Certified Deep Learning Specialist (CDLS) practice questions
Deep Learning · CDLS · 300 questions
Original, similar-concept practice questions for the Certified Deep Learning Specialist (CDLS) exam, covering neural network foundations, the math behind deep learning, core architectures (CNN, RNN/LSTM, Transformer, autoencoder), training and optimization, deep learning frameworks and workflow, and generative and reinforcement learning.
This course contains the use of artificial intelligence.
Practice Quizzes
Test your knowledge with standard 20-question practice sets.
Quiz 1
Quiz 2
Quiz 3
Quiz 4
Quiz 5
Quiz 6
Quiz 7
Quiz 8
Quiz 9
Quiz 10
Quiz 11
Quiz 12
Quiz 13
Quiz 14
Quiz 15
Browse by Domain
Study specific topics at your own pace.
Neural Network Foundations · 50 questions
- What is the primary function of a convolutional layer in a CNN?
- In a very deep neural network, what problem does a residual (skip) connection primarily address?
- A CNN has already extracted features through several convolutional and pooling layers. What does the fully connected layer near the end of the network typically do?
- Which statement best describes backpropagation in a neural network?
- What is the main objective behind using a deep neural network rather than a shallow one?
- A neuron computes a weighted sum of its inputs plus a bias term before applying an activation function. What role does the bias play?
- How does a recurrent neural network (RNN) fundamentally differ from a standard feedforward network?
- A machine learning engineer notices that after several convolutional layers, the feature maps have shrunk considerably. Which technique is designed to prevent this shrinkage by preserving the spatial size of the input?
- What is a perceptron in the context of neural networks?
- What core mechanism gives the Transformer architecture its ability to model relationships across an entire input sequence?
- Why is a convolutional neural network generally preferred over a fully connected network for image classification tasks?
- A team training a generative adversarial network (GAN) notices that the generator keeps producing only a handful of nearly identical outputs, no matter how long training continues. What is this well-known GAN training problem called?
- What specific architectural property allows a CNN to efficiently capture local spatial features in an image?
- How does a dense (fully connected) layer differ structurally from a convolutional layer?
- A team is choosing between a CNN and an RNN for two different projects: one involves classifying photographs, and the other involves predicting the next word in a sentence. Which statement correctly matches architecture to task?
- What does the learning rate hyperparameter primarily control during neural network training?
- What is a key distinction between deep learning and traditional machine learning approaches?
- In the context of deep learning frameworks, what is a tensor?
- What is the primary role of a generative model, such as a GAN?
- What is the main objective of using the Adam optimizer during neural network training?
- A model achieves very high accuracy on training data but performs noticeably worse on new, unseen data. Which of the following is a common technique to address this problem?
- What is the purpose of a cost (loss) function in training a neural network?
- In a convolutional layer, what does the stride parameter control?
- What does the process of hyperparameter tuning involve?
- What is a distinguishing property of the tanh (hyperbolic tangent) activation function compared to sigmoid?
- In a generative adversarial network, what is the primary job of the generator network?
- In a convolutional layer, what best describes the kernel (also called a filter)?
- A colleague asks you to explain the difference between a shallow neural network and a deep neural network. What is the key distinction?
- Which of the following is a classic example of a supervised learning task?
- A network's final layer produces one raw score per class for a 10-class image classification problem. What is the purpose of applying softmax to these scores?
- What is a learning-rate schedule in the context of training a neural network?
- A machine learning engineer chooses ReLU as the activation function for the hidden layers of a new network. Which property of ReLU is she relying on?
- In the context of an artificial neural network, what is a neuron?
- A team notices their deep network performs very well on training data but poorly on validation data. They add a dropout layer to help. What does a dropout layer actually do?
- What is the main purpose of an autoencoder?
- Why did the Transformer architecture become so influential for natural language processing tasks?
- A team is building a system that takes an English sentence as input and produces its French translation as output. Which type of model architecture is best suited to this task?
- During training of a multi-class classifier, why does the output layer typically apply the softmax function rather than leaving the raw scores untouched?
- A team splits their labeled data into three groups before training a neural network. What is the main reason they carve out a validation set separate from the training and test sets?
- In the context of training a neural network, what does the term 'epoch' refer to?
- A model's final layer needs to output something that can be interpreted as the probability of belonging to the positive class in a two-class problem. Which activation function is the classic choice for that output neuron, and why?
- A deep network is training unstably, with loss swinging wildly between mini-batches, and convergence is painfully slow. An engineer suggests inserting batch normalization layers between the linear layers and activations. What problem is batch normalization primarily designed to address?
