Deep Learning

Simulating the Human Mind in Thinking

AI Programming & Machine Learning 4.4(91)12h 40m64 lectures5 sections

What you'll learn

Understand the neural network and binary classification
Use the Logistic Regression algorithm
Compute the cost function and derivatives
Build the shallow neural network and apply gradient descent
Build the deep neural network
Use the Keras library for prediction and classification
Work with TensorFlow and its applications
Apply the convolutional neural network
Build an LSTM model and the recurrent network
Use Autoencoders

About this course

This course teaches Deep Learning and neural networks step by step, from the first idea to building models in code. You start with the neural network, supervised learning, binary classification and the Logistic Regression algorithm, then the cost function, computing derivatives and the computation graph. Next you learn the shallow neural network, activation functions and gradient descent, then the deep neural network, forward propagation and matrix dimensions. You move on to practice using the Keras library for prediction and classification, then TensorFlow and its basics, then the convolutional neural network, then sequential data, the recurrent neural network and the LSTM model, and you finish with Autoencoders. Every part comes with a hands-on application and downloadable files. The course suits anyone who knows Python basics and wants to enter the world of artificial intelligence. After finishing, you will understand how neural networks work and be able to build prediction and classification models using Keras and TensorFlow.

Expected outcomes

Master the use of neural networks to build efficient deep learning models.
Analyze data and apply deep learning to solve complex problems.
Design and implement real projects using advanced deep learning tools.
Understand how to optimize model performance using advanced techniques.
Prepare yourself to enter the field of artificial intelligence and deep learning as a professional.

Learner reviews

These reviews are from learners who bought the content and completed at least 50% of it — there are no reviews from anyone who hasn't gone through the material.

4.4

★★★★☆

91 reviews

5★

4★

3★

2★

1★

★★★★★احمد م.·منذ 4 شهرشاهد 100٪
كورس مفيد جدا وعملي ونظري في نفس الوقت فالكورس يدمج المفاهيم النضريه مع التطبيق العملي

تم عرض كل التقييمات (1)

