IBM AI Engineering Professional Certificate Quiz Answers 2026 (102+ MCQs)

ibm ai engineering professional certificate answers - Feature Image
102+
Quiz MCQs
13
Courses Covered
100%
Free Access
2026
Updated For
The IBM AI Engineering Professional Certificate on Coursera is a 13-course programme that takes you from machine learning with Python through deep learning in Keras, TensorFlow and PyTorch to generative AI, fine-tuning transformers, RAG and LangChain. Every course has graded module quizzes and a final exam. The 102+ solved MCQs below follow the certificate course by course, with the correct option marked and a short explanation under every question, so you can revise before each graded quiz. New to AI? Start with IBM AI Fundamentals first. A free PDF of the full set is at the end of the page.

ย 

Question 01

In machine learning, what is the difference between supervised and unsupervised learning?

ASupervised uses labelled data; unsupervised finds patterns in unlabelled data โœ…
BSupervised is faster
CUnsupervised needs more labels
DThere is no difference
๐Ÿ’ก ExplanationSupervised models learn a mapping from inputs to known outputs; unsupervised models discover structure such as clusters.

Question 02

Which of the following is a regression problem?

APredicting whether an email is spam
BPredicting the CO2 emission of a car from its engine size โœ…
CGrouping customers into segments
DDetecting objects in an image
๐Ÿ’ก ExplanationRegression predicts a continuous value; the course uses the CO2 emissions dataset for this.

Question 03

Which of the following is a classification problem?

APredicting house prices
BPredicting whether a tumour is benign or malignant โœ…
CForecasting temperature
DEstimating salary
๐Ÿ’ก ExplanationClassification predicts a discrete category.

Question 04

In simple linear regression, the model fits:

AA curve through all points
BA straight line that minimises the sum of squared errors โœ…
CA decision boundary
DClusters
๐Ÿ’ก ExplanationOrdinary least squares finds the line with the lowest mean squared error.

Question 05

Which metric is commonly used to evaluate a regression model?

AAccuracy
BMean Squared Error (MSE) โœ…
CF1 score
DJaccard index
๐Ÿ’ก ExplanationMSE, RMSE, MAE and R-squared are regression metrics.

Question 06

What does R-squared measure?

AClassification accuracy
BThe proportion of variance in the target explained by the model โœ…
CThe number of features
DThe learning rate
๐Ÿ’ก ExplanationR-squared of 1 means a perfect fit; 0 means no better than the mean.

Question 07

What is the purpose of splitting data into training and test sets?

ATo make the dataset smaller
BTo estimate how the model performs on unseen data โœ…
CTo speed up training
DTo remove outliers
๐Ÿ’ก ExplanationOut-of-sample evaluation shows how well the model generalises.

Question 08

Which algorithm classifies a point based on the majority class of its nearest neighbours?

ALogistic regression
BK-Nearest Neighbours (KNN) โœ…
CK-means
DLinear regression
๐Ÿ’ก ExplanationKNN is a lazy, instance-based classifier.

Question 09

In KNN, what happens if K is too small?

AThe model underfits
BThe model becomes sensitive to noise and overfits โœ…
CTraining takes longer
DAccuracy always improves
๐Ÿ’ก ExplanationA small K captures noise; a very large K over-smooths.

Question 10

A decision tree chooses the attribute to split on by:

ARandom selection
BMaximising information gain (reducing entropy) โœ…
CAlphabetical order
DMinimising the number of nodes
๐Ÿ’ก ExplanationEntropy measures impurity; the best split lowers it most.

Question 11

Logistic regression is used for:

APredicting continuous values
BBinary classification, outputting a probability โœ…
CClustering
DDimensionality reduction
๐Ÿ’ก ExplanationIt applies the sigmoid function to a linear combination of features.

Question 12

The sigmoid function outputs values between:

A-1 and 1
B0 and 1 โœ…
C0 and infinity
D-infinity and infinity
๐Ÿ’ก ExplanationThe output can be read as a probability.

Question 13

What does a Support Vector Machine (SVM) try to find?

AThe shortest tree
BThe hyperplane with the maximum margin between classes โœ…
CThe nearest neighbours
DThe centroid of each cluster
๐Ÿ’ก ExplanationSVM maximises the distance between the hyperplane and the closest points (support vectors).

