the photos without using their labels to train the model.
A dataset has 10,000 records, and only five are missing a nonessential value. Give one reasonable way to handle those records and one possible drawback.
One option is to remove the five records, which is simple but discards their other information. Another is to estimate the missing values, but the estimates may not reflect the true values. The best choice depends on the analysis.
A dataset contains 50 features describing each customer. A team uses a technique to represent the data with fewer dimensions while trying to retain important patterns.
Dimensionality reduction.
A team uses labeled examples to teach a model to recognize handwritten digits. Afterward, they ask it to recognize digits it has never seen. Why is it important that these new examples were not used during training?
Unseen examples help evaluate whether the model generalizes to new data, rather than only performing well on examples it learned from.
A company has customer data and wants to do two things: group customers by shopping behavior and predict whether a customer will cancel their membership. The historical data includes whether past customers canceled. Which learning approach fits each task?
Unsupervised learning for grouping customers and supervised learning for predicting cancellation from historical labeled examples.
A dataset has missing income values primarily for people with higher incomes. Why might replacing every missing income with the overall average produce misleading results?
The missing values may not be representative of the observed values. Replacing them with the overall average could underestimate higher incomes and distort the distribution or relationships in the data.
A student selects 100 records from a dataset of 10,000 students. The sample contains only first-year students, even though the full dataset includes all four class years. Which preprocessing technique did the student use, and why might the result be misleading?
The student used sampling. The sample may not represent the full student population, so patterns found in it may not apply to students in other class years.