Skip to main content
Version: Next

What is AI Contents

Updated 2025.03.19

What is AI Contents?​

AI Contents is a collection of resources built upon the rich experience and technical expertise of data scientists. By leveraging the AI Contents provided in Mellerikat, users can rapidly develop AI Solutions necessary for executing AI projects. The code architecture of AI Contents is structured to run directly within Mellerikat. By modifying experiment documents, users can tailor the solutions to their domain and utilize them within the platform. This minimizes trial and error and increases development productivity, accelerating the application of AI to real business problems.


mellerikat


Key Features​

AI Contents serves as the default AI Solution operating on AI Learning Organizer (ALO), and even novice analysts without deep AI knowledge can use it to develop deployable AI Solutions. Key highlights include:

Convenient AI Solution Development​

By editing the provided experiment plan, users can run various model experiments tailored to their data and domain. Without deep AI knowledge, users can still tune parameters and carry out effective experiments. AI Contents helps users produce high-quality, deployable AI models even without extensive expertise.

Rich Functionality Through Asset Combination​

AI Contents offers a variety of assets designed to solve challenges across different domains. These include sampling assets for imbalanced data, preprocessing assets that auto-handle complex data formats, and Graph Feature Engineering (GFE) assets for missing data or learning relationships across features. When default assets aren’t sufficient, users can combine additional assets or include custom ones for specific use cases.

Memory Efficiency and Optimized Training/Inference​

All AI Contents are optimized for memory efficiency and performance. For instance, in Vision Classification, the HRVI model uses attention mechanisms to efficiently process high-resolution images. Compared to MobileNetV1, it reduces parameters by 85%, cuts training time in half, and improves performance by 7%, making it ideal for high-resolution inspection tasks. For Graph-based Classification & Regression, Mellerikat provides efficient GNN models, using proprietary techniques to make them lightweight and production-ready.


User Scenario​

AI Contents can be used as follows:

  1. Select Content & Review Manual: Browse the available content and select one to use. Review its manual for usage instructions.

  2. Installation & Environment Setup: Install ALO and the selected content on your development environment (local PC, server, or cloud), and set up the virtual environment.

  3. Edit Experiment Plan: Update the Experimental_plan.yaml file with data paths, x_columns, and other key parameters.

  4. Run Experiments: Execute experiments via Jupyter Notebook for interactive work, or use a terminal if only reviewing results.

  5. Set Parameters: Define operational parameters (ui_args) and adjust ui_args_detail for detailed configuration.

  6. Register AI Solution: Following the registration guide, submit your experiment results and parameters to complete AI Solution development.


AI Contents List​

Anomaly Detection (AD)​

Learns normal patterns from continuously collected index or numerical data and detects anomalies in real time. Useful for equipment monitoring, early fault detection, and multivariate anomaly detection. Uses statistical point anomaly detection.

Forecasting (FCST)​

Predicts future trends and patterns by learning past time series data and external factors. Easily applies deep learning-based forecasting for domains like sales, inventory, and demand prediction. Currently supports the Nbeats algorithm.

Graph-powered Classification & Regression (GCR)​

Improves inference by modeling not only feature values but also their relationships as a graph. Ideal for datasets with high missing values or categorical variables. Based on PyTorch BigGraph and includes proprietary lightweight GNN and Graph XAI technologies.

Tabular Anomaly Detection (TAD)​

Detects anomalies in structured tabular data (e.g., spreadsheets, databases). Supports both numerical and categorical data using deep learning, machine learning, and statistical approaches. Applies to fraud detection, quality control, cybersecurity, and healthcare.

Tabular Classification & Regression (TCR)​

Solves classification and regression problems using tabular data. Automatically selects optimal parameters based on user configuration. Includes data filtering, preprocessing, model selection, and Explainable AI (XAI) features.

Vision Anomaly Detection (VAD)​

Identifies anomalies in image data without labeled training examples using models like FastFlow and PatchCore. Suitable for tasks such as defect detection and inspection automation.

Vision Classification (VC)​

Classifies pre-labeled image types using deep learning (e.g., normal/abnormal). Quickly trains models using TensorFlow V2. Supports lightweight MobileNetV1 and high-resolution HRVI models for tasks like visual inspection or defect classification.


AI Contents Access​

AI Contents is managed in a private Git repository.
To download the code, please request an access token from the Mellerikat manager.
Contact: contact@mellerikat.com

LGE users can check access details at the link below:
Access for LGE


Release Notes​

Check out the latest version updates for AI Contents:


Topics​