Apple Study Explores How AI Predicts To Improve Bugs

The study of AI developed by Apple (photo: dock. unsplash)

JAKARTA Apple recently published three interesting research studies on how Artificial Intelligence (AI) technology can improve workflow, quality, and software productivity.

This study was conducted by a number of Apple researchers. In the study, Apple provides important knowledge of how sophisticated AI can work. Here are the details of the results of the study, according to 9to5mac.

Predicting Defects On Software

The first study was carried out on a new AI model named ADE-QVAET designed to predict software or bug defects. The model was specially designed to overcome the limitations of the Large Language Model (LLM) in analyzing large-scale code bases.

The ADE-QVAET model combines four AI techniques to improve the accuracy of bug predictions. The techniques used include Adaptive Differential Evolution, Quantum Variational Autoencoder, Transformer layer, and Adaptive Noise Reduction and Augmentation.

This LM does not analyze the code directly, but rather analyzes the metric and data about the code. In a test, ADE-QVAET achieved high accuracy, precision, recall, and F1 scores, each exceeding 98 percent compared to traditional models.

Software Testing Automatically

The second study discusses how AI can simplify the time of engineers by automating the entire testing process. The researchers developed a system that utilizes LLM and Autonomous AI Agents.

The system is designed to produce and manage testing artifacts, from plans to reports of validation, automatically. The results of this study are considered promising with an increase in testing accuracy from 65 percent to 94.8 percent.

Agent Training for Bug Repairs

The third study is based on SWE-Gym, an AI model designed to train AI agents to fix bugs independently. This agent is trained to read, edit, and verify real codes.

SWE-Gym was built using 2,438 real-world Python tasks from open source repositories. Agents trained using SWE-Gym managed to complete 72.5 percent of the tasks correctly beyond the previous benchmark of more than 20 percentage points.