- A small startup has only 2,000 labeled medical images, nowhere near enough to train a deep CNN from scratch. Instead, they start from a network already trained on millions of general-purpose photos, then fine-tune its later layers on their small medical dataset. What is this strategy called?
- A team is choosing an architecture to classify tens of thousands of product photos into categories like 'shoes,' 'bags,' and 'watches.' Which type of network is generally the best default choice for this kind of image classification task?
- In an NLP model, raw text is first tokenized into integer IDs and then passed through an embedding layer before reaching the rest of the network. What role does that embedding layer serve?
- What key architectural characteristic sets a Recurrent Neural Network (RNN) apart from a standard feedforward network when processing sequential data like sentences or time series?
- Before training even begins, an engineer must decide on values like the learning rate, the number of layers, and the batch size. What is the general term for these kinds of values?
- Standard RNNs struggle to learn dependencies across long sequences because gradients shrink to near zero as they're propagated back through many time steps. Which component of an LSTM (Long Short-Term Memory) unit is primarily responsible for mitigating this vanishing gradient problem?
- Inside a convolutional layer, a learned filter slides across the input image and produces a 2D grid of output values. What is this resulting grid of values called?
- During the training loop, after the network makes a prediction for a batch of examples, something has to quantify how wrong that prediction was so the optimizer knows which direction to adjust the weights. What is the name of that component, and what role does it play?
Math & Classical ML Foundations · 50 questions
- A data scientist plots the distribution of a feature and confirms it is a perfect normal (Gaussian) distribution. Based on this alone, what can she conclude about the relationship between the mean, median, and mode of the feature?
- In the context of the mathematics used by deep learning frameworks, what does taking the tensor product of two tensors do?
- During backpropagation, a training framework computes the gradient of the loss function with respect to each weight in the network. What is the fundamental purpose of computing this gradient?
- A team trains a deep neural network that achieves 99% accuracy on the training set, but only 61% accuracy on new, unseen data from the same problem domain. Which phenomenon best explains this gap?
- An engineer is choosing between L1 and L2 regularization for a model and wants some of the less useful weights to be driven exactly to zero, effectively performing feature selection. Which statement correctly distinguishes the two approaches for this goal?
- A layer in a neural network multiplies a 3x2 matrix by a 2x4 matrix. What are the dimensions of the resulting matrix?
- In linear algebra as applied to machine learning (for example, in principal component analysis), what is the significance of a matrix's eigenvectors and eigenvalues?
- A dataset of household incomes shows most values clustered at the lower end, with a small number of very high-income outliers stretching the distribution's tail far out to the right. How would a data scientist describe the shape of this distribution?
- A square weight matrix needs to be inverted as part of a closed-form solution. Under which condition does that inverse actually exist?
- A team trains a deep neural network and finds it scores 99% accuracy on the training set but only 68% on the held-out test set. What best explains this gap?
- How does logistic regression fundamentally differ from linear regression in what it outputs?
- Which mathematical concept allows backpropagation to compute how the loss at the output layer should adjust the weights buried deep in earlier layers of a network?
- A data scientist is choosing between several candidate decision boundaries for a support vector machine classifier. What is the SVM's primary objective when selecting among them?
- What does the statistical measure of variance actually quantify about a dataset?
- In linear algebra, what does taking the transpose of a matrix produce?
- A regression model ends up with several coefficients that are extremely large in magnitude, making the model unstable and prone to overfitting. Which technique directly addresses this?
- In the underlying math of a neural network, what role do matrices primarily play?
- An engineer is reviewing why a similarity-scoring layer multiplies two embedding vectors element-wise and sums the results. Conceptually, what does this dot product operation give you?
- Suppose h(x) = f(g(x)), where g is an inner function and f is an outer function applied to g's output. Which expression correctly gives h'(x)?
- A researcher needs to know whether a 2x2 weight transformation matrix can be inverted before using it in a normalizing-flow layer. Which property of the matrix should the researcher compute to answer this?
- A colleague shows you a training script where the loss function is defined as Mean Squared Error (MSE) between the model's predictions and the target values. Based on this loss choice alone, what kind of task is the model most likely being trained for?
- While running 5-fold cross-validation, a data scientist notices that the model's accuracy swings wildly between folds — scoring 96% on one fold and only 71% on another, despite the folds being drawn from the same overall dataset. What does this high variance across folds most likely indicate?