Question 14

What is the ‘kernel trick’ in SVM?

ARemoving features
BMapping data into a higher-dimensional space to make it linearly separable โœ…
CSpeeding up training
DReducing the dataset
๐Ÿ’ก ExplanationCommon kernels: linear, polynomial, RBF, sigmoid.

Question 15

K-means is an example of:

ASupervised classification
BUnsupervised clustering โœ…
CRegression
DReinforcement learning
๐Ÿ’ก ExplanationK-means partitions data into K clusters around centroids.


Question 16

In K-means, the algorithm stops when:

AAccuracy reaches 100%
BCentroids no longer move significantly โœ…
CK equals the number of points
DThe tree is fully grown
๐Ÿ’ก ExplanationIt iterates assign-and-update until convergence.

Question 17

Which metric evaluates a classification model by comparing predicted and actual labels in a table?

AR-squared
BConfusion matrix โœ…
CMSE
DSilhouette score
๐Ÿ’ก ExplanationFrom it you compute precision, recall and F1.

Question 18

Precision is defined as:

ATP / (TP + FN)
BTP / (TP + FP) โœ…
CTN / (TN + FP)
D(TP + TN) / total
๐Ÿ’ก ExplanationPrecision is the share of predicted positives that are correct.

Question 19

Recall is defined as:

ATP / (TP + FP)
BTP / (TP + FN) โœ…
CTN / (TN + FN)
DFP / (FP + TN)
๐Ÿ’ก ExplanationRecall is the share of actual positives that were found.

Question 20

Which Python library is mainly used for classical machine learning in the course?

ATensorFlow
Bscikit-learn โœ…
CPyTorch
DKeras
๐Ÿ’ก Explanationscikit-learn provides regression, classification, clustering and evaluation tools.

Question 21

What is overfitting?

AThe model is too simple
BThe model fits training data too closely and performs poorly on new data โœ…
CThe model trains too fast
DThe data is too large
๐Ÿ’ก ExplanationHigh training accuracy with low test accuracy signals overfitting.

Question 22

Which technique helps evaluate a model more reliably than a single train/test split?

AUsing all data for training
BK-fold cross-validation โœ…
CRemoving the test set
DIncreasing K in KNN
๐Ÿ’ก ExplanationEach fold serves once as the test set.

Next: Introduction to Deep Learning and Neural Networks with Keras โ†’

ย 
๐Ÿง 

Course 2 โ€” Introduction to Deep Learning and Neural Networks with Keras (Questions 23โ€“42)

Neurons, forward and backpropagation, activations, Keras models, CNNs, RNNs and autoencoders

Question 23

What is an artificial neuron’s output computed from?

AThe sum of inputs only
BA weighted sum of inputs plus a bias, passed through an activation function โœ…
CA random number
DThe number of layers
๐Ÿ’ก Explanationz = w.x + b, then a = f(z).

Question 24

What is forward propagation?

AUpdating weights
BPassing input data through the network layer by layer to produce an output โœ…
CShuffling data
DRemoving neurons
๐Ÿ’ก ExplanationEach layer transforms the previous layer’s output.

Question 25

What is backpropagation?

AMoving data backwards to the user
BComputing gradients of the loss with respect to weights and propagating them backward to update weights โœ…
CDeleting layers
DNormalising inputs
๐Ÿ’ก ExplanationIt applies the chain rule from the output layer back to the input.

Question 26

What does gradient descent do?

AIncreases the loss
BIteratively adjusts weights in the direction that reduces the loss โœ…
CChooses the activation function
DAdds layers
๐Ÿ’ก ExplanationThe learning rate controls the step size.

Question 27

What is the ‘vanishing gradient’ problem?

AGradients become too large
BGradients become very small in early layers, slowing or stopping learning โœ…
CWeights are deleted
DThe dataset is too small
๐Ÿ’ก ExplanationIt occurs with sigmoid/tanh activations in deep networks.

Question 28

Which activation function helps overcome the vanishing gradient problem?

ASigmoid
BReLU โœ…
CTanh
DStep function
๐Ÿ’ก ExplanationReLU outputs max(0, x) and does not saturate for positive inputs.

