42698 A ST: ML EXPT BME RES Monday, May 4, 2026 01:00pm - 04:00pm In
Person WEH 7500 WASHBURN NEWELL (washburn) carnegie mellon university
150+
SECTION 1: FOUNDATIONS OF ML IN BME (Questions 1-40)
Question 1
What is the primary difference between supervised and unsupervised learning in
the context of analyzing patient EKG data?
A) Supervised learning requires labeled data (e.g., "heart attack" or "normal"),
while unsupervised learning finds hidden patterns without labels.
B) Unsupervised learning requires labeled data, while supervised learning does
not.
C) Supervised learning is used exclusively for regression, while unsupervised
learning is for classification.
D) There is no difference; both methods require labeling.
Answer: A)
Rationale: Supervised learning relies on labeled input-output pairs to train a
model (e.g., diagnosing a condition from a signal). Unsupervised learning
,identifies inherent structures (e.g., clustering similar EKG morphologies) without
predefined labels .
Question 2
Which type of machine learning is most appropriate for discovering new subtypes
of a disease from genetic expression data without prior knowledge of the
subtypes?
A) Supervised classification
B) Supervised regression
C) Unsupervised clustering
D) Reinforcement learning
Answer: C)
Rationale: Unsupervised clustering is used to find natural groupings within data.
In BME, this is essential for discovering novel disease phenotypes or patient strata
without ground truth labels.
Question 3
A researcher has a dataset of MRI scans labeled "tumor" or "no tumor." What is
the appropriate learning task?
A) Unsupervised clustering
,B) Dimensionality reduction
C) Supervised classification
D) Association rule mining
Answer: C)
Rationale: Because the dataset contains labels (ground truth), the task is
supervised. The goal is classification (discrete categories: tumor vs. no tumor).
Question 4
In the context of biomedical signal processing, which of the following is
considered a "label" for supervised learning?
A) The raw voltage values of an EEG signal
B) A binary indicator of seizure onset annotated by a neurologist
C) The principal components of an ECG signal
D) The Fourier transform of a PPG signal
Answer: B)
Rationale: A label is the ground truth or outcome variable we want to predict.
Clinical annotations (e.g., seizure start/stop times) serve as labels for training
classifiers.
, Question 5
Which of the following best describes a "feature" in a machine learning model for
prostate cancer detection from histopathology images?
A) The final diagnosis (cancerous/benign)
B) The patient's age and medical record number
C) The pixel intensity, texture, or morphology of cell nuclei extracted from the
image
D) The training loss value
Answer: C)
Rationale: Features are measurable properties or characteristics of the raw data
used as input to the model. In medical imaging, these are often derived
representations like texture, shape, or intensity.
Question 6
A deep neural network trained to detect pneumonia on chest X-rays achieves 99%
accuracy on the training set but only 70% on a held-out test set. This is a classic
sign of:
A) Underfitting
B) Overfitting
Person WEH 7500 WASHBURN NEWELL (washburn) carnegie mellon university
150+
SECTION 1: FOUNDATIONS OF ML IN BME (Questions 1-40)
Question 1
What is the primary difference between supervised and unsupervised learning in
the context of analyzing patient EKG data?
A) Supervised learning requires labeled data (e.g., "heart attack" or "normal"),
while unsupervised learning finds hidden patterns without labels.
B) Unsupervised learning requires labeled data, while supervised learning does
not.
C) Supervised learning is used exclusively for regression, while unsupervised
learning is for classification.
D) There is no difference; both methods require labeling.
Answer: A)
Rationale: Supervised learning relies on labeled input-output pairs to train a
model (e.g., diagnosing a condition from a signal). Unsupervised learning
,identifies inherent structures (e.g., clustering similar EKG morphologies) without
predefined labels .
Question 2
Which type of machine learning is most appropriate for discovering new subtypes
of a disease from genetic expression data without prior knowledge of the
subtypes?
A) Supervised classification
B) Supervised regression
C) Unsupervised clustering
D) Reinforcement learning
Answer: C)
Rationale: Unsupervised clustering is used to find natural groupings within data.
In BME, this is essential for discovering novel disease phenotypes or patient strata
without ground truth labels.
Question 3
A researcher has a dataset of MRI scans labeled "tumor" or "no tumor." What is
the appropriate learning task?
A) Unsupervised clustering
,B) Dimensionality reduction
C) Supervised classification
D) Association rule mining
Answer: C)
Rationale: Because the dataset contains labels (ground truth), the task is
supervised. The goal is classification (discrete categories: tumor vs. no tumor).
Question 4
In the context of biomedical signal processing, which of the following is
considered a "label" for supervised learning?
A) The raw voltage values of an EEG signal
B) A binary indicator of seizure onset annotated by a neurologist
C) The principal components of an ECG signal
D) The Fourier transform of a PPG signal
Answer: B)
Rationale: A label is the ground truth or outcome variable we want to predict.
Clinical annotations (e.g., seizure start/stop times) serve as labels for training
classifiers.
, Question 5
Which of the following best describes a "feature" in a machine learning model for
prostate cancer detection from histopathology images?
A) The final diagnosis (cancerous/benign)
B) The patient's age and medical record number
C) The pixel intensity, texture, or morphology of cell nuclei extracted from the
image
D) The training loss value
Answer: C)
Rationale: Features are measurable properties or characteristics of the raw data
used as input to the model. In medical imaging, these are often derived
representations like texture, shape, or intensity.
Question 6
A deep neural network trained to detect pneumonia on chest X-rays achieves 99%
accuracy on the training set but only 70% on a held-out test set. This is a classic
sign of:
A) Underfitting
B) Overfitting
