⚙️

Data Intelligence Lab

The Data Intelligence Lab conducts a variety of artificial intelligence research that abstracts, interprets, predicts, and optimizes complex data. Artificial intelligence can extract core insights from complex data that cannot be interpreted by humans, help difficult-to-judge decisions, and provide data-based rationales. It also challenges human thinking, judgment and creative abilities, including playing Go and playing games like Starcraft. This also means that artificial intelligence can play a huge role in solving business problems in many fields, including manufacturing, service, and logistics. If the future can be accurately predicted using complex historical data, work efficiency can be greatly improved or cost can be significantly reduced.

For example, the vast amount of data collected from various sensors in the manufacturing process is time-consuming and expensive for humans to fully understand and analyze, and the accuracy is often low. It is virtually impossible for a human to accurately analyze both historical sales data or market economic data to predict demand or sales. However, the development of computing resources and the development of deep learning-based artificial intelligence technology made it possible to express these data well on the manifold, which became the basis for greatly improving anomaly detection and future prediction performance. In addition, based on past purchase history, it is now possible to predict a person's propensity and recommend products that fit his or her personal taste.

Artificial intelligence is also capable of more accurately performing complex optimizations that are difficult for humans to perform. PCB routing is a difficult problem in which you have to connect all the complex wires, but artificial intelligence can self-examine variously and suggest the most optimal route. Artificial intelligence algorithms such as Neural combinatorial optimization or Deep reinforcement learning can help you make the most optimal decision when operating a factory, making a purchase, or operating logistics.

In addition, deep learning-based Representation learning enabled artificial intelligence to better understand past music, and generative models enabled it to generate plausible music. These technologies have enabled artificial intelligence to perform creative activities such as composing.

The Data Intelligence Lab aims to create artificial intelligence that can make business decisions together by conducting challenging predictive and optimizing AI research with outstanding fellow researchers.

Anomaly Detection

Anomaly detection is a field related to checking whether the data stays in a normal range and whether an abnormal situation has occurred or how likely it can occur. It can be applied to a wide range of areas from abnormal product status to abnormal operating status of factory facility. Through the anomaly prediction, it can also prevent accidents in advance. We research how to increase the performance of anomaly detection by utilizing generative models such as Adversarial Autoencoder and GAN.

Prediction

Prediction is an area of study where AI algorithms are used to model correlation between input data and output data, which cannot be interpreted by humans due to its extreme complexity. Our research objective is to reach the level of prediction accuracy that traditional system identification algorithms could not reach by performing regression that is based on the latest Deep Learning techniques such as Variational Autoencoder and Adversarial Autoencode.

Time-Series Forecasting

Time series forecasting is an area of study where causal relationships according to time are modeled in order to make predictions for future phenomena. In order to predict product demand or cost of raw materials, key features need to be extracted from past patterns while considering change in the data related to the target variable. The latest Deep Learning techniques can be used to create complex models that could not be easily developed through conventional methods. Our research is about using AI technology to make more precise and sophisticated predictions for the future.

Optimization

Optimization is an area of study that enables computers to make accurate judgments about various complex situations and find the most appropriate actions by themselves. The focus of traditional optimization was finding an optimal solution based on mathematical modeling. However, it is actually very difficult to develop a model that is sophisticated enough to reflect the complex real world. In recent years, AI algorithms that perform modeling based on data and find optimal solutions have been receiving more attention. Through such technique, AI can now produce optimal actions at the level that was previously not possible, such as playing the board game Go. LG AI Research studies AI to optimize it to various real-world situations such as factories and SCM as well as the board game Go.

Recommendation

Recommendation means suggesting things that customers are likely to find interesting. Customers can save time by receiving suggestions without having to go through a complicated search process to find something they need or want. If they are satisfied with the recommended products, they continue to use the recommended service. In order to make recommendations, the program needs to automatically identify individual customers’ tendencies and classify customers with similar tendencies in order to make predictions on similar patterns of behavior. Conventionally, this was achieved by using various features that indicate customers. Recently, however, Deep Learning algorithms such as Variational Autoencoder have been used to use hidden features. LG AI Research studies Deep Learning-based recommendation algorithms that can be applied to various areas.

Data Creation

Data creation refers to the AI’s ability to learn from various data by itself and create similar data on its own. What people learn and apply something can be considered a process in which information is abstracted during the learning process and has got created in a whole new way. Similarly, AI can learn to abstract data through Representation learning and create similar yet new data through Generation. LG AI Research is researching AI algorithms that can be used to create various content, such as music composition, and that can help human beings develop new perspectives through future data creation according to the changes in conditions.

Unsupervised Representation Learning

In order to better process data that is difficult for humans to interpret or label intuitively, like time series data, one must be able to discover and create a method to represent the data themselves from the training data. In particular, as there are many cases that cannot be labeled, it is very important and difficult to perform Unsupervised learning. If there is a way to effectively discover hidden representations that can map such complex data onto a space that we can handle, we can use this to more accurately perform various tasks, such as making more accurate predictions or classifications. LG AI Research aims to make more accurate predictions and optimization through research to learn meaningful representations from complex data.

Causal AI

Being able to explain the cause of the prediction results is essential to utilize the prediction results more effectively, including anomaly detection, prediction, and time series prediction. Prediction results will be more reliable if AI can be used to explain causal relationships that go beyond simple correlation between inputs and outputs. In addition, opportunities to newly discover the root cause factors that people could not previously identify may arise, too. LG AI Research is conducting Causal AI research that can explain the causal relationships between various factors and outputs to utilize prediction results better and is researching AI that can help us make decisions to obtain the results we desire through causal models.

Meet our leader!

Woohyung Lim, Head of Data Intelligence Lab

“Data Intelligence Lab is working on the latest AI technologies for prediction and optimization. As calculators and computers are currently helping humans to do many things efficiently, AI that analyzes and predicts very complex data that is difficult for humans to interpret will help mankind solve many difficulties and change the world in the future.

We look forward to receiving a lot of support from many people who are willing to grow together through challenging research and create a social value through this .”