Comment from Drreddys Laboratories, Hyderabad, India
AnonymousSupportIndividual
Summary: The commenter provides a detailed technical breakdown of machine learning (ML), deep learning (DL), and generative AI (GenAI) to clarify their differences and applications in data science. They suggest specific guidelines for the draft guidance, including distinguishing between GMP and non-GMP operations and outlining a multi-step process for model development and evaluation.
Need to brief about ML and DL and their model building. Also need validation approach as per GAMP5, if we use AI prompt.
ML:
To Perform specific tasks or solve problems, such as data analysis, predictive future analytics.
Data analysis Past data analysis
Machine learning Prediction of future using past data (only numeric data)
Data analysis+ Machine learning Data science (only numeric data)
Data science Data analysis using ML, Power BI, Python, statistics (Descriptive and inferential) etc.,
Data types Structured data base such as Excel, CSV file, SQL, Oracle etc.,
Major models Decision tree, Random forest, Time series, LDA (Linear Discriminant Analysis)
and PCA(Principal Component Analysis). LDA and PCA using for dimensionality reduction.
DL/GEN AI:
To perform large amounts of data and can generate new content based on their patterns. Gen AI systems can create original content like images, audio, video, text, or software code. Gen AI is used in content creation and design, scientific research etc.,
Technology used in Gen.AI is Deep learning Advanced part of machine learning which deals mostly un-Structured data.
Deep learning is a powerful network using neural concept as like our brain neurons.
Data types Un-Structured data base such as MongoDB, images, audio, video, text, etc.,
Major models ANN (Artificial Neural Network), CNN(Convolutional Neural Network), GNN (Graph neural network),
LSTM(Long Short-Term Memory networks), RNN((Recurrent Neural Network),
NLP (Natural Language Processing), LLM (Large Language Model) using weight and bias.
DL+ Computer vision (Self driving car & lane assistance)
AI software name called as AGI(Artificial General Intelligence), it’s like machine/Roboot act as human brain.
Learning methodology:
Two types of learning (Supervised and unsupervised)
Supervised learning Input and output we know before deploy model.
Unsupervised learning output unknown.
Supervised Linear regression contains continuous variables and Logistic regression Classification(categorical) discrete variables, decision trees, random forest, PCA etc.,
Unsupervised Clustering and LDA
To plot any model need independent and dependent variable.
AI Process Flow:
Step-1 Understanding problem statement
Step-2 Data collection
Step-3 Data processing, structuring (Future selection, extraction) and cleaning using EDA (Exploratory Data Analysis)
Step-4 Explore & Visualize data by training and testing via google colab, python, power BI etc.,
Step-5 Apply ML/DL model on the data and verify performance
Step-6 Tuning the performance of ML/DL (Hyper tuning, optimization etc.,)
Step-7 Publish the results
Evaluate model performance:
In GMP critical operations, only the ML model at Biologics) is permitted; GenAI and LLMs are not allowed.
In non-GMP operations, Generative AI and Large Language Models may be used, where it will not impact on patient safety, product quality and data integrity.
Before implementing any models, personnel must be trained, documentation must be completed, and risk assessment must be conducted.
To evaluate model performance, acceptance criteria such as confusion matrix, sensitivity, accuracy, precision, and F1 score may be applied.