Chapter 6Artificial Intelligence~1 min read

Deep Learning — TensorFlow आणि PyTorch

Advanced Neural Networks

Deep Learning म्हणजे many layers असलेले neural networks — images, audio, text, video समजणे. Classical ML जिथे fail होतो (raw images, natural language) तिथे Deep Learning excellent results देतो.

TensorFlow vs PyTorch

▸TensorFlow (Google): Production deployment साठी excellent. Keras API — beginners साठी simple.
▸PyTorch (Meta): Research साठी most popular. Pythonic, flexible, easy to debug.
▸Beginners साठी: Keras (TensorFlow वर) — simplest API.
▸Research साठी: PyTorch — maximum flexibility.

Image Classification with Keras — MNIST digits

python

import tensorflow as tf
from tensorflow.keras import layers, models
import numpy as np

# MNIST dataset load — 28x28 handwritten digits
(X_train, y_train), (X_test, y_test) = tf.keras.datasets.mnist.load_data()

# Normalize — 0-255 → 0-1
X_train = X_train.astype('float32') / 255.0
X_test = X_test.astype('float32') / 255.0

# Flatten 28x28 → 784
X_train = X_train.reshape(-1, 784)
X_test = X_test.reshape(-1, 784)

# Model build करा
model = models.Sequential([
    layers.Dense(256, activation='relu', input_shape=(784,)),
    layers.Dropout(0.3),          # Overfitting रोखतो
    layers.Dense(128, activation='relu'),
    layers.Dropout(0.3),
    layers.Dense(10, activation='softmax')  # 10 digits (0-9)
])

model.compile(
    optimizer='adam',
    loss='sparse_categorical_crossentropy',
    metrics=['accuracy']
)

# Train करा
history = model.fit(
    X_train, y_train,
    epochs=10,
    batch_size=128,
    validation_split=0.1
)

# Evaluate
test_loss, test_acc = model.evaluate(X_test, y_test)
print(f"Test Accuracy: {test_acc:.2%}")  # ~98%+

CNN — Convolutional Neural Networks (Images साठी)

CNN for image classification

python

model = models.Sequential([
    # Convolutional layers — image features extract करतात
    layers.Conv2D(32, (3,3), activation='relu', input_shape=(28, 28, 1)),
    layers.MaxPooling2D((2,2)),       # Downsampling
    layers.Conv2D(64, (3,3), activation='relu'),
    layers.MaxPooling2D((2,2)),
    layers.Flatten(),
    layers.Dense(64, activation='relu'),
    layers.Dense(10, activation='softmax')
])
# CNN: Dense NN पेक्षा images साठी खूप better — 99%+ accuracy on MNIST

💡

Transfer Learning वापरा — ImageNet वर pre-trained models (ResNet, VGG, MobileNet) fine-tune करा. Scratch पासून train करणे expensive असते. Keras मध्ये `keras.applications.MobileNetV2()` directly available.

✅ Key Points — लक्षात ठेवा

▸Keras: beginner-friendly, TensorFlow वर
▸PyTorch: research-friendly, flexible
▸CNN: images साठी, RNN: sequences साठी
▸Dropout: overfitting रोखतो
▸Transfer Learning: pre-trained models fine-tune करा

0/11 chapters पूर्ण

मागे — Previous

Scikit-learn — Classical ML

पुढे — Next

Generative AI आणि LLMs