- A neural network's final layer applies a softmax activation to produce output for a 3-class image classification problem. Which of the following best describes what the resulting output vector looks like?
- In a fully connected neural network layer, each neuron computes a weighted sum of its inputs before applying an activation function. What role does the bias term play in this computation?
- During training, an engineer observes that the derivative of the loss function with respect to a particular weight has reached exactly zero at the current point. Which statement correctly interprets what this means on its own?
- A data scientist has a dataset with five numeric features and wants a single mathematical object that summarizes how much each feature varies and how each pair of features moves together. Which object should they compute?
- A team needs to predict the next word in a sentence, where each prediction depends on the words that came before it in order. Which model class is classically built for this kind of sequential, order-dependent prediction task?
- While profiling a numeric feature before training, an analyst finds that its mean is noticeably higher than its median. What does this gap between mean and median most directly suggest about the feature's distribution?
- The idea of finding an optimal separating hyperplane between two classes, with the widest possible margin on either side, is the central concept behind which classical algorithm?
- In a convolutional neural network, the repeated stacking of convolution layers followed by pooling layers is, at its core, a practical implementation of which broader concept?
- A retailer has purchase history for thousands of customers but no predefined labels for them, and wants to automatically group similar customers together to design targeted marketing campaigns. Which category of machine learning task is this?
- Which classical model makes a prediction for a new data point by looking at the k training examples that are closest to it in feature space and basing its output on them (majority vote for classification, or averaging for regression)?
- A subscription company wants to predict whether each customer will churn (yes/no) in the next month using account features like tenure, usage, and support tickets. Which classical algorithm is a common, natural first choice for this kind of binary outcome prediction?
- After computing the covariance between two features, X and Y, an analyst finds it is a large positive number. What does this positive covariance indicate about the relationship between X and Y?
- During training, a framework computes the gradient vector of the loss function with respect to the model's weights. In which direction does this raw gradient vector actually point?
- Given two 2D vectors v = [v1, v2] and w = [w1, w2], which expression correctly computes their dot product v · w?
- A dataset has 200 numeric features, many of which are correlated, and a team wants to compress it into a much smaller set of features while retaining as much of the original variance as possible before feeding it into a downstream model. Which technique is the standard choice for this?
- A researcher plots the training loss curve for a new model and sees it spike up and down wildly from epoch to epoch instead of trending downward, even though the architecture and data pipeline are known to be correct. What is the most likely explanation?
- During backpropagation, why does the algorithm need to compute the derivative of each neuron's activation function at its current input value?
- In one sentence, what is gradient descent fundamentally trying to accomplish when training a model?
- In a logistic regression model, what is the role of the log-likelihood (log-loss) function used during training?
- Which core calculus concept provides the mathematical foundation that makes it possible to train a multi-layer neural network by propagating error signals backward through the network?
- A team wants an activation function that squashes any real-valued input into an output strictly between -1 and 1, with an output of 0 when the input is 0. Which activation function fits this description?
- A team trains a CNN on a dataset of images that are each labeled either "cat" or "dog," and the network learns to predict that label for new images. What type of machine learning task is this?
- You are building a network with a softmax output layer that must choose among ten mutually exclusive digit classes (0 through 9). Which loss function is the most appropriate choice for training this model?
- An engineer builds a network with a single sigmoid output neuron trained to answer one question for each input image: is this digit a "9" or not? What kind of task is this?
- In precise mathematical terms, how is the dot product of two vectors A and B, each with n components, formally defined?
- A data scientist is exploring a feature such as annual household income, where most values cluster at the lower end but a small number of values extend far into the millions. Under what condition is it generally most appropriate to apply a log scale (or log transform) to a feature like this?
- The Naive Bayes classifier, popular for tasks like spam filtering, derives its predictions from which foundational mathematical principle?
- In probability notation, what does the expression P(A|B) represent?
Core Architectures (CNN, RNN, Transformer) · 50 questions
- A team training a deep feedforward network notices that training loss oscillates wildly and convergence is painfully slow, even after tuning the learning rate. A colleague suggests inserting a normalization layer after each hidden layer's activations. Why would this help?
- In a generative adversarial network, two networks are trained together in competition. What is the specific job of the generator network in this setup?
- A product team wants to build a system that generates long, coherent, human-sounding paragraphs of text from a prompt. Which type of model is best suited to this task?
- During training of a recurrent neural network on long sequences, the loss suddenly spikes to NaN and the weight values become enormous. What phenomenon is most likely responsible?
