Camilo Andres Rodriguez Garzon – Artificial Intelligence

The information is an intent to summarises the theoretical concepts, techniques, and computational methods developed across coursework in artificial intelligence, machine learning, knowledge engineering, natural language processing, computer vision, and data science.

The projects investigated focus on applying advanced modelling techniques and analytical methods to complex problems across multiple domains.

---

Core Areas of Knowledge

The coursework develops knowledge across the following areas:

• Machine Learning and Statistical Modelling
• Deep Learning
• Computer Vision
• Natural Language Processing
• Knowledge Engineering and Knowledge Graphs
• Ai for Health Analysis
• Machine Learning - Financial problem
• Feature Engineering and Dimensionality Reduction

---

1. Machine Learning Foundations

The coursework applies core machine learning theory and modelling techniques used for predictive analytics.

Techniques and Concepts

• Supervised learning
• Unsupervised learning
• Regression modelling
• Model generalisation
• Bias variance tradeoff
• Feature selection
• Model evaluation

Methods Implemented

• Linear Regression
• Ridge Regression
• SVM
• K-means
• AdaBoost
• Lasso Regression
• kNN: k-Nearest Neighbors
• Decision Trees

Optimisation Methods

• Gradient descent
• Regularisation techniques
• Hyperparameter tuning

Evaluation Techniques

• Training and test set validation
• Model performance comparison
• Error analysis

---

2. Deep Learning and Computer Vision

Deep learning methods were applied to image classification tasks.

Core Concepts

• Neural network architectures
• Convolutional neural networks
• Feature hierarchies in images
• Transfer learning

Techniques Used

Neural Networks

• Feedforward neural networks
• Backpropagation
• Non-linear activation functions

Convolutional Neural Networks (CNNs)

Key components:

• Convolution layers
• Feature maps
• Pooling layers
• Fully connected layers

Transfer Learning

Using pretrained models to improve performance and reduce training cost

---

3. Natural Language Processing

The coursework explores computational approaches for analysing and modelling natural language.

Text Processing Techniques

• Tokenisation
• Text normalisation
• Lemmatization
• Corpus analysis

Linguistic Statistical Analysis

The statistical structure of language was explored through:

• Word frequency distributions
• Rank-frequency analysis
• Zipf’s Law

Zipf’s Law describes the inverse relationship between a word’s frequency and its rank within a corpus.

---

4. Representation Learning and Word Embeddings

Text data was represented numerically using vector based models.

Core Concepts

• Distributional semantics
• Vector space representations
• Context-based word representations

Models Studied

• Word2Vec
• Transformer-based embeddings

These methods encode semantic meaning by representing words as vectors in a high dimensional space.

---

5. Transformer Models and Large Language Models

Modern NLP models based on transformer architectures were explored.

Key Concepts

• Self-attention mechanisms
• Contextual word representations
• Bidirectional language modelling

Transformer Architecture Components

• Attention layers
• Encoder blocks
• Positional embeddings
• Token embeddings

Training Approach

• Transfer learning
• Fine tuning pretrained language models

These models allow contextual understanding of text by learning long range dependencies within sequences.

---

6. Knowledge Engineering and Knowledge Graphs

The coursework explores the representation of structured knowledge using graph based models.

Core Concepts

• Knowledge representation
• Ontologies
• Semantic relationships
• Graph data structures

Knowledge Graph Techniques

• Entity relationship modelling
• Graph traversal
• Indirect relationship discovery
• Semantic reasoning

Knowledge graphs represent complex systems through nodes (entities) and edges (relationships), allowing discovery of hidden connections across domains.

---

7. High-Dimensional Data Analysis

Many real-world datasets contain thousands of features. The coursework explores techniques for managing and analysing high-dimensional data.

Challenges

• Curse of dimensionality
• Feature redundancy
• Computational complexity

Techniques Used

Dimensionality Reduction

• Principal Component Analysis (PCA)
• Non-negative Matrix Factorization (NMF)

Feature Filtering

• Variance filtering
• Correlation filtering

These methods reduce dimensionality while preserving meaningful structure in the data.

---

8. Bioinformatics Data Analysis

Machine learning techniques were applied to genomic datasets.

Analytical Concepts

• High dimensional biological data
• Genomic feature analysis
• Data preprocessing pipelines

Techniques Used

• Missing value imputation
• Dimensionality reduction
• Pattern discovery in biological data

This work demonstrates how machine learning can support biomedical research and biological data analysis.

---

9. Financial Machine Learning

Machine learning methods were applied to financial time series analysis.

Key Concepts

• Time series modelling
• Market behaviour analysis
• Feature engineering for financial indicators

Techniques Used

• Technical indicator construction
• Feature selection
• Dimensionality reduction
• Predictive modelling

These techniques allow the modelling of complex financial systems and identification of structural patterns in markets.

---

10. Feature Engineering

Feature engineering is critical for improving model performance.

Techniques Used

• Feature construction
• Feature transformation
• Feature selection
• Correlation filtering

Good features allow machine learning models to capture important patterns within the data.

11. Reinforcement Learning and Decision Modelling

Reinforcement learning is a machine learning paradigm in which an agent learns to make decisions by interacting with an environment and receiving feedback through rewards.

Unlike supervised learning, reinforcement learning does not rely on labelled data. Instead, the agent learns optimal behaviour through sequential interactions and reward signals.

Key Concepts

• Agent
• Environment
• State
• Action
• Reward
• Policy

The objective is to learn a policy that selects actions that maximise long-term cumulative rewards.

Markov Decision Processes (MDPs)

Sequential decision problems can be modelled using Markov Decision Processes (MDPs), defined by:

• States (S) – possible configurations of the environment
• Actions (A) – choices available to the agent
• Transition probabilities (P) – likelihood of moving between states
• Rewards (R) – feedback received after actions

MDPs provide a mathematical framework for modelling and analysing decision making processes in dynamic systems.

Learning Outcome

Understanding reinforcement learning enables modelling of sequential decision making problems, where agents must learn strategies that optimise long term outcomes.

---

Programming and Computational Tools

The coursework required extensive use of scientific computing tools.

Programming

• Python
• Jupyter Notebooks

Data Science Libraries

• NumPy
• Pandas
• Scikit learn
• Matplotlib
• Seaborn

Deep Learning Frameworks

• PyTorch
• HuggingFace Transformers

NLP Libraries

• NLTK
• Gensim

---

Summary of Technical Skills

The coursework develops knowledge in several key areas of artificial intelligence and data science.

Core Skills

• Machine Learning
• Deep Learning
• Natural Language Processing
• Computer Vision
• Knowledge Graph Analysis
• Bioinformatics Data Processing
• Financial Data Modelling
• Feature Engineering
• Dimensionality Reduction
• Statistical Analysis

---

Applications

The techniques developed were applied to multiple domains including:

• Image analysis
• Text processing
• Knowledge graphs
• Genomic data analysis
• Financial markets

---

Links

Side Projects

• Timetabling problem January 2020 – December 2021
  Timetabling university problems is my thesis for the master's degree in applied mathematics that I am finishing.