Machine Learning Deep Learning and Neural Network Syllabus

Deep Learning and Neural Network syllabus — full chapter and topic list

Expand any chapter to see its topics and sub-topics. This is the whole examinable outline for Deep Learning and Neural Network in Machine Learning, not a summary of it.

1. 1

   Introduction to Deep Learning

   1 topic

   • Overview of Deep Learning
     • Definition and History
     • Difference Between Machine Learning and Deep Learning
     • Applications of Deep Learning
2. 2

   Introduction to Neural Networks

   overview

   Examined as a single unit within Deep Learning and Neural Network — no further topic split in the official outline.

3. 3

   Neural Networks

   3 topics

   • Basic Concepts
     • Perceptron
     • Activation Functions
     • Feedforward Neural Networks
   • Training Neural Networks
     • Loss Functions
     • Gradient Descent
     • Backpropagation
   • Improving Neural Networks
     • Regularization Techniques
     • Optimization Algorithms
     • Hyperparameter Tuning
4. 4

   Deep Learning Architectures

   3 topics

   • Convolutional Neural Networks (CNNs)
     • Convolutional Layers
     • Pooling Layers
     • Applications of CNNs
   • Recurrent Neural Networks (RNNs)
     • Recurrent Layers
     • Long Short-Term Memory (LSTM)
     • Gated Recurrent Unit (GRU)
   • Generative Adversarial Networks (GANs)
     • Generator and Discriminator
     • Training GANs
     • Applications of GANs
5. 5

   Advanced Topics in Deep Learning

   3 topics

   • Transfer Learning
     • Pre-trained Models
     • Fine-tuning
     • Applications of Transfer Learning
   • Reinforcement Learning
     • Markov Decision Process
     • Q-Learning
     • Deep Q-Networks (DQN)
   • Natural Language Processing (NLP)
     • Word Embeddings
     • Sequence-to-Sequence Models
     • Transformers
6. 6

   Practical Implementation

   3 topics

   • Deep Learning Frameworks
     • TensorFlow
     • PyTorch
     • Keras
   • Model Deployment
     • Saving and Loading Models
     • Serving Models
     • Scalability and Performance
   • Case Studies and Projects
     • Image Classification
     • Natural Language Processing
     • Reinforcement Learning Projects

Deep Learning and Neural Network flashcards for Machine Learning

19 of 50 cards from the Deep Learning and Neural Network deck — real questions with worked answers.

1. What is Deep Learning and how does it relate to machine learning?

   Deep Learning is a subfield of machine learning that uses artificial neural networks with multiple hidden layers (deep architectures) to automatically learn hierarchical feature representations from raw data, rather than relying on hand-engineered features.

2. What is the universal approximation theorem?

   It states that a feedforward neural network with a single hidden layer containing a finite number of neurons and a non-linear activation can approximate any continuous function on a compact subset of $\mathbb{R}^{n}$ to arbitrary accuracy, given enough hidden units.

3. Write the equation for a single artificial neuron's output (forward computation).

   $$y = \phi\left(\sum_{i=1}^{n} w_{i} x_{i} + b\right) = \phi(\vec{w}^{\top}\vec{x} + b)$$ where $\phi$ is the activation function, $\vec{w}$ the weights, $\vec{x}$ the inputs, and $b$ the bias.

4. Define the sigmoid activation function and its output range.

   $$\sigma(x) = \frac{1}{1 + e^{-x}}$$ Its output lies in the range $(0, 1)$, making it useful for representing probabilities.

5. Define the hyperbolic tangent (tanh) activation function and its range.

   $$\tanh(x) = \frac{e^{x} - e^{-x}}{e^{x} + e^{-x}}$$ Its output range is $(-1, 1)$, and it is zero-centered, unlike the sigmoid.

6. Define the ReLU activation function and state one key advantage.

   $$\text{ReLU}(x) = \max(0, x)$$ It mitigates the vanishing gradient problem for positive inputs and is computationally cheap, which speeds up training.

7. What problem does Leaky ReLU solve, and what is its formula?

   It addresses the 'dying ReLU' problem where neurons output zero permanently. $$\text{LeakyReLU}(x) = \begin{cases} x & x > 0 \\ \alpha x & x \leq 0 \end{cases}$$ with a small constant such as $\alpha = 0.01$.

8. Write the softmax function used for multi-class classification output.

   $$\text{softmax}(z_{i}) = \frac{e^{z_{i}}}{\sum_{j=1}^{K} e^{z_{j}}}$$ It converts a vector of $K$ logits into a probability distribution summing to $1$.

9. What is the cross-entropy loss for multi-class classification?

   $$L = -\sum_{i=1}^{K} y_{i} \log(\hat{y}_{i})$$ where $y_{i}$ is the true (one-hot) label and $\hat{y}_{i}$ the predicted probability for class $i$.

10. Write the mean squared error (MSE) loss formula used in regression.

    $$\text{MSE} = \frac{1}{n}\sum_{i=1}^{n}(y_{i} - \hat{y}_{i})^{2}$$ where $y_{i}$ is the true value and $\hat{y}_{i}$ the prediction.

11. What is backpropagation?

    An algorithm that computes gradients of the loss with respect to every weight by applying the chain rule of calculus backward through the network, layer by layer, enabling gradient-based optimization.

12. State the gradient descent weight update rule.

    $$w \leftarrow w - \eta \frac{\partial L}{\partial w}$$ where $\eta$ is the learning rate and $\frac{\partial L}{\partial w}$ the gradient of the loss with respect to the weight.

13. Compare batch, stochastic, and mini-batch gradient descent.

    Batch GD uses the entire dataset per update (stable but slow); Stochastic GD (SGD) uses one sample per update (noisy, fast); Mini-batch GD uses a small subset (e.g. 32-256 samples), balancing efficiency and stability.

14. What is the role of the learning rate $\eta$, and what happens if it is too large or too small?

    It controls step size in parameter updates. Too large causes divergence or oscillation around the minimum; too small causes very slow convergence and risk of getting stuck.

15. Describe the momentum optimization update.

    $$v_{t} = \gamma v_{t-1} + \eta \nabla L, \qquad w \leftarrow w - v_{t}$$ Momentum accumulates a velocity term (typically $\gamma = 0.9$) to accelerate descent in consistent directions and dampen oscillations.

16. What is the Adam optimizer and what does it combine?

    Adam (Adaptive Moment Estimation) combines momentum (first moment of gradients) and RMSProp (second moment / adaptive per-parameter learning rates), using bias-corrected estimates $\hat{m}_{t}$ and $\hat{v}_{t}$ for the update $w \leftarrow w - \eta \frac{\hat{m}_{t}}{\sqrt{\hat{v}_{t}} + \epsilon}$.

17. What is the vanishing gradient problem?

    In deep networks, gradients shrink exponentially as they propagate backward through many layers (especially with sigmoid/tanh), so early layers learn extremely slowly or not at all.

18. What is the exploding gradient problem and one common remedy?

    Gradients grow exponentially during backpropagation, causing unstable, divergent updates. A common remedy is gradient clipping, which rescales gradients when their norm exceeds a threshold.

19. Distinguish overfitting from underfitting.

    Overfitting: model fits training data too closely (low train error, high test error) capturing noise. Underfitting: model is too simple to capture the underlying pattern (high train and test error).

See more Deep Learning and Neural Network flashcards →

Planning Deep Learning and Neural Network for Machine Learning

Deep Learning and Neural Network (Machine Learning) FAQ