¹û¶³Ó°Ôº

Aims

This module aims to introduce advanced machine learning techniques for financial applications ranging from derivatives pricing, model calibration, portfolio allocation and hedging, credit scoring, fraud detection, investment decision, and risk-management. Text mining will also be introduced for the use of increasingly available alternative data sources.

Intended learning outcomes

On successful completion of the module, a student will be able to:

1. Familiarise with a range of advanced machine learning algorithms, including convolutional neural networks, recurrent neural networks and reinforcement learning.
2. Identify the appropriate machine learning algorithm for a given financial problem.
3. Optimise pricing and calibration methods.
4. Formulate numerically complex dynamic control problems in finance.
5. Construct flexible models for financial predictions.
6. Investigate textual data algorithmically.
7. Write and adapt advanced machine learning algorithms in Python code with Keras and/or Tensorflow for financial applications.

Indicative content:

This module seeks to introduce advanced machine learning concepts in the financial services industry, utilising neural networks, recurrent neural networks, reinforcement learning and text mining with applications ranging from credit scoring, fraud detection, market anomalies, trading engines to asset pricing. The module is based on a data-driven approach in which machine learning techniques are implemented using either simulated or real data. This module enables students to familiarise themselves with advanced machine learning techniques, to handle noisy datasets and extract value from them. This course will also introduce text mining methods for quantitatively analysing text data.

The following are indicative of the topics the module will typically cover:

General Introduction to Neural Networks and Deep Learning:

• History and background.
• Classification of Neural Networks.
• Types of Network Layers.
• Overview of applications.

Optimisation Algorithms:

• Loss functions.
• Regularisation.
• Stochastic Gradient Descent.
• Hessian matrices.
• Energy Landscapes.
• Generalisation Errors.
• Overfitting.

• Supervised Learning:
  • Convolutional Neural Networks.
  • Recurrent Neural Networks.
  • Long-Short Term Memory Model.
• Unsupervised Learning:
  • Principal Component Analysis.
  • Autoencoder.
  • Reinforcement Learning.

Data structures and Applications:

• Deep Neural Networks for Regression Problems.
• Graph Neural Networks.
• Text Mining.
• Time-Series Prediction.

Sample projects:

• Network Filtering and Anomaly Detection.
• Link Prediction in Temporal Networks.
• Autoencoders for Payment Network Analysis.
• Return Time-Series Prediction with Recurrent Neural Networks.
• Sentiment Analysis of Text Data.

Requisites:

To be eligible to select this module as an optional or elective, a student must: (1) be registered on a programme and year of study for which it is formally available; (2) understand basic levels of probability theory, linear algebra, and multivariate calculus; (3) be able to write a reasonably non-trivial computer program in a programming language suitable for numerical computing; and (4) have taken Machine Learning with Applications in Finance (COMP0050 or COMP0198).