3: Deep Learning and Neural Networks (Perceptrons, Multilayer Perceptrons - MLP, Activation Functions: ReLU/Sigmoid/Softmax, Backpropagation & Gradient Descent, CNNs, RNNs & Transformers, Generative AI & LLMs) — البرمجة والذكاء الاصطناعي الصف الثاني الثانوي لغات

درس «3: Deep Learning and Neural Networks (Perceptrons, Multilayer Perceptrons - MLP, Activation Functions: ReLU/Sigmoid/Softmax, Backpropagation & Gradient Descent, CNNs, RNNs & Transformers, Generative AI & LLMs)» فى مادة البرمجة والذكاء الاصطناعي لـالصف الثاني الثانوي لغات من وحدة «Artificial Intelligence and Machine Learning». This lesson provides an advanced examination of Deep Learning. It dissects the Artificial Perceptron, forward propagation, and non-linear activation functions (ReLU, Sigmoid, Softmax). It details the mathematical training pipeline: loss functions, Backpropagation via the Calculus Chain Rule, and Gradient Descent optimization. Finally, it surveys major deep...

الدرس ده من وحدة «Artificial Intelligence and Machine Learning» فى مادة البرمجة والذكاء الاصطناعي لـالصف الثاني الثانوي لغات على منصة دروسات.

الدرس متاح بالاشتراك: ترم كامل أو المادة لوحدها.

ملخص الدرس

This lesson provides an advanced examination of Deep Learning. It dissects the Artificial Perceptron, forward propagation, and non-linear activation functions (ReLU, Sigmoid, Softmax). It details the mathematical training pipeline: loss functions, Backpropagation via the Calculus Chain Rule, and Gradient Descent optimization. Finally, it surveys major deep architectures: Convolutional Neural Networks (CNNs) for vision, Recurrent Neural Networks (RNNs/LSTMs) for sequences, and the Self-Attention Transformer architecture powering modern Large Language Models (LLMs) and Generative AI.

هتتعلم إيه فى الدرس ده

• Deep Learning uses multilayer neural networks to learn hierarchical representations.
• The perceptron computes a weighted sum plus bias: z = sum(w*x) + b, followed by an activation f(z).
• ReLU (max(0, z)) is the standard hidden layer activation, mitigating vanishing gradients.
• Backpropagation calculates loss gradients using the chain rule to update weights via gradient descent.
• CNNs use convolution filters and pooling layers for computer vision tasks.
• RNNs and LSTMs process sequential time-series data using internal memory loops.
• Transformers utilize parallel Self-Attention (Q, K, V), forming the backbone of modern LLMs.