Question 29

The softmax function is typically used in:

AHidden layers
BThe output layer for multi-class classification โœ…
CThe input layer
DRegression outputs
๐Ÿ’ก ExplanationIt converts scores into probabilities that sum to 1.

Question 30

In Keras, which class is used to build a model layer by layer?

ASequential โœ…
BFunctional
CPipeline
DLayerwise
๐Ÿ’ก Explanationmodel = Sequential(); model.add(Dense(…)).


Question 31

Which Keras layer is a fully connected layer?

AConv2D
BDense โœ…
CMaxPooling2D
DLSTM
๐Ÿ’ก ExplanationDense connects every input to every output neuron.

Question 32

For a regression model in Keras, a common loss function is:

Acategorical_crossentropy
Bmean_squared_error โœ…
Cbinary_crossentropy
Dhinge
๐Ÿ’ก ExplanationMSE is standard for continuous targets.

Question 33

For multi-class classification in Keras, a common loss function is:

Amean_squared_error
Bcategorical_crossentropy โœ…
Cmean_absolute_error
Dhuber
๐Ÿ’ก ExplanationUsed with a softmax output layer.

Question 34

Which optimiser is most commonly used as a default in Keras?

ASGD
BAdam โœ…
CAdagrad
DMomentum
๐Ÿ’ก ExplanationAdam adapts the learning rate for each parameter.

Question 35

What is an ‘epoch’?

AOne weight update
BOne complete pass through the entire training dataset โœ…
COne layer
DOne batch
๐Ÿ’ก ExplanationTraining usually runs for many epochs.

Question 36

What is a ‘shallow’ vs ‘deep’ neural network?

AShallow has one hidden layer; deep has many โœ…
BShallow uses images; deep uses text
CShallow is slower
DNo difference
๐Ÿ’ก ExplanationDepth lets networks learn hierarchical features.

Question 37

What is a Convolutional Neural Network (CNN) best suited for?

ATabular data
BImage data โœ…
CTime series only
DText only
๐Ÿ’ก ExplanationConvolution and pooling layers extract spatial features.

Question 38

What does a pooling layer do in a CNN?

AAdds parameters
BReduces spatial dimensions and computation โœ…
CIncreases image size
DNormalises colours
๐Ÿ’ก ExplanationMax pooling keeps the largest value in each window.

Question 39

What is a Recurrent Neural Network (RNN) designed for?

AImages
BSequential data such as text and time series โœ…
CTabular data
DClustering
๐Ÿ’ก ExplanationRNNs keep a hidden state across time steps.

Question 40

What is an LSTM?

AA type of CNN
BA recurrent network with gates that handles long-term dependencies โœ…
CAn optimiser
DA loss function
๐Ÿ’ก ExplanationLong Short-Term Memory networks mitigate vanishing gradients in sequences.

Question 41

What is an autoencoder?

AA supervised classifier
BAn unsupervised network that learns to compress and reconstruct its input โœ…
CAn optimiser
DA pooling layer
๐Ÿ’ก ExplanationUsed for dimensionality reduction and denoising.

Question 42

What does ‘dropout’ do?

ADeletes the dataset
BRandomly disables neurons during training to reduce overfitting โœ…
CSpeeds up inference
DIncreases learning rate
๐Ÿ’ก ExplanationA regularisation technique.

Next: Deep Learning with Keras and TensorFlow โ†’

ย 
๐Ÿ”ถ

Course 3 โ€” Deep Learning with Keras and TensorFlow (Questions 43โ€“54)

TensorFlow 2, transfer learning, callbacks, Functional API, transformers, GANs and diffusion

Question 43

In TensorFlow 2, which mode is enabled by default?

AGraph mode
BEager execution โœ…
CLazy mode
DStatic mode
๐Ÿ’ก ExplanationEager execution evaluates operations immediately.

Question 44

What is a tensor?

AA type of neuron
BA multi-dimensional array โœ…
CA loss function
DA dataset
๐Ÿ’ก ExplanationScalars, vectors and matrices are tensors of rank 0, 1 and 2.

Question 45

What is transfer learning in deep learning?

ATraining from scratch
BReusing a pre-trained model’s weights for a new but related task โœ…
CMoving data between GPUs
DCopying datasets
๐Ÿ’ก ExplanationCommon with ImageNet-trained models such as VGG16 and ResNet.


