Introducing Foundational AI Concepts

Summarizing Neural Networks, Deep Learning, and Langauge Models are interconnected

Objective

After completing this lesson, you will be able to summarize how neural networks, deep learning, language models, and LLMs process complex data

Neural Networks and Deep Learning

One common approach to machine learning uses neural networks.

Neural networks are mathematical systems loosely inspired by the way neurons in the human brain are connected. They are made up of layers of small processing units that analyze information and pass results to the next layer. Each layer can learn to recognize different parts of a pattern.

For example, when analyzing an invoice image:

  • An early layer may detect lines, shapes, and characters.
  • Another layer may recognize words, numbers, and table structures.
  • A later layer may identify fields such as the supplier name, invoice number, amount, and due date.

When a neural network contains many layers, this is called deep learning. The word deep refers to the number of processing layers, not to the system having deep understanding.

Deep learning is especially useful for working with large amounts of complex or unstructured information, including:

  • Images
  • Speech
  • Documents
  • Natural language
  • Video
  • Large and complex datasets
Graphic comparing neural networks and deep learning

In an ERP environment, deep learning may support capabilities such as:

  • Extracting information from invoices and purchase orders
  • Recognizing objects or damage in images
  • Understanding spoken or written user requests
  • Identifying complex patterns in transactions
  • Predicting business outcomes from large datasets

Deep learning is also the foundation of most modern large language models and generative AI systems.

Language Models and Large Language Models (LLMs)

What is a Language Model?

A language model is trained on large amounts of text and learns patterns in how words, sentences, questions, and ideas fit together. A simple language model might predict the next word in a sentence.

For example:

Graphic listing what LMs are capable of

What are Large Language Models (LLMs)?

Deep learning made it possible to train much larger and more capable language models. These models can identify patterns not only in individual words, but across sentences, documents, topics, and different forms of business communication.

Large language models (LLMs) use this capability to interpret prompts and generate responses.

Graphic listing what LLMs can perform

In an ERP implementation, an LLM might be asked:

Explain the difference between a purchase requisition and a purchase order for a new business user.

It might produce a clear explanation based on the language patterns it learned during training.

It could also be asked:

Create a draft test script for a procure-to-pay scenario that includes purchase order creation, goods receipt, invoice posting, and payment.

The model generates the response one piece at a time by predicting what content is most likely to form a useful answer. This is more sophisticated than basic autocomplete, but the underlying principle is similar: the model predicts what should come next based on patterns in its training data and the context provided by the user.