GEN AI

 GEN AI

Generative AI is a type of artificial intelligence that uses machine learning to create new patterns to create new content based on patterns and relationships in existing data. Neural networks are used to learn from data, and then use those learned patterns to generate new content.

Steps to create your own Gen AI model:

1. Defining your goal
2. Data collection
3. Data preparation and preprocessing
4. Choose the framework you would like to work with(e.g. Tensorflow, PyTorch, Keras)
5. Choose the model architecture(e.g. Transformers or GAN)
6. Training the model
7. Evaluate and refine

Objective: To develop an end-to-end gen AI model for medical summarization.

Input: MRI images of the brain

Output: Summarized medical report 

Here we already have a deep learning model based on MRI images of the brain trained on Parkinson's and Alzheimer's disease. To develop a gen AI model based on these we have to follow these steps:

1. Choosing the text summarization method. There are two approach 
• Extractive Summarization: Identifies and combines the most important sentences to create a summary.
• Abstractive Summarization: Generates a new, shorter version using the model's understanding of the text.
2. Selecting a pre-trained model
• Exploring pre-trained transformer models like BART (https://huggingface.co/facebook/bart-large-cnn) or T5 (https://huggingface.co/docs/transformers/en/model_doc/t5) specifically designed for text summarization tasks.
3. Preparing the data
• Collecting a dataset of text documents and their corresponding written summaries related to Alzheimer's and Parkinson's disease.
• Preprocessing the data: Clean the data. removing irrelevant information, and tokenizing the text using NLP.
4. Fine-tuning the pre-trained model
• Use libraries like Transformers to fine-tune the pre-trained model on your prepared dataset.
• Transformers link: https://huggingface.co/docs/transformers/en/index 
5. Evaluate the model's performance using metrics like the ROUGE score
6. Refine the model by adjusting hyperparameters or trying different pre-trained models based on the evaluation results.

Before developing the model some of the pre-requisite are:

1. NLP
2. Hugging face transformers
3. Text summarization
4. Fine-tuning models

Tutorials for these are available on Hugging Face:

• Hugging Face Transformers Course: https://huggingface.co/learn/nlp-course/chapter1/1
• Fine-tuning Text Summarization Models: https://huggingface.co/docs/transformers/en/training
• Text Summarization with Transformers Tutorial: https://huggingface.co/docs/transformers/en/tasks/summarization
• NLP Course: https://huggingface.co/learn/nlp-course/chapter1/1