24 min Intermediate Prerequisite: Ensemble Methods

Deep Learning Intro

Neural Networks samjho � AI ka asli foundation!

Yeh Lesson Kyun important hai?

Deep learning neural networks ka advanced form hai � images, text, speech sab deep learning se process hota hai. Aaj kal ka AI revolution deep learning ki wajah se hua hai. GPT, DALL-E, self-driving cars � sab deep learning pe based hain.

NEURAL NETWORK

Brain jaisa structure jo data seekhta hai

LAYERS

Input ✓ Hidden ✓ Output layers

🔧

ACTIVATION

Non-linearity add karta hai

BACKPROPAGATION

Error se seekhne ka tarika

Neural Network Kya Hai?

Neural network inspired hai human brain se. Jaise brain mein neurons ek dusre se connected hote hain, waise hi artificial neural network mein nodes (neurons) layers mein arranged hote hain.

Input Hidden 1 Hidden 2 Output

Layers samjho

Input Layer: Yahan data aata hai � numbers, pixels, features.

Hidden Layers: Yahan actual kaam hota hai. Har neuron data ko process karta hai aur apna contribution deta hai. Multiple hidden layers hone se Deep Learning kehte hain.

Output Layer: Final prediction ya classification milta hai.

Activation Functions

Agar sirf linear transformations karte toh network sirf linear relationships seekh paata. Activation functions non-linearity add karte hain � isliye network complex patterns seekh sakta hai.

Function Formula Kab Use Hoti Hai
ReLU max(0, x) Hidden layers mein sabse popular
Sigmoid 1 / (1 + e^(-x)) Binary classification output
Tanh (e^x - e^(-x)) / (e^x + e^(-x)) Scaled output, zero-centered
Softmax e^(xi) / Se^(xj) Multi-class classification output

Backpropagation � Learning Ka Tarika

Neural network kaise seekhta hai? Backpropagation se!

Yeh 3 steps mein kaam karta hai:

Step-by-Step Process

  • Forward Pass: Data input se output tak jaata hai, prediction milta hai
  • Loss Calculate: Prediction aur actual value ke beech ka error nikalte hain
  • Backward Pass: Error ko output se input tak wapas bhejte hain aur weights update karte hain

Har iteration mein yeh process hota hai jab tak network ka prediction improve na ho jaaye. Isi ko training kehte hain!

Code Example: sklearn se Neural Network

Pehle sklearn ka MLPClassifier dekhte hain � yeh simple aur beginner-friendly hai:

Python � MLPClassifier 
# Simple neural network with sklearn
from sklearn.neural_network import MLPClassifier
from sklearn.model_selection import train_test_split
from sklearn.datasets import make_classification
import numpy as np

# Sample data banao
X, y = make_classification(n_samples=200, n_features=5, random_state=42)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)

# Neural network banao aur train karo
model = MLPClassifier(hidden_layer_sizes=(100, 50), max_iter=500)
model.fit(X_train, y_train)

# Accuracy check karo
print(f"Accuracy: {model.score(X_test, y_test):.2%}")

TensorFlow/Keras Preview

Yeh real deep learning framework hai. Keras API bahut clean aur readable hai:

Python � TensorFlow/Keras 
# TensorFlow/Keras preview
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

try:
 import tensorflow as tf

 # Sequential model banao
 model = tf.keras.Sequential([
 tf.keras.layers.Dense(64, activation='relu', input_shape=(5,)),
 tf.keras.layers.Dense(32, activation='relu'),
 tf.keras.layers.Dense(1, activation='sigmoid')
 ])

 # Model compile karo
 model.compile(
 optimizer='adam',
 loss='binary_crossentropy',
 metrics=['accuracy']
 )

 print("Keras model created!")
 model.summary()

except ImportError:
 print("TensorFlow not installed - preview only")

Kya samjhe?

  • Sequential: Layers ek ke baad ek stack hain
  • Dense: Fully connected layer � har neuron agle layer ke sabse connected hai
  • activation='relu': Hidden layers mein ReLU use hota hai
  • activation='sigmoid': Binary output ke liye (0 ya 1)
  • adam optimizer: Sabse popular optimizer � fast aur efficient

✓ Interactive: Apna Neural Network Banao

Neeche code likho ya edit karo aur "Run" dabao:

// Output yahan dikhega...

Exercise

Sawaal: Neural network mein backpropagation kya hai?

Jawab: Error ko backward pass karke weights update karna. Yeh output layer se input layer tak error propagate karta hai aur gradient descent use karke weights adjust karta hai taaki next time prediction better ho.

Key Takeaways

Yaad Rakho

  • Neural network layers mein neurons hote hain � input, hidden, output
  • Deep learning = multiple hidden layers
  • Activation functions non-linearity add karte hain
  • Backpropagation se network seekhta hai � error se weights update hote hain
  • sklearn se basics seekho, TensorFlow/PyTorch se real projects banao
✓ Previous: Ensemble Methods Course Complete!