The field of tabular data management and analysis is witnessing significant advancements, driven by the integration of large language models and graph-based techniques. Researchers are exploring innovative methods to infer taxonomy from tabular data, leveraging the semantic information contained within the data to improve schema inference and data exploration. Furthermore, there is a growing interest in developing efficient algorithms for computing QR and SVD decompositions over join matrices, with a focus on GPU implementations to achieve improved performance. The use of pre-trained language models is also being investigated for tasks such as number of distinct values estimation, aiming to reduce data access costs and improve estimation accuracy. Additionally, graph-table-RAG frameworks are being proposed to enhance cross-table question answering capabilities. Notable papers in this area include: PLM4NDV, which proposes a learned method incorporating pre-trained language models for number of distinct values estimation, reducing data access costs and providing accurate estimates. GTR, which introduces a graph-table-RAG framework for cross-table question answering, demonstrating superior performance and high deployment efficiency.