- In a typical CNN image classifier, the convolutional and pooling layers output a stack of 2D feature maps, but the final classification layers are fully connected and expect a 1D input vector. What layer bridges this gap?
- You need an activation function for the output neuron of a binary classifier so the raw score can be interpreted directly as a probability between 0 and 1. Which activation function fits this requirement?
- What is the core purpose of the self-attention mechanism inside a Transformer block?
- An artificial neuron computes a weighted sum of its inputs, adds a bias term, and passes the result through a nonlinear activation function. What is this process fundamentally intended to model?
- When comparing the internal gating structures of a GRU and an LSTM cell, which statement correctly describes the difference?
- A researcher trains a plain (vanilla) RNN on sequences several hundred time steps long and finds it struggles to learn dependencies between early and late elements in the sequence. What is the primary architectural reason for this limitation?
- A medical imaging team needs a model that labels every individual pixel of an MRI scan as belonging to a specific tissue type, producing a full-resolution segmentation mask rather than a single image-level label. Which architecture is purpose-built for this pixel-wise task?
- In a typical image-classification CNN, the convolutional layers extract low- and mid-level visual features like edges and textures. Which part of the network is responsible for combining those extracted features into the final, high-level representation used to make the classification decision?
- In a basic autoencoder architecture, how do the encoder and decoder halves of the network relate to one another?
- In a convolutional layer, what does the stride parameter actually control?
- A data scientist is building a network to predict house prices, a continuous dollar value with no fixed range. Which activation function should sit on the output layer's single neuron?
- A fraud team has millions of unlabeled, high-dimensional transaction embeddings and wants to flag unusual records without any labeled examples of fraud. Which unsupervised architecture is the classic fit?
- What is the core difference between max pooling and average pooling in a CNN's downsampling layer?
- How does connectivity work between neurons in a fully connected (dense) layer and the layer before it?
- You inherit a trained model and notice its output layer uses a plain linear activation with no squashing at all. What kind of task was this model most likely built for?
- What advantage does a GRU offer compared to a standard (vanilla) RNN when working with longer sequences?
- Mechanically, how is the output of an attention mechanism produced?
- A model performs great on training data but noticeably worse on validation data. A colleague suggests adding a dropout layer. What is dropout actually doing to address this?
- Within an autoencoder, what job does the decoder half of the network perform?
- Which gate inside an LSTM cell is responsible for deciding what information should be discarded from the cell state?
- Since a Transformer processes all tokens in a sequence simultaneously rather than one at a time, why is positional encoding added to the input embeddings?
- A team is building a system that automatically transcribes customer support calls into written text. Which type of model architecture is best suited to this task?
- In a convolutional neural network, what is the main purpose of adding padding around the input feature map before applying a convolution?
- A researcher is choosing between a sigmoid and a tanh activation function for a hidden layer. In which situation would tanh typically be preferred over sigmoid?
- In the context of a convolutional neural network, what does the term 'receptive field' refer to?
- What is the main purpose of a kernel (filter) in a convolutional layer?
- In a feedforward neuron, why is a bias term added alongside the weighted sum of inputs?
- What is the 'bottleneck layer' in an autoencoder?
- In a generative adversarial network (GAN), what is the role of the discriminator network?
- What is a key advantage of a Transformer architecture over a recurrent neural network (RNN) for natural language processing tasks?
- Why do most image-classification CNN architectures end with one or more fully connected layers after the convolutional and pooling layers?
- A team training a GAN to generate images of handwritten digits notices the generator keeps producing only a handful of nearly identical-looking digit images, regardless of the random noise input. What phenomenon is this?
- A company wants to build an unsupervised model that learns to remove background noise from audio recordings, without relying on labeled clean/noisy pairs for every possible noise type. Which architecture is well suited to this?
- What is the main purpose of a variational autoencoder (VAE), as opposed to a standard (vanilla) autoencoder?
- A team has a small dataset of only 800 labeled medical images and wants strong classification performance. They decide to start from a CNN such as VGG or ResNet that was pre-trained on a large general-purpose image dataset. What is the main goal of this transfer learning approach?
- What structurally distinguishes a recurrent neural network from a feedforward neural network?
- What is the primary objective of an encoder-decoder architecture used for sequence-to-sequence (seq2seq) tasks, such as machine translation?
- In a variational autoencoder (VAE), what does the encoder learn to output for a given input, as opposed to a single fixed latent vector?
- What is a key feature of a classic RNN-based encoder-decoder architecture used for machine translation?
