Unit 11/1

Unit 11 Part 2 - CNN Model Activity

Throughout this project, I systematically demonstrated the application of machine learning knowledge to solve a real-world problem: object recognition using the CIFAR-10 dataset. This work showcases my understanding of legal, social, ethical, and professional issues in the context of machine learning, along with technical and collaborative achievements.

Legal, Social, Ethical, and Professional Considerations

  • Critically evaluated broader implications of machine learning technologies, such as their impact on data privacy, algorithmic bias, and fairness.
  • Ensured model development adhered to ethical guidelines, avoiding overfitting and addressing transparency and explainability requirements.
  • Acknowledged the importance of these considerations in high-stakes scenarios, such as autonomous systems or security applications.

Applicability and Dataset Challenges

  • Worked with the CIFAR-10 dataset, addressing unique challenges like low resolution and class similarities (e.g., automobiles vs. trucks).
  • Applied preprocessing techniques, including normalization and one-hot encoding, to prepare the data for the CNN model.
  • Introduced data augmentation (e.g., flipping, rotation, shifting) to enhance generalization and address overfitting.

Application and Critical Appraisal of Machine Learning Techniques

  • Designed and implemented a convolutional neural network (CNN) with advanced enhancements:
    • Batch Normalization: Stabilized training and improved convergence.
    • L2 Regularization: Reduced overfitting by penalizing large weights.
    • Learning Rate Scheduling: Dynamically adjusted the learning rate for optimal convergence.
  • Achieved a validation accuracy of 71.32% over 20 epochs, while identifying opportunities for improvement (e.g., adopting deeper architectures like ResNet).
  • Compared model results to a Kaggle implementation, critically evaluating trade-offs between simplicity and robustness.

Collaboration and Professional Development

  • Engaged in team collaboration by integrating peer feedback and participating in discussions that refined my model design and evaluation practices.
  • Maintained clear and comprehensive documentation of the development process, reflecting real-world machine learning team practices.
  • Demonstrated critical reflection by analyzing outcomes and proposing refinements for future iterations.

Evaluation of Final Project vs. Initial Proposal

The final project (Unit 11) presented significant technical advancements compared to the initial proposal (Unit 6). The Unit 6 project, focused on Airbnb business analytics, emphasized exploratory data analysis, statistical modeling, and visualization. In contrast, the CIFAR-10 task in Unit 11 required the design and implementation of advanced deep learning techniques, including convolutional neural networks (CNNs). This progression marked a shift from high-level data insights to hands-on application of machine learning algorithms.

While both projects benefited from Python as a programming language, Unit 11 was notably easier to implement due to Python’s specialized libraries like TensorFlow and Keras, which streamlined neural network design and optimization. In contrast, Unit 6 relied heavily on pandas and matplotlib for manual data manipulation and visualization, requiring more effort to achieve actionable insights. The structured approach enabled by Python in Unit 11 allowed me to focus more on improving model performance and less on handling raw data, highlighting the advantage of domain-specific libraries for complex tasks.

Overall, the transition from Unit 6 to Unit 11 demonstrated my growth in both conceptual understanding and technical expertise, showcasing my ability to tackle diverse machine learning problems and refine my skills across different domains.

Code Showcase

						
import numpy as np
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropout
from tensorflow.keras.utils import to_categorical
from sklearn.model_selection import train_test_split

# Step 1: Load CIFAR-10 dataset
import pickle
import os

# Define the folder containing the CIFAR-10 files
cifar_folder = 'cifar-10-batches-py'

def load_cifar10_batch(file_path):
	with open(file_path, 'rb') as file:
		batch = pickle.load(file, encoding='bytes')
	data = batch[b'data']
	labels = batch[b'labels']
	return data, labels

def load_cifar10_data(folder_path):
	train_data = []
	train_labels = []
	for i in range(1, 6):  # Files: data_batch_1 to data_batch_5
		data, labels = load_cifar10_batch(os.path.join(folder_path, f'data_batch_{i}'))
		train_data.append(data)
		train_labels.extend(labels)
	train_data = np.concatenate(train_data)
	train_labels = np.array(train_labels)

	# Load test batch
	test_data, test_labels = load_cifar10_batch(os.path.join(folder_path, 'test_batch'))

	return train_data, train_labels, test_data, test_labels

# Load the CIFAR-10 dataset
train_data, train_labels, test_data, test_labels = load_cifar10_data(cifar_folder)

# Step 2: Preprocess the data
train_data = train_data.reshape(-1, 3, 32, 32).transpose(0, 2, 3, 1) / 255.0
test_data = test_data.reshape(-1, 3, 32, 32).transpose(0, 2, 3, 1) / 255.0


# Split training data into training and validation sets
train_data, val_data, train_labels, val_labels = train_test_split(
	train_data, train_labels, test_size=0.1, random_state=42
)

# One-hot encode the labels
train_labels = to_categorical(train_labels, num_classes=10)
val_labels = to_categorical(val_labels, num_classes=10)
test_labels = to_categorical(test_labels, num_classes=10)

# Step 3: Define the CNN model
model = Sequential([
	Conv2D(32, (3, 3), activation='relu', input_shape=(32, 32, 3)),
	MaxPooling2D(pool_size=(2, 2)),
	Conv2D(64, (3, 3), activation='relu'),
	MaxPooling2D(pool_size=(2, 2)),
	Flatten(),
	Dense(128, activation='relu'),
	Dropout(0.5),
	Dense(10, activation='softmax')
])

# Step 4: Compile the model
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

# Step 5: Train the model
history = model.fit(
	train_data, train_labels,
	validation_data=(val_data, val_labels),
	epochs=20,
	batch_size=64
)

print("Model training completed successfully.")