INTRODUCING
Key to Precision and Efficiency in AI/ML Model Training
In the dynamic realm of machine learning (ML) and Artificial Intelligence (AI), GENIE emerges as a trailblazing tool by AvaWatz, crafted to redefine the approach to AI training. This comprehensive solution tackles the fundamental challenges in preparing datasets for machine learning, offering a suite of features that enhance efficiency, accuracy, reliability, and trustworthiness in ML model training.
Case Study 1
In this report, we provide detailed case-study of how GENIE has shown impact in reducing the labelling cost and time, and in create models with required performance in a very short amount of time.
Foreign Object Debris (FOD) Detection on Runways
Foreign Object Debris (FOD) on runways is a critical safety concern in aviation. Traditional methods for detecting and classifying such debris are often inefficient and labour-intensive. GENIE, an advanced data labelling and machine learning tool, was implemented to tackle this challenge effectively.
Challenge
The task was to develop a machine learning model capable of detecting and classifying various types of FOD on runways, including small particles like nails, metal scraps, pebbles, stones, coins, and tool parts, across 22 different classes. This required a substantial amount of accurately labelled data.
GENIE’s Approach
GENIE utilized a blend of intelligent seed set selection, synthetic data augmentation, active learning, and automatic labelling to enhance the data labelling process’s efficiency and accuracy. The following were the four functionalities of GENIE used in this project:
- Intelligent Seed Set Selection: GENIE started with a diverse and representative set of initial examples for training the model.
- Synthetic Data Augmentation: Focused on copy-paste augmentation to create realistic composite images by overlaying FOD items onto various runway backgrounds. This was particularly used for the rare classes of objects and using copy-paste augmentation, we increased the number of labeled examples for rare classes.
- Active Learning: GENIE continuously refined the model by focusing on labeling data points that were most challenging or uncertain.
- Automatic Labelling: Advanced algorithms were used for automatically labelling large portions of the dataset, reducing manual effort significantly.