Question 46

When fine-tuning a pre-trained model, ‘freezing’ layers means:

ADeleting them
BKeeping their weights fixed during training โœ…
CDoubling them
DRandomising them
๐Ÿ’ก ExplanationUsually early layers are frozen and later layers trained.

Question 47

What is data augmentation?

ABuying more data
BCreating modified versions of training data, such as rotated or flipped images โœ…
CRemoving outliers
DCompressing data
๐Ÿ’ก ExplanationIt improves generalisation.

Question 48

What does a Keras ‘callback’ such as EarlyStopping do?

AStarts training early
BStops training when a monitored metric stops improving โœ…
CSkips layers
DDeletes checkpoints
๐Ÿ’ก ExplanationPrevents overfitting and saves time.

Question 49

What is the Keras Functional API used for?

AOnly simple stacks
BModels with multiple inputs/outputs or non-linear topology โœ…
CData loading
DPlotting
๐Ÿ’ก ExplanationMore flexible than Sequential.

Question 50

What is batch normalisation?

ANormalising labels
BNormalising layer inputs per mini-batch to stabilise and speed training โœ…
CShuffling batches
DReducing batch size
๐Ÿ’ก ExplanationIt also has a mild regularising effect.

Question 51

What is a Transformer’s key mechanism?

AConvolution
BSelf-attention โœ…
CPooling
DRecurrence
๐Ÿ’ก ExplanationSelf-attention relates every token to every other token.

Question 52

What is a GAN?

AA regression model
BTwo networks, a generator and a discriminator, trained in competition โœ…
CA clustering algorithm
DAn optimiser
๐Ÿ’ก ExplanationGenerative Adversarial Networks produce realistic synthetic data.

Question 53

What is a diffusion model?

AA classifier
BA generative model that learns to reverse a gradual noising process โœ…
CA pooling method
DA tokeniser
๐Ÿ’ก ExplanationUsed for image generation.

Question 54

What is ‘reinforcement learning’ with Keras/TensorFlow typically used for in the course?

AImage classification
BTraining agents (e.g. Q-learning) to maximise reward โœ…
CClustering
DTranslation
๐Ÿ’ก ExplanationDeep Q-Networks combine Q-learning with neural networks.

Next: Neural Networks and Deep Learning with PyTorch โ†’

ย 
๐Ÿ”ฅ

Courses 4โ€“5 โ€” Neural Networks and Deep Learning with PyTorch (Questions 55โ€“70)

Tensors, autograd, nn.Module, DataLoader, losses, optimisers and CNNs in PyTorch

Question 55

In PyTorch, what is a tensor’s ‘requires_grad’ attribute for?

APrinting the tensor
BTelling autograd to track operations for gradient computation โœ…
CChanging its shape
DMoving it to GPU
๐Ÿ’ก ExplanationSet requires_grad=True for parameters you want to optimise.

Question 56

Which method computes gradients in PyTorch?

A.forward()
B.backward() โœ…
C.step()
D.zero_grad()
๐Ÿ’ก Explanationloss.backward() populates .grad on tensors.

Question 57

What does optimizer.step() do?

AComputes the loss
BUpdates parameters using the computed gradients โœ…
CResets gradients
DLoads data
๐Ÿ’ก ExplanationCalled after loss.backward().

Question 58

Why call optimizer.zero_grad() each iteration?

ATo delete the model
BBecause gradients accumulate by default โœ…
CTo increase speed
DTo shuffle data
๐Ÿ’ก ExplanationWithout it, gradients from previous batches add up.

Question 59

Which PyTorch class do you subclass to define a custom model?

Atorch.Tensor
Btorch.nn.Module โœ…
Ctorch.optim.SGD
Dtorch.utils.Data
๐Ÿ’ก ExplanationYou define layers in __init__ and the computation in forward().

Question 60

What do Dataset and DataLoader do in PyTorch?

ADefine the model
BDataset wraps the data; DataLoader batches and shuffles it โœ…
CCompute gradients
DPlot results
๐Ÿ’ก ExplanationDataLoader(dataset, batch_size=…, shuffle=True).


Question 61

Which loss function is used for multi-class classification in PyTorch?

Ann.MSELoss
Bnn.CrossEntropyLoss โœ…
Cnn.L1Loss
Dnn.BCELoss
๐Ÿ’ก ExplanationIt combines LogSoftmax and NLLLoss.