- For translating long documents, why is a Transformer generally preferred over an LSTM-based encoder-decoder?
- What is the primary purpose of a residual (skip) connection in a ResNet architecture?
- What is the 'latent space' in an autoencoder?
- In a neural network, what are 'weights'?
- A team wants to build a model that predicts the most likely next word given the words that came before it in a sentence. Which type of architecture is best suited to this task?
- Which activation function is commonly used in deep networks specifically to help mitigate the vanishing gradient problem, compared to sigmoid or tanh?
- What is a key advantage of shared weights (weight sharing) within a convolutional layer of a CNN?
Training, Optimization & Regularization · 50 questions
- What does the mean absolute error (MAE) metric measure for a regression model?
- A team trains a model on a dataset with tens of millions of examples using standard batch gradient descent. What is the main drawback they will run into?
- A model's training loss barely moves epoch after epoch and stays stubbornly high. Weight updates are being computed correctly and gradients are non-zero. What is the most likely cause?
- What is the main advantage of mini-batch gradient descent over full batch gradient descent when training on a large dataset?
- What is the main downside of setting the learning rate far too low when training a neural network?
- What is learning-rate decay in the context of training a deep learning model?
- What does the mean squared error (MSE) loss function do?
- What is the main purpose of a loss function in training a deep learning model?
- What is a key defining feature of the RMSprop optimizer?
- A model's loss oscillates wildly and never settles down, even growing over time. Gradients are correct. What hyperparameter is most likely set too high?
- What does L2 regularization add to a model's loss function?
- How does stochastic gradient descent (SGD), updating on one example at a time, typically compare to full batch gradient descent?
- A bank is building a fraud detection model. Missing an actual fraudulent transaction (a false negative) is far more costly to the bank than flagging a legitimate transaction for review (a false positive). Which metric should the team prioritize when evaluating the model?
- Which optimizer gives each parameter its own learning rate but has a well-known drawback of that per-parameter rate shrinking too aggressively over long training runs?
- What is cross-validation?
- In classification evaluation, what is a false negative (type II error)?
- A classifier is trained on a dataset where 98% of examples belong to the negative class. Which pair of metrics gives a more informative picture of performance than accuracy alone?
- During training, the training loss keeps falling steadily, but the validation loss starts rising after a certain epoch. What does this pattern indicate?
- What is the benefit of using a mini-batch size greater than one, compared to updating weights after every single training example?
- What is the main goal of hyperparameter tuning in a deep learning workflow?
- What is a practical benefit of using the RMSprop optimizer over plain SGD with a single fixed learning rate?
- What is the main strength of the F1 score as an evaluation metric for a classifier trained on imbalanced classes?
- A classifier achieves a recall of 1.0 on the positive class. What does this tell you?
- When is it most appropriate to apply early stopping during training?
- What is an advantage of batch gradient descent over stochastic gradient descent (SGD)?
- A newcomer is picking an optimizer for a fresh image classification project and wants a solid default that adapts its step size per parameter using running averages of both the gradients and their squares. Which optimizer are they describing?
- In gradient-based training, what is the primary role of the learning rate hyperparameter?
- A team wants a regularization method that can push some of their model's weights all the way to exactly zero, effectively removing those features from the model. Which comparison correctly describes the difference between L1 and L2 regularization for this goal?
- A team builds a classifier for a rare disease where only 1% of patients in the dataset actually have the condition. The model predicts 'no disease' for every single patient and reports a very high score on one metric. Which metric is dangerously misleading in this scenario if used as the only measure of performance?
- In classification evaluation, what does the accuracy metric represent?
- A bank builds a model to flag transactions as fraudulent, and every falsely flagged legitimate transaction locks out a real customer and triggers an expensive manual review. Which metric should the team prioritize to keep those false alarms as low as possible?
- When inspecting a multi-class confusion matrix, what does it indicate if the values along the main diagonal are large relative to the off-diagonal values?
- A team is training a neural network to forecast next week's electricity demand as a continuous numeric value from historical usage data. Which loss function is best suited to this time-series forecasting task?
- In a binary classification confusion matrix, what does a true negative represent?
- A hospital deploys a model to screen scans for a serious tumor, and missing an actual tumor case (letting it go undetected) is far more dangerous than sending a healthy patient for an unnecessary follow-up. Which metric should this team prioritize?
- Which regularization technique works by randomly setting a fraction of neuron outputs to zero during each training pass, forcing the network to avoid relying too heavily on any single neuron?
- What is the effect of applying L1 (lasso) regularization to a model's weights during training?
