Industry grapples with ways to collect, store, and process big data.
High-performance with MathWorks, MATLAB, and AutoML
Support for automated development of machine learning-based solutions


Many companies recognize the business value of big data, but lack of technical expertise and data accessibility/siloing issues prevent them from devising appropriate big data collection, storage, and processing strategies for their businesses.

On the 18th, MathWorks Korea held a two-part online seminar titled "Data Science and Machine Learning Applications Using MATLAB." In the first lecture, MathWorks Korea Manager Seong Ho-hyeon introduced high-speed big data processing using simple functions in a parallel/distributed computing architecture and how this approach can support applications.

“MathWorks supports the simplification and automation of big data processing and data feature extraction processes,” said Vice President Seong Ho-hyeon. “Through this, it presents solutions to the problems of lack of skilled technology and low data quality, which are obstacles to utilizing big data.”

In the second lecture, MathWorks Korea's Deputy Manager, Gyu-Hwan Jang, explained the pain points of the machine learning workflow and MathWorks' automated machine learning development solution to resolve them. Vice President Jang emphasized that even users without prior knowledge or experience with domain data or algorithms can develop high-performance machine learning systems.

Below is a summary of the first and second lectures of the seminar held that day.

◇ MATLAB, Generating Training Data for Machine Learning-Based Solutions

MATLAB provides a convenient big data processing environment for experts and engineers in specific industries. First, it supports cluster and Spark/Hadoop-based distributed data processing to support streaming applications through the 'Tall function' suitable for processing big data larger than the memory capacity of the big data processing platform.

Next, we implement a cluster-based parallel processing method through a 'distributed function' that enables high-speed in-memory operations, enabling smooth execution of mathematical and matrix operations that require a large amount of computing work.

Volkswagen adopted MATLAB to build the technological foundation for providing personalized car features and services to drivers, and successfully developed an individual driver recognition algorithm based on driving habit data.

MATLAB significantly reduces the time required for processing large-scale video and image data and extracting features, which account for approximately 70-80% of AI development time. Accelerate the processing of image data of increasing scale and complexity in industries such as medical, defense, and aerospace by implementing video streaming environments, processing large-scale image segmentation, supporting parallel computing and GPUs, analyzing pixels, and detecting and tracking objects.

For text data analysis, MathWorks offers the Text Analytics Toolbox. This toolbox analyzes the sentiments contained in human language and analyzes, classifies, and summarizes data, thereby enhancing the efficiency of natural language processing.

MathWorks helps customers develop maintenance applications that predict failure time and remaining useful life (RUL) of enterprise assets by leveraging their existing IoT and streaming data.

Additionally, when rapid environmental changes make it difficult to continuously collect sufficient amounts of data, it generates virtual error data. In particular, MathWorks supports the development of highly accurate prediction models that produce results most similar to data values measured in real environments through 'Simulink', which is equipped with an automatic tuning function.

Energy technology company Baker Hughes partnered with MathWorks to develop machine learning-based, highly accurate predictive maintenance software that can easily collect, categorize, and analyze terabytes of diverse sensor data.

Lacking data on shale oil extraction pump failures, Baker Hughes developed a model that mimics how the pumps operate and developed a machine learning algorithm that derives the RUL by virtually introducing errors into the model, 10 times faster than before.

◇ AutoML, Automating Industry-Specific Machine Learning Model Development

Machine learning is a technology that allows models, which derive outputs from input data, to learn and update based on new data. Currently, it's being used in various applications, including: speech recognition, object detection, and engine life prediction, all based on nonlinear input-output structures; stock market and weather forecasting based on dynamic programming based on real-time data fluctuations; and IoT, taxi fare systems, and flight delay analysis, all based on scalable big data.

Currently, there is a growing demand in many industries, including defense, electronics, and automotive, to perform a series of machine learning development processes on a single platform. To address this, MathWorks provides a toolbox that supports the entire machine learning workflow, from data access to modeling and system deployment. It also addresses pain points that exist at each stage of the machine learning workflow.

MathWorks' automated machine learning development solution, AutoML, automates the feature extraction process. Feature extraction involves generating (feature generation), selecting (feature selection), and transforming (feature transformation) key data features during data preparation so that machine learning models can learn from them.

AutoML allows for easy extraction of key features even with limited understanding of domain data. It prevents "model overfitting," which leads to suboptimal results due to modeling based on inappropriate data. It also improves "model interpretability," the interpretability of selected features, and reduces model size.

Additionally, AutoML automatically finds and recommends the optimal algorithm applicable to the characteristics of data prepared through automated machine learning model training.

Today, engineers are challenged by the sheer number of algorithms available, making it difficult to choose the right one. For example, classification and regression models often have over ten and six algorithms available, respectively. AutoML improves the efficiency and productivity of machine learning development by eliminating the time spent training individual algorithms.

Furthermore, by tuning hyperparameters, which are parameters that remain constant during training, models can be easily and quickly optimized. Furthermore, MATLAB Coder, Compiler, and Compiler SDK support code generation and deployment in C, C++, HDL, PLC, CUDA, Java, Python, and other languages for machine learning models, tailored to machine learning applications on embedded hardware and enterprise systems.