Question 62

Which loss is used for linear regression in PyTorch?

Ann.CrossEntropyLoss
Bnn.MSELoss โœ…
Cnn.NLLLoss
Dnn.BCELoss
๐Ÿ’ก ExplanationMean squared error.

Question 63

What is mini-batch gradient descent?

AUsing the whole dataset per update
BUpdating weights using small subsets of the data โœ…
CUsing one sample per update
DNo updates
๐Ÿ’ก ExplanationA balance between batch and stochastic gradient descent.

Question 64

What does torch.nn.Linear(in_features, out_features) create?

AA convolution layer
BA fully connected (linear) layer โœ…
CAn activation
DA loss
๐Ÿ’ก ExplanationIt applies y = xA^T + b.

Question 65

What is the purpose of a validation set?

AFinal testing
BTuning hyperparameters and monitoring overfitting during training โœ…
CTraining
DDeployment
๐Ÿ’ก ExplanationThe test set is kept for the final unbiased estimate.

Question 66

How do you move a PyTorch model to a GPU?

Amodel.gpu()
Bmodel.to(‘cuda’) โœ…
Cmodel.fast()
Dmodel.device()
๐Ÿ’ก ExplanationTensors must be moved to the same device.

Question 67

What is ‘momentum’ in optimisation?

AA type of loss
BA term that accelerates gradient descent by accumulating past gradients โœ…
CA regulariser
DA data loader
๐Ÿ’ก ExplanationIt helps escape shallow minima and smooths updates.

Question 68

What is the softmax regression model used for?

ARegression
BMulti-class classification โœ…
CClustering
DGeneration
๐Ÿ’ก ExplanationIt generalises logistic regression to many classes.

Question 69

In a CNN in PyTorch, nn.Conv2d’s ‘kernel_size’ sets:

AThe number of classes
BThe size of the filter window โœ…
CThe learning rate
DThe batch size
๐Ÿ’ก ExplanationStride and padding control the output size.

Question 70

What does model.eval() do?

ATrains the model
BSets layers like dropout and batch norm to evaluation mode โœ…
CDeletes gradients
DSaves the model
๐Ÿ’ก ExplanationUse torch.no_grad() as well during inference.

Next: Generative AI, LLMs and Fine-Tuning Transformers โ†’

ย 
๐Ÿ’ฌ

Courses 7โ€“11 โ€” Generative AI, LLMs and Fine-Tuning Transformers (Questions 71โ€“88)

Tokenization, embeddings, GPT vs BERT, attention, PEFT, LoRA, QLoRA and quantization

Question 71

What is a large language model (LLM)?

AA database
BA transformer-based model trained on huge text corpora to understand and generate language โœ…
CA search index
DA tokenizer
๐Ÿ’ก ExplanationExamples: GPT, Llama, IBM Granite.

Question 72

What is tokenization?

AEncrypting text
BSplitting text into tokens (words or sub-words) that a model can process โœ…
CTranslating text
DSummarising text
๐Ÿ’ก ExplanationMethods include word-based, character-based and sub-word (BPE, WordPiece).

Question 73

What is an embedding?

AA compressed file
BA dense vector that represents the meaning of a token or text โœ…
CA type of loss
DA layer of dropout
๐Ÿ’ก ExplanationSimilar meanings have nearby vectors.

Question 74

Which architecture uses only a decoder and is used for text generation?

ABERT
BGPT โœ…
CWord2Vec
DCNN
๐Ÿ’ก ExplanationGPT models are autoregressive decoders.

Question 75

Which architecture uses only an encoder and suits understanding tasks?

AGPT
BBERT โœ…
CGAN
DRNN
๐Ÿ’ก ExplanationBERT is trained with masked language modelling.


Question 76

What is positional encoding in a transformer?

AA loss function
BInformation added to embeddings so the model knows token order โœ…
CA tokeniser
DAn optimiser
๐Ÿ’ก ExplanationSelf-attention alone is order-agnostic.

Question 77

What is self-attention?

APaying attention to the user
BA mechanism that weighs the relevance of every token to every other token in a sequence โœ…
CA pooling method
DA regulariser
๐Ÿ’ก ExplanationIt captures long-range dependencies.

