Advanced Topics in Machine Learning

Endang Wahyu Pamungkas, Ph.D.

Introduction to Neural Networks
● Definition: A neural network is a series of algorithms that
endeavors to recognize underlying relationships in a set of
data through a process that mimics the way the human
brain operates.
● Structure of Neural Networks:
○ Composed of nodes (neurons) organized in layers: input
layer, hidden layers, and output layer.
○ Neurons are connected by weights, and these
connections are used to propagate input data through
the network.
Introduction to Neural Networks
● Functionality:
○ Each neuron receives input, applies a function (typically
non-linear) to it, and passes the output to the next
layer.
○ The network learns by adjusting the weights based on
the error of the output compared to the expected result.
Introduction to Neural Networks
What is Deep Learning?
● Definition: Deep learning is a subset of machine learning
in artificial intelligence (AI) that has networks capable of
learning unsupervised from data that is unstructured or
unlabeled.
● Key Characteristics:
○ Uses multiple layers of neural networks to extract
progressively higher-level features from the raw input.
○ Can automatically discover the representations needed
for feature detection or classification from raw data.
What is Deep Learning?
What is Deep Learning?
● Types of Deep Neural Networks:
○ Convolutional Neural Networks (CNNs): Used primarily
for image processing, classification, and segmentation.
○ Recurrent Neural Networks (RNNs): Suited for
sequential data like time series or natural language.
○ Autoencoders: Used for learning efficient data codings
in an unsupervised manner.
What is Deep Learning?
● Recurrent Neural Networks (RNNs):
○ Designed to process sequential data, like time series or
natural language.
○ Can remember previous inputs in memory by creating
loops in the network.
○ Issue of vanishing and exploding gradients, addressed
by LSTM (Long Short-Term Memory) and GRU (Gated
Recurrent Unit) networks.
Recurrent Neural Network
What is Deep Learning?
● Convolutional Neural Networks (CNNs):
○ Ideal for processing grid-like data such as images.
○ Utilize convolutional layers to automatically detect
features and patterns like edges and shapes.
○ Key components include convolutional layers, pooling
layers, and fully connected layers.
Convolutional Neural Network
Cutting-edge Developments in Deep Learning
● Attention Mechanisms and Transformers:
○ Attention mechanisms allow neural networks to focus
on specific parts of the input sequentially, improving
performance in tasks like translation and image
recognition.
○ Transformers, built on attention mechanisms, have
revolutionized natural language processing (NLP) with
models like BERT and GPT, facilitating breakthroughs in
generating human-like text.
Cutting-edge Developments in Deep Learning
● Neural Network Pruning and Compression:
○ Techniques to reduce the size of neural networks,
making them more efficient for deployment on devices
with limited hardware resources.
○ Involves removing neurons or connections that
contribute least to the network's output, thus reducing
computational complexity and memory usage.
Application of Deep Learning
● Image and Video Recognition:
○ Facial recognition in security systems.
○ Object and scene recognition in photographs and
videos.
● Natural Language Processing (NLP):
○ Language translation services like Google Translate.
○ Speech recognition systems like Amazon’s Alexa and
Apple’s Siri.
Application of Deep Learning
● Medical Field:
○ Disease detection and diagnosis through medical
imaging analysis.
○ Drug discovery and genomics.
● Autonomous Vehicles:
○ Used in self-driving cars for making real-time
navigation decisions.
Advance Deep Learning Technique
● Transfer Learning:
○ Utilizing a pre-trained model on a new problem to
leverage learned features, reducing the need for a large
dataset.
○ Common in image and speech recognition tasks where
models trained on large datasets can be fine-tuned for
specific tasks.
Advance Deep Learning Technique
● Generative Adversarial Networks (GANs):
○ A method of AI algorithms used in unsupervised
machine learning, implemented by a system of two
neural networks contesting with each other in a
zero-sum game framework.
○ Applications include creating realistic images, art, and
enhancing low-resolution photos.
Advance Deep Learning Technique
● Reinforcement Learning in Deep Networks:
○ Combining deep learning with reinforcement learning
to create systems that can make intelligent decisions
based on complex inputs.
○ Examples include AI agents that learn to play video
games or optimize trading strategies in finance.
Reinforcement Learning in Deep Learning
Challenges and Ethical Considerations in Deep Learning
● Challenges:
○ Data Requirements: Deep learning models often require
massive amounts of data, making them impractical for
tasks with limited data.
○ Computational Cost: Training deep learning models can
be resource-intensive, requiring significant computing
power and time.
○ Interpretability: Deep neural networks are often viewed
as "black boxes" due to their complex structure, making
it difficult to understand how decisions are made.
Challenges and Ethical Considerations in Deep Learning
● Ethical Considerations:
○ Bias and Fairness: Ensuring that AI systems do not
perpetuate or amplify social biases.
○ Privacy: Addressing concerns related to the collection
and use of large datasets, especially personal data.
○ Autonomy and Control: Ensuring that decisions made by
AI systems are transparent and can be overridden or
controlled by humans when necessary.
Challenges and Ethical Considerations in Deep Learning