- What is the main purpose of building a confusion matrix after evaluating a classifier on a test set?
- In a regression model's evaluation report, what does the R-squared value actually tell you?
- A fraud-detection model shows high precision but low recall on the test set. What does this combination actually indicate about its behavior?
- How does the Adam optimizer relate to RMSprop in terms of what it borrows from each technique?
- What is the primary objective the Adam optimizer is trying to achieve during training?
- A spam filter classifies a legitimate email as spam. In classification terminology, what is this specific kind of mistake called?
- An object detection model trains on images where the background class vastly outnumbers rare foreground objects. Which loss function was specifically designed to address this kind of imbalance?
- Why would a data scientist report the F1 score instead of just reporting precision and recall separately?
- During model development, why do teams hold out a separate validation set instead of just training and testing on the same data split?
- A researcher training a deep network for 200 epochs notices validation loss stopped improving around epoch 60 and slowly worsened afterward. What technique would have helped avoid wasting the remaining epochs and overfitting the model?
- In the context of neural network training, what precisely is one epoch?
- A team is training a very deep neural network with dozens of layers and wants an optimizer that handles this scale well. Why is Adam typically recommended for this kind of architecture?
- A team wants a regression metric that punishes large prediction errors much more severely than small ones. Which metric fits that requirement?
Frameworks & Practical Workflow · 50 questions
- In PyTorch, what is the primary role of the torch.optim module?
- A team needs to run a trained image classifier on a low-power embedded sensor that has no GPU. What is the most effective way to speed up inference in this environment?
- In a PyTorch training script, what is the main job of the DataLoader class?
- A team is building a sentiment classifier from raw customer product reviews. Before feeding the text into any embedding or vectorization step, what preprocessing is most commonly applied first?
- While fine-tuning a pretrained Keras model, an engineer wants to keep the weights of the early convolutional layers fixed so only the new top layers learn. Which property accomplishes this on a per-layer basis?
- What is the main objective when a data scientist applies standardization to a numeric feature before training?
- A new team member asks what the term 'pipeline' refers to in the context of a deep learning project. What is the best description?
- In TensorFlow, what is a tf.Variable primarily used for?
- During data preprocessing with pandas, when is it most appropriate to use a Python lambda function with .apply()?
- A dataset includes a categorical feature with exactly 5 possible, unordered values (e.g., shipping method). Which encoding is generally best before feeding this feature into a neural network?
- When building and training a Sequential model in Keras, which order of method calls reflects the correct standard workflow?
- A Keras training run sometimes overfits in later epochs. Which built-in callback should be used to automatically save only the model weights from the epoch with the best validation loss?
- What is the fundamental purpose of hyperparameter tuning in a deep learning workflow?
- Why does the file size and parameter count of a trained model matter so much specifically when deploying to a mobile phone?
- A dataset has a feature that ranges from 1 to 1000, with no significant outliers, and the team wants all input features bounded within a fixed, known range for a neural network. Which scaling technique fits best?
- What does one-hot encoding fundamentally accomplish when applied to a categorical feature?
- When implementing a custom loss function as a class in PyTorch, which base class should it inherit from to integrate properly with the training loop?
- An engineer wants Keras training to automatically halt if the validation loss has not improved for a specified number of consecutive epochs. Which callback provides this behavior?
- Which specific technique transforms a numeric feature so that it ends up with a mean of 0 and a standard deviation of 1?
- A vision team applies random rotations, flips, and slight color jitter to their training images before each epoch. What is the main goal of this technique?
- A junior engineer proposes tuning the model's hyperparameters directly against the test set to save time, skipping a separate validation set. What is the main problem with this approach?
- In a typical deep learning workflow, what role does the pandas library primarily serve?
- Why do many data scientists prefer developing and exploring a deep learning model inside a Jupyter Notebook rather than a plain script?
- A student asks what makes a value inside a neural network count as a 'trainable parameter.' What is the best definition?
- A computer vision team needs to efficiently load thousands of images in shuffled mini-batches during PyTorch training, ideally using multiple worker processes in parallel. Which class is designed for this?
- In a standard PyTorch training loop, after loss.backward() has computed gradients for every parameter, what is the job of the optimizer object (for example, an instance of torch.optim.SGD)?
- A research team is designing a novel architecture with a custom, non-standard training loop that changes behavior based on intermediate outputs. They choose PyTorch over a high-level framework like Keras. What is the main advantage driving that choice?
- In NumPy terms commonly used to describe deep learning data, what distinguishes a vector from a scalar?