Question 78

What is fine-tuning an LLM?

ATraining from scratch
BFurther training a pre-trained model on task-specific data โœ…
CShrinking a model
DTokenising data
๐Ÿ’ก ExplanationAdapts general knowledge to a domain.

Question 79

What is PEFT?

AA dataset
BParameter-Efficient Fine-Tuning: updating a small number of parameters instead of the whole model โœ…
CA tokenizer
DA metric
๐Ÿ’ก ExplanationIt cuts compute and storage costs.

Question 80

What is LoRA?

AA loss function
BLow-Rank Adaptation: adding small trainable low-rank matrices to frozen weights โœ…
CA dataset format
DA prompt style
๐Ÿ’ก ExplanationA popular PEFT method.

Question 81

What is QLoRA?

ALoRA with quantised (e.g. 4-bit) base model weights to reduce memory โœ…
BA query language
CA tokenizer
DA new architecture
๐Ÿ’ก ExplanationIt enables fine-tuning large models on modest hardware.

Question 82

What is quantization?

AAdding layers
BReducing the numerical precision of model weights to shrink size and speed inference โœ…
CTokenising
DClustering
๐Ÿ’ก ExplanationFor example from 32-bit floats to 8-bit or 4-bit.

Question 83

What is instruction tuning?

ATuning hyperparameters
BFine-tuning a model on instruction-response pairs so it follows prompts better โœ…
CPruning
DQuantising
๐Ÿ’ก ExplanationIt improves zero-shot task following.

Question 84

What is RLHF?

AA tokenizer
BReinforcement Learning from Human Feedback โœ…
CA dataset
DA pooling layer
๐Ÿ’ก ExplanationHuman preference signals train a reward model that guides the LLM.

Question 85

Which library is widely used to load pre-trained transformers?

Ascikit-learn
BHugging Face Transformers โœ…
COpenCV
DMatplotlib
๐Ÿ’ก ExplanationIt provides models, tokenizers and pipelines.

Question 86

What is in-context learning?

ARetraining the model
BThe model adapting to a task from examples given in the prompt without weight updates โœ…
CFine-tuning
DQuantising
๐Ÿ’ก ExplanationFew-shot prompting is in-context learning.

Question 87

What is perplexity?

AA tokeniser
BA measure of how well a language model predicts a sample; lower is better โœ…
CA layer
DAn optimiser
๐Ÿ’ก ExplanationIt is the exponential of the average negative log-likelihood.

Question 88

What does ‘temperature’ control during generation?

AHardware heat
BThe randomness of the sampled output โœ…
CModel size
DBatch size
๐Ÿ’ก ExplanationLower values make output more deterministic.

Next: RAG, LangChain and AI Agents โ†’

ย 
๐Ÿ”—

Courses 12โ€“13 โ€” RAG, LangChain and AI Agents (Questions 89โ€“102)

Retrieval-augmented generation, vector databases, chains, agents and the capstone app

Question 89

What does RAG stand for?

ARandom Answer Generator
BRetrieval-Augmented Generation โœ…
CRecurrent Attention Graph
DRapid AI Gateway
๐Ÿ’ก ExplanationRAG retrieves relevant documents and passes them to the LLM as context.

Question 90

Why use RAG?

ATo make models smaller
BTo ground answers in up-to-date or private data and reduce hallucination โœ…
CTo speed up training
DTo replace tokenisation
๐Ÿ’ก ExplanationThe model does not need retraining when the knowledge changes.


Question 91

What is a vector database used for in RAG?

AStoring images
BStoring embeddings and retrieving the most similar chunks to a query โœ…
CTraining models
DTokenising text
๐Ÿ’ก ExplanationExamples: Chroma, FAISS, Milvus.

Question 92

What is ‘chunking’ in a RAG pipeline?

ADeleting documents
BSplitting documents into smaller pieces before embedding โœ…
CMerging answers
DCompressing models
๐Ÿ’ก ExplanationChunk size and overlap affect retrieval quality.

Question 93

What is LangChain?

AA blockchain
BA framework for building LLM applications by chaining prompts, models, retrievers and tools โœ…
CA tokenizer
DA GPU
๐Ÿ’ก ExplanationIt provides chains, agents, memory and document loaders.

Question 94

In LangChain, what is a ‘chain’?

