SK Shieldus said on Thursday that a paper by Lim Jeong-hoon (임정훈), a senior researcher at its cybersecurity AI research organisation, Cybersecurity AI Labs, was accepted by ICML 2026 (International Conference on Machine Learning).

The company said the research received high marks for taking a new approach to the problem of "irregular time series", a feature of real cyber attacks.

The company said existing cybersecurity detection techniques typically analyse data based on the assumption that attacks continue in a steady flow. It said real cyber attacks vary, with their timing and intervals not constant, sometimes concentrated in a short time or appearing over a long period. That makes it difficult for existing methods to fully reflect such changing patterns, resulting in missed signs of attack or lower detection accuracy, it said.

SK Shieldus presented a new AI technology called QuITE (Query-based Irregular Time-series Embedding) that can analyse an irregularly continuing attack flow as it is.

QuITE is an analysis method designed to effectively represent data with different time intervals so it can more naturally reflect real attack flows. It can also flexibly combine with existing AI models, giving it scalability that can be applied to various security detection systems.

In performance tests, QuITE showed an improvement of up to 45.9 percent over existing time-series analysis methods on a global public benchmark dataset, the company said.

Lim said, "In the AI academic community, how to effectively handle imperfect real-world environment data is being discussed as an important task." He said, "This research is meaningful in that it enabled existing AI models to learn irregular attack patterns more precisely."