Course content

Fundamentals of Deep Learning and Neural Networks

31 lectures

IntroductionOutlines the first chapter of the deep learning course: the neural network concept, programming basics with forward and backward propagation, building a one-hidden-layer network and then a deep network.1:56
Neural NetworkDefines a neural network through a house price prediction example, where neurons form functions linking inputs like size and room count to the output price through hidden layers the network builds itself.8:13
Supervised Learning and Neural NetworksExplains supervised learning with neural networks through examples like real estate, online ads, translation and self-driving cars, the network type suited to each, and structured versus unstructured data.7:52
Binary ClassificationIntroduces binary classification in supervised learning, showing how an image becomes an input vector X with outputs Y, plus the notation for training and test data and matrix dimensions.13:55
Load Symbol List
Logistic Regression AlgorithmExplains the Logistic Regression algorithm for binary classification: the parameters W and B, and the sigmoid function that bounds the output between zero and one, including its behavior for large positive and negative values.8:51
Cost FunctionExplains the Cost Function in binary classification: the difference between the loss of one example and the average loss over all test data, its log-based formula, and its role in improving W and B.11:43
Calculate Variable ValuesExplains how gradient descent finds the best W and B by minimizing the convex cost function: starting from initial values then moving iteratively using the derivative and the learning rate.14:00
CalculusExplains the derivative: the slope of a straight line as change in y over change in x, the rule for a squared function, and the derivative of the logarithm, since most of the course relies on it.13:34
Computational GraphExplains the computation graph through a function of three variables, showing how it orders the steps of forward propagation to compute the value, ahead of backward propagation and derivatives.6:23
Computational Graph for Derivative CalculationUses the computation graph to compute derivatives through backward propagation from right to left, working out how changing each variable affects the function output with the same example.15:24
Derivative Calculation for Algorithm ApplicationShows how to compute the derivatives dz, dw1, dw2 and db to apply gradient descent in Logistic Regression on a single example with two features, using the computation graph and the rule dz = a - y.12:51
Derivative Calculation for Several ExamplesExtends derivative computation to m examples: accumulating dw1, dw2, db and dz in a for loop, dividing by m for the average and updating W and B, while noting the loop is slow.10:34
Understanding DirectionsIntroduces vectorization as a replacement for for loops, comparing in Python with NumPy the np.dot product against a loop, which shows the vectorized version is far faster.15:59
Load Practical Example
Using Directions in Linear Regression AlgorithmApplies vectorization to Logistic Regression: dz as a vector equal to A - Y, computing dw and db without a for loop, then updating W and B with the learning rate.14:45
Distribution in PythonExplains Broadcasting in Python and NumPy using a food calorie table: summing columns, dividing each row by the totals, and using reshape to make sure of the array shape.14:29
Load Practical Example
Introduction to Hollow Neural NetworkIntroduces the shallow neural network with one hidden layer, the notation for layers and nodes, and computing Z and A for each layer from the inputs to the output.11:08
Hollow Neural Network Representation and Output CalculationShows how to represent a shallow neural network and compute the output of its first layer for one node and then all nodes using vectors, with dimensions of W, b and Z such as a 4 by 3 matrix and a 4 by 1 column.17:30
Output Calculation for Several ExamplesExplains computing the output for m training examples instead of one, stacking the examples as columns in a matrix X and computing the Z and A matrices of the hidden layer in one go.10:25
Activation FunctionCompares activation functions: sigmoid for binary classification, tanh ranging from negative one to one, ReLU, and Leaky ReLU to fix the problem with negative values.11:23
Calculus for Activation FunctionsComputes the derivatives of activation functions: sigmoid with g(1-g), tanh with one minus the square of the function, and ReLU whose derivative is zero for negative values and one for positive ones.15:58
Loading Function Format
Implementing Gradient Descent for the Hollow Neural NetworkApplies gradient descent to the shallow network: forward propagation with its four equations, then backward propagation to compute dz2, dw2, db2, dz1, dw1 and db1 for updating the parameters.18:27
Introduction to Deep Neural NetworksIntroduces the deep neural network: its definition with more than one hidden layer, the notation L for the number of layers and n for the nodes in each layer, and a review of forward and backward propagation.11:19
Forward Propagation in Deep Neural NetworksExplains forward propagation in a deep neural network: computing Z and A layer by layer, the general rule Z = W·A(l-1) + b then A = g(Z), and applying it to the matrix of all examples.13:52
How to Calculate Matrix DimensionsExplains how to work out matrix dimensions in a deep neural network for W, b, Z and A, helping avoid common mistakes, noting that dW and db share the dimensions of W and b.22:52
Core Components for Implementing Deep Neural NetworksPresents the building blocks for implementing a deep network: a forward step that takes A(l-1) and stores Z in a cache, and a backward step that uses the cache to compute dW, db and dZ before updating W and b.14:14
Implementing Core Components for Gradient Descent CalculationExplains the Backward Propagation formulas for computing DW, DB and DA at each layer using vectorization, how to calculate them in Python, and leads into the topic of Hyperparameters.11:22
HyperparametersClarifies the difference between Parameters such as W and B and Hyperparameters such as the learning rate, number of layers, activation function and iterations, and how tuning them improves performance.4:44

Developing Deep Neural Networks and Training Models

6 lectures

Libraries Used for DevelopmentCompares the popular deep learning libraries TensorFlow, PyTorch and Keras and when each fits, with Keras for quick simple builds and the others for more control or academic research.7:08
Practical Example of Using Keras Library for PredictionA theoretical walkthrough of using Keras to predict concrete strength from its ingredients, covering hidden layers with the ReLU function and splitting the data into training and test sets.11:12
First Practical ApplicationA first hands-on lab on Kaggle for writing code without installing software, building a Keras model that predicts concrete strength, training it and using it to predict for new concrete.26:35
Loading Data File
Practical Example of Using Keras Library for ClassificationA practical example of classification with the Keras library: one-hot encoding the classes and building a network of Dense layers whose output layer has four nodes, one per class, producing probabilities.13:51
Second Practical ApplicationA second hands-on image classification with Keras on MNIST: loading the data and checking its shape, converting labels with to_categorical, then building and training the model with fit to reach about 97% accuracy.35:06

