Learning Objectives:

  • Explain the architecture of neural networks.

  • Understand deep learning applications in banking.

  • Explain transformers and large language models.

5.1 Neural Network Fundamentals

Neural networks are the foundation of deep learning. The RPTPU course introduces:

  • Single Neuron: The basic building block.

  • Single Hidden Layer Network: One layer between input and output.

  • Multi-Layer Perceptron (MLP): Multiple hidden layers.

  • Backpropagation: The algorithm for training neural networks .

The HKU SPACE course covers the basics of neural networks and deep learning architectures . Neural networks are able to model complex non-linear relationships that traditional statistical models may miss.

5.2 Deep Learning Architectures for Finance

  • Feedforward Neural Networks: Used for tabular data in credit scoring and fraud detection.

  • Convolutional Neural Networks (CNNs): Used for image-based document processing.

  • Recurrent Neural Networks (RNNs): Used for time series forecasting.

  • Transformers: The architecture behind large language models, increasingly used in finance for document processing.

The ETH Zürich course covers neural networks, including stochastic gradient descent, autoencoders, graph neural networks, and transformers . Transformers are a key technology in Generative AI.

5.3 Practical Implementation

The RPTPU course includes practical sessions on building neural networks using TensorFlow and Keras, covering advanced architectures, backpropagation, optimization techniques, and hyperparameter tuning . The University of Warsaw course includes deep learning and AI concepts in its curriculum .

 

PreviousNext