- During model development, a team checks performance on a held-out validation set after every epoch and adjusts hyperparameters based on the results. What is the primary purpose of that validation set?
- In a Keras workflow, after a model has been built and compiled, what does calling model.fit(xtrain, ytrain) do?
- A dataset has one feature where 99% of values fall between 0 and 100, but a handful of extreme outliers reach 100,000. If min-max scaling is applied to squeeze this feature into [0, 1], what problem results?
- A colleague adds np.random.seed(42) and torch.manualseed(42) at the top of a training script before any weight initialization or data shuffling happens. Why bother setting a random seed like this?
- A team has finished all hyperparameter tuning and model selection using their training and validation sets. They now evaluate the final chosen model on a completely separate set of data for the first and only time. What is the main purpose of this last set?
- Before feeding pixel values and other numeric features into a neural network, a data scientist rescales them so all values fall within a common range such as [0, 1]. What is this preprocessing step generally called, and what does it accomplish?
- A dataset is far too large to fit entirely into GPU memory at once. What is the standard, memory-efficient approach for training a neural network on data like this?
- Why is a dataset typically split into training, validation, and test sets rather than using all available data for training?
- A developer wants a quick printed overview of every layer in a Keras model, including each layer's output shape and how many trainable parameters it contributes. Which method provides this?
- In TensorFlow, what fundamentally is a Tensor, the central data structure the framework is named after?
- What defines a Dense (fully connected) layer in a neural network?
- A categorical feature has more than 50 distinct values (for example, a 'city' column with hundreds of possible cities). If this feature is one-hot encoded, what problem is likely to arise?
- A developer needs a NumPy array filled entirely with zeros, with shape (3, 4, 5) to use as a placeholder tensor. Which line of code correctly creates it?
- After defining a Keras model's layers, which method must be called to configure it with an optimizer, loss function, and metrics before training can begin?
- Training crashes partway through with a CUDA out-of-memory warning. The team wants to resolve it quickly without redesigning the model architecture. What is the simplest effective fix?
- What is the main role that the NumPy library plays in a typical deep learning workflow?
- A junior engineer splits a dataset into train and test sets simply by taking the first 80% of rows as train and the last 20% as test, without shuffling first. What common pitfall does this risk?
- A batch of text sequences fed into an RNN has varying lengths — some 10 words, some 40 words. Since a batch tensor requires a single, uniform shape, what technique is used to make all sequences in the batch the same length?
- A Python script uses the built-in pickle library to persist a trained scikit-learn model object to disk, then loads it back later in a separate script. What is pickle's role in this workflow?
- A team wants to augment their image training set by randomly flipping images horizontally during training, without writing custom flip logic from scratch. Which of these offers this built in?
- In frameworks like Keras, what is a callback in the context of model training?
- Beyond the high-level Keras API, what is the main purpose of the TensorFlow core library itself?
Generative & Reinforcement Learning · 50 questions
- A team is training a sequence model on long text passages and finds that a standard RNN struggles to retain information from the beginning of the sequence by the time it reaches the end. Which architectural change most directly addresses this problem?
- An image classifier needs to correctly identify a cat whether it appears in the top-left corner of a photo or the bottom-right. Which architecture is inherently best suited to handle this kind of variation in feature location?
- What is the primary objective of the policy-gradient method in reinforcement learning?
- In a reinforcement learning system, what role does the policy play?
- A reinforcement learning engineer describes their algorithm as 'off-policy.' What does that mean?
- What is the main role of the environment in a reinforcement learning setup?
- A team training a GAN on a dataset of diverse handwritten digits notices that the generator has started producing only a handful of nearly identical-looking digit images, no matter what noise vector is fed in. What is this phenomenon called?
- What is the main purpose of a variational autoencoder (VAE)?
- What is the main strength of reinforcement learning compared with supervised learning?
- In the standard reinforcement learning framework, who or what is the 'agent'?
- In a standard GAN, what does the generator's loss typically represent during training?
- What role does multi-head attention play in a Transformer architecture?
- In an actor-critic reinforcement learning algorithm, what is the goal of the actor component?
- For an image classification task, why is max pooling often preferred over average pooling?
- A data science team has a large collection of unlabeled images and wants a deep learning model to help group similar images together. Which type of model is commonly used for this?
- What is the main role of the convolutional layer in a CNN?
- What is a key advantage of the Transformer architecture over an RNN for natural language processing tasks?