Using TensorFlow

9 lectures

Introduction to TensorFlowAn introduction to TensorFlow: its C and C++ backend, running on CPU or GPU, and the data flow graph concept that connects multi-dimensional arrays with operations.16:13
Fundamentals Part OneExplains TensorFlow basics: the computation graph, multi-dimensional arrays (tensors) and their ranks from scalar to vector and matrix, with an example of multiplying W and X and adding B.9:04
Fundamentals Part TwoCovers TensorFlow basics: importing the library, defining constant values with the constant function, building a function that adds values and takes a parameter, and telling constants apart from variables.17:50
Fundamentals Part ThreeShows how to create multi-dimensional arrays in TensorFlow, check their shape, and multiply matrices with the matmul function, then display the result.9:00
Application on Prediction Part OneA quick review of linear regression and its TensorFlow implementation on CO2 emission versus engine size data, defining the values A and B and computing the difference between predictions and actual values.21:33
Application on Prediction Part TwoContinues linear regression in TensorFlow: setting initial values and the learning rate, training the model with gradient descent, and plotting the resulting line over the actual data.26:13
Download Data File
Application on Classification Part OneApplies Logistic Regression classification in TensorFlow on the Iris data: loading and splitting the data, defining W and B as variables, and applying the activation function to pick the most probable of three classes.32:22
Application on Classification Part TwoFinishes the TensorFlow classification application: the training loop with a Keras loss function, computing weights, bias and optimization, tracking accuracy, then wrapping up the TensorFlow basics section.24:28

Convolutional Neural Network

8 lectures

Introduction to Convolutional Neural NetworksIntroduces convolutional neural networks (CNN), contrasting them with the traditional image classification approach of hand-crafted features, and listing applications such as skin cancer detection and self-driving cars.7:54
Basic PrinciplesExplains the core principles of a CNN for object detection: it extracts features from images automatically and combines them step by step to recognize an object, as in the example of identifying buildings.5:46
Architecture of Convolutional Neural NetworksExplains the basic architecture of a CNN: representing an image as a matrix, applying a filter or kernel, picking the highest value, and finally classifying the image through probabilities.18:26
Practical Application Using Traditional Method Part OneA hands-on traditional approach to image classification: one-hot encoding labels, defining weights W and bias B, the Softmax function, and setting an SGD optimizer with a learning rate.27:20
Practical Application Using Traditional Method Part TwoContinues the traditional approach by building the training step with GradientTape and cross entropy to update W and B, then tracking accuracy across epochs until it reaches about 84%.14:01
Download the Application
Practical Application on the Convolutional Neural NetworkA hands-on CNN lesson: applying filters and pooling layers, training with the Adam algorithm, and raising accuracy to 94% compared with 84% from the traditional approach.29:50
Download the Application

Building Different Models for Deep Learning

10 lectures

Introduction to Building Different Models for Deep LearningIntroduces recurrent neural networks (RNN) for sequential data, with example applications such as stock price prediction, sentiment analysis, word suggestion, translation and text-to-speech.4:45
Time Series DataExplains sequential data and why traditional neural networks cannot handle it, since they do not remember earlier outputs, and how an RNN solves this by retaining them.3:06
Recurrent Neural NetworkExplains the RNN architecture and how it handles sequential data using a previous state as memory, its uses, and its drawbacks such as sensitivity and instability, leading into LSTM.6:48
LSTM ModelExplains the LSTM model, a recurrent network that fixes RNN problems using a memory cell and three gates for input, reading and forgetting, with the computation repeating until the data ends.6:45
Practical Application of LSTM Model BasicsA hands-on lesson on LSTM and Staged LSTM basics in Kaggle: defining the inputs and previous state, building one cell and then two stacked cells, and printing the outputs and memory.23:14
Download the Application
Introduction to New AutoencodersIntroduces Autoencoders as neural networks for unsupervised learning and their uses: feature extraction, data compression and dimensionality reduction, with a comparison to PCA on the MNIST data.5:20
Architecture of AutoencodersExplains the Autoencoder architecture in two parts: an encoder that compresses image features into fewer dimensions and a decoder that rebuilds the image, with the compressed values reused for tasks like classification.4:02
Practical Application on AutoencodersA hands-on Autoencoder lesson on MNIST: building an encoder with layers of 256, 128 and 32 units and a decoder restoring 784 inputs, trained with RMSProp to rebuild images close to the originals.19:13
Download the Application