AA dataset
BA sequence of calls (prompt, model, parser) linked together โœ…
CA GPU cluster
DA loss function
๐Ÿ’ก ExplanationLCEL uses the pipe operator to compose them.

Question 95

What is an AI agent?

AA human operator
BAn LLM-driven system that decides which tools or actions to use to reach a goal โœ…
CA dataset
DA tokenizer
๐Ÿ’ก ExplanationAgents plan, call tools and observe results in a loop.

Question 96

What is a ‘prompt template’?

AA fixed answer
BA reusable prompt with placeholders filled at run time โœ…
CA model checkpoint
DA tokenizer file
๐Ÿ’ก ExplanationIt keeps prompts consistent across calls.

Question 97

What does a ‘retriever’ do in RAG?

AGenerates text
BFinds the most relevant documents for a query โœ…
CTrains the model
DEvaluates accuracy
๐Ÿ’ก ExplanationUsually via similarity search over embeddings.

Question 98

Which similarity measure is common for comparing embeddings?

AHamming distance
BCosine similarity โœ…
CJaccard index
DEdit distance
๐Ÿ’ก ExplanationIt compares the angle between vectors.

Question 99

What is ‘memory’ in a LangChain application?

AGPU RAM
BStored conversation history that gives the model context across turns โœ…
CDisk space
DThe training set
๐Ÿ’ก ExplanationBuffer and summary memory are common types.

Question 100

What is Gradio used for in the capstone projects?

ATraining models
BBuilding a simple web interface for a model or app โœ…
CTokenising
DQuantising
๐Ÿ’ก ExplanationIt lets you demo an app in a few lines of Python.

Question 101

What is a hallucination in a RAG system usually a sign of?

AToo many GPUs
BPoor retrieval or the model ignoring the provided context โœ…
CLow temperature
DToo many chunks
๐Ÿ’ก ExplanationBetter chunking, retrieval and prompts reduce it.

Question 102

What is the role of an ‘output parser’ in LangChain?

ATokenising input
BConverting the model’s raw text into a structured format such as JSON โœ…
CEmbedding documents
DTraining
๐Ÿ’ก ExplanationIt makes outputs usable by downstream code.

๐Ÿ“š The 13 courses in the IBM AI Engineering Professional Certificate

1. Machine Learning with Python
2. Introduction to Deep Learning & Neural Networks with Keras
3. Deep Learning with Keras and TensorFlow
4. Introduction to Neural Networks and PyTorch
5. Deep Learning with PyTorch
6. AI Capstone Project with Deep Learning
7. Generative AI and LLMs: Architecture and Data Preparation
8. Gen AI Foundational Models for NLP & Language Understanding
9. Generative AI Language Modeling with Transformers
10. Generative AI Engineering and Fine-Tuning Transformers
11. Generative AI Advanced Fine-Tuning for LLMs
12. Fundamentals of AI Agents Using RAG and LangChain
13. Project: Generative AI Applications with RAG and LangChain

Course line-up as listed on Coursera; IBM updates it periodically, so check the certificate page for the current order.

๐Ÿ“ฅ Download all 102+ AI Engineering quiz MCQs as PDF (free)

๐ŸŽฏ More IBM and AI certification MCQs

Question banks updated regularly to match the latest course content

Frequently Asked Question

Can I download these quiz answers as a PDF?

Yes, free, from the button at the end of the page

How many courses are in the IBM AI Engineering Professional Certificate?

Thirteen courses on Coursera, from Machine Learning with Python through Keras, TensorFlow and PyTorch to generative AI, fine-tuning, RAG and LangChain, ending with applied projects.

Is the IBM AI Engineering Professional Certificate free?

You can audit most course videos free on Coursera, but graded quizzes and the certificate need a Coursera subscription or financial aid.

Is AI Engineering harder than IBM AI Fundamentals?

A: Yes. AI Fundamentals is a beginner, no-code course. AI Engineering assumes Python and covers building and training models in scikit-learn, Keras, PyTorch and Hugging Face.

Which topics come up most in the graded quizzes?

Regression and classification metrics, backpropagation and activation functions, Keras and PyTorch training loops, transformer architecture, LoRA and PEFT, and RAG components.

About the author

MCQS TOP

2 Comments

Leave a Comment