AutoKE: Automated Knowledge Extraction

Today machine learning is one of the most relevant information technologies. Among modern methods, supervised learning shows the highest practical applicability. However, it has significant limitations, including

• need for large samples of labeled data
• need for an expert designing learning pipelines
• insufficient inerpretability of learning results

This prevents the widespreading of intellectual technologies and the advent of Industry 4.0.

---

AutoKE project is aimed at overcoming these limitations based on technologies:

• unsupervised learning
• automated machine learning
• ontology learning

Our mission is the creation of intelligent technologies of a new generation, capable of solving more complex problems than the capabilities of modern AI allow.

---

Our tasks:

• development of software products and systems that democratize intelligent technologies
• obtaining fundamental theoretical results in the field of Computer Science and Artificail Intelligence

---

Examples of our research:

• regularizing clustering models with adjusted entropy
• accelerating hyper-parameter optimization in clustering based on objective prior information criteria
• automated clustering model selection for ontology learning

---

Key papers:

• A Bayesian Information Criterion for Unsupervised Learning Based on an Objective Prior
• Entropy adjustment for model selection in automated clustering (preprint)
• OPIC: the Method of Machine Learning Automation Based on Objective Prior Information Criteria (preprint)

---

Contacts: Ildar Baimuratov, baimuratov.i@gmail.com