<![CDATA[Deep learning is interpreted differently in educational and computational contexts. The purpose of this study is to systematically examine how deep learning is conceptualized and applied in mathematics education research, and to evaluate the extent to which these applications align with the Indonesian Ministry of Education and Culture’s (Kemendikbud) definition of deep learning. A systematic literature review (SLR) was conducted using PRISMA guidelines. Searches in ScienceDirect, Scopus, Springer Link, and ProQuest produced 1,881 records containing the term “deep learning,’” published between 2015 and 2025. After duplicate removal and screening, 16 peer-reviewed journal articles explicitly addressing deep learning in mathematics education were included. Two main interpretations were found. Eleven studies framed deep learning pedagogically, focusing on conceptual understanding, problem-solving, collaborative learning, and real-world application. Five studies adopted a computational framing, using deep neural networks and other machine learning techniques for prediction, error analysis, adaptive instruction, and automated feedback. While the majority aligned with Kemendikbud’s pedagogical emphasis, some studies treated deep learning purely as a technical method, without explicit links to student-centered outcomes. The review highlights a conceptual gap between pedagogical and computational uses of deep learning in mathematics education. Bridging this gap requires interdisciplinary collaboration between educators and technology developers to ensure technological applications support meaningful learning. The findings provide a reference for aligning global research on deep learning with national education policy, ensuring relevance for curriculum design and classroom practice. ]]>