- A GAN trained to generate faces starts producing only a small set of nearly identical-looking faces regardless of the random input it's given. What GAN-specific problem does this describe?
- What is the main goal of transfer learning in deep learning?
- In Q-learning, what does the Q-value Q(s, a) represent?
- A team wants to compress a large set of unlabeled images into compact representations and later reconstruct them, without using any labeled data. Which architecture is the standard choice?
- How does reinforcement learning fundamentally differ from unsupervised learning?
- Which of the following is a common real-world application of reinforcement learning?
- In a typical CNN architecture, what role does the fully connected layer play after the convolutional and pooling layers?
- What is the main role of the discriminator in a Generative Adversarial Network (GAN)?
- A multi-class image classifier's final layer produces raw, unbounded scores called logits. What does applying a softmax function to these logits accomplish?
- An engineer describes an RL agent that learns action-value estimates from experience gathered under an exploratory behavior policy, while the estimates themselves converge toward the optimal greedy policy. Which classic algorithm best matches this description?
- In reinforcement learning, what does it mean to use a 'policy gradient' method to train an agent?
- A team is deciding which of their four active AI projects best exemplifies a reinforcement-learning problem. Which project is it?
- A researcher stacks 150 convolutional layers and finds training accuracy actually gets worse than with a shallower 50-layer version. A colleague suggests adding skip connections, as in a ResNet architecture. Why would this help?
- When configuring a convolutional layer, an engineer sets the padding parameter to 'same' instead of 'valid'. What does this padding parameter actually control?
- A quant team wants to forecast next-day closing prices using several years of daily historical price and volume data. Which type of architecture is best suited to this time-series prediction task?
- In a reinforcement learning setup for a warehouse robot that picks and moves boxes, what role does the reward function serve?
- What fundamental relationship does the Bellman equation express in reinforcement learning?
- A team has trained a GAN to generate synthetic faces and wants to judge how realistic the output images are. Since there's no simple accuracy metric like in classification, what is a common, practical way to evaluate GAN output quality?
- For an image classification task, an engineer compares a convolutional neural network (CNN) against a plain fully connected feedforward network (FFNN) fed the flattened pixel values. What is the main architectural difference that makes the CNN better suited to images?
- While training a vanilla RNN on long text sequences, an engineer notices the loss occasionally spikes to NaN and weight updates become huge and erratic. What is the standard, straightforward fix for this exploding-gradient problem?
- An autoencoder is trained to reconstruct 256x256 images after passing them through a narrow bottleneck layer of only 32 values. Which component of the autoencoder is responsible for compressing the image down to that low-dimensional representation?
- A robotics team's RL agent first learns to predict how the environment will transition given a state and action, then uses that learned model to simulate and plan ahead before acting in the real world. What kind of RL approach is this?
- In an actor-critic reinforcement learning algorithm, what specific role does the critic play?
- An RL researcher describes a method that updates a state's estimated value immediately after each single step, using the observed reward plus the current estimated value of the next state, rather than waiting for an entire episode to finish. What learning approach is being described?
- A team switches from a simple vanilla RNN to an LSTM for a language modeling task involving paragraphs hundreds of words long. What is the main advantage this switch provides?
- Which characteristic is most central to defining reinforcement learning as a distinct machine learning paradigm?
- A team training a robotic arm to perform complex manipulation tasks first has it practice on simplified, easier versions of the task before gradually introducing full-complexity scenarios. What training strategy is this?
- A newcomer to deep learning asks how a GAN fundamentally differs from an autoencoder, since both are described as 'generative' in some sense. What is the most accurate distinction?
- A marketing team wants to generate entirely new, realistic-looking product photos in various settings that were never actually photographed, to use in ad campaigns. Which architecture is best suited to this image-generation task?
- A neural network with an input layer, three hidden layers, and an output layer has no loops or connections feeding backward between layers during a forward pass. What key characteristic does this describe?
- A convolutional layer applies the exact same filter weights as it slides across every position of an input image, rather than learning a completely separate set of weights for each location. What benefit does this parameter sharing provide?
- Summarized in a single sentence, what is the central idea underlying reinforcement learning as a field?
- A team wants to classify a video clip's overall action (such as 'running' versus 'jumping') by processing its sequence of frames in order, letting information from earlier frames inform interpretation of later ones. Which network type is best suited to this sequential frame classification task?
These questions are original practice material and are NOT actual exam questions or brain-dump content. All certification names and marks are the property of their respective owners. This site is independent and not affiliated with, endorsed by, or sponsored by any certification vendor.