Towards Democratized Machine Learning: A Semantic Web Approach

The rapid growth of machine learning (ML) research has produced a vast and expanding collection of algorithms, datasets, and pipelines available on the Web. However, fragmented and dispersed documentation of these resources hampers accessibility, transparency, and effective use, posing challenges for users seeking to understand, adapt, and create ML pipelines. To address these challenges, we leverage Knowledge Graphs (KGs) and ontologies to represent ML pipelines as executable KGs (ExeKGs). This approach fosters an intuitive understanding of pipeline components and their relationships while defined constraints streamline the creation of valid and efficient pipelines. Furthermore, the structure of our KGs enables intelligent exploration and discovery of relevant ML artifacts, including pipelines and datasets. By incorporating KG-based ML techniques, we enhance the discovery and reuse of these artifacts. To consolidate these functionalities and provide users with an intuitive interface, we are developing ExeKGLab, a GUI-based platform for interacting with ExeKGs. This thesis explores the potential of KGs to democratize the ML landscape. We present our ongoing efforts to build a KG for ML, emphasizing its role in simplifying pipeline design, enhancing comprehension, and enabling smart exploration. By creating a structured and interconnected framework, our approach seeks to bridge gaps in accessibility and foster a more collaborative ML ecosystem. We invite discussion and feedback to advance this promising direction for future ML research.