TY - JOUR TI - Editorial AU - Schüle, Maximilian E. AU - Klettke, Meike AU - Störl, Uta PY - 2026 SP - 183-186 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/115103 LA - deu ER - TY - CPAPER TI - Back to the Roots : Decision Trees within Modern Database Systems AU - Wolfschmidt, Lena Renate AU - Schüle, Maximilian Emanuel PY - 2026 SP - 687-695 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/115101 LA - eng AB - HUK-COBURG relies on decision trees to classify e-mails into topics. As the data is stored within relational database systems, this paper investigates the performance improvement of in-database evaluation of decision trees. The paper first summarises the theory behind binary decision trees, then describes the implementation as SQL case-when statements. The evaluation shows the benefit of the in-database application of decision trees by eliminating expensive extraction-transform-load time. In the future, we intend to optimise the training as well as inference of decision trees in SQL to fully eliminate the need for data extraction. ER - TY - CPAPER TI - DuoLingo-AutoDiff : In-Database Automatic Differentiation with MLIR AU - Gutjahr, Kevin AU - Ruck, Clemens AU - Schüle, Maximilian E. PY - 2025 SP - 1-8 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/111827 LA - eng AB - Forward and reverse mode automatic differentiation evaluate the gradient of a model function efficiently by caching the results of partial derivatives. Just-in-time compilation improves the runtime of automatic differentiation by eliminating function calls and storing partial derivatives in virtual registers. This paper discusses the first open-source implementation of automatic differentiation with MLIR and LingoDB. The evaluation compares optimizations applied to forward and reverse modes. It showed that sub-expressions, that appear frequently within the calculation, will be reused after MLIR performs its optimization. Additionally, reverse mode outperforms forward mode due to less generated code. KW - Automatic Diferentiation KW - In-Database Machine Learning KW - Query Compilation KW - MLIR ER - TY - CPAPER TI - DuoLingo-AutoDiff : In-Database Automatic Differentiation with MLIR AU - Gutjahr, Kevin AU - Ruck, Clemens AU - Schüle, Maximilian E. PY - 2025 T2 - DEEM '25: Proceedings of the Workshop on Data Management for End-to-End Machine Learning SP - 1-8 PB - ACM CY - New York SN - 979-8-4007-1924-0 DO - 10.1145/3735654.3735943 UR - https://fis.uni-bamberg.de/handle/uniba/111824 LA - eng AB - Forward and reverse mode automatic differentiation evaluate the gradient of a model function efficiently by caching the results of partial derivatives. Just-in-time compilation improves the runtime of automatic differentiation by eliminating function calls and storing partial derivatives in virtual registers. This paper discusses the first open-source implementation of automatic differentiation with MLIR and LingoDB. The evaluation compares optimizations applied to forward and reverse modes. It showed that sub-expressions, that appear frequently within the calculation, will be reused after MLIR performs its optimization. Additionally, reverse mode outperforms forward mode due to less generated code. KW - Automatic Diferentiation KW - In-Database Machine Learning KW - Query Compilation KW - MLIR ER - TY - JOUR TI - The Duck’s Brain : Training and Inference of Neural Networks within Database Engines AU - Schüle, Maximilian AU - Neumann, Thomas AU - Kemper, Alfons PY - 2025 SP - 209-221 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/107115 LA - eng AB - Although database systems perform well in data access and manipulation, their relational model hinders data scientists from formulating machine learning algorithms in SQL. Nevertheless, we argue that modern database systems perform well for machine learning algorithms expressed in relational algebra. To overcome the barrier of the relational model, this paper shows how to transform data into a coordinate relational representation for training neural networks in SQL: We first describe building blocks for data transformation, model training and inference in SQL-92 and their counterparts using an extended array data type. Then, we compare the implementation for model training and inference using array data types to the one using a coordinate relational representation in SQL-92 only. The evaluation in terms of runtime and memory consumption proves the suitability of modern database systems for matrix algebra, although specialised array data types perform better than matrices in coordinate relational representation. KW - SQL-92 KW - Neural Networks KW - Automatic Differentiation KW - In-Memory Database Systems KW - Machine Learning ER - TY - CPAPER TI - Back to the Roots : Decision Trees within Modern Database Systems AU - Wolfschmidt, Lena Renate AU - Schüle, Maximilian Emanuel PY - 2025 T2 - Datenbanksysteme für Business, Technologie und Web: Bamberg, 3.-7. März 2025 : Proceedings SP - 687-695 PB - Gesellschaft für Informatik CY - Bonn DO - 10.18420/btw2025-37 UR - https://fis.uni-bamberg.de/handle/uniba/106799 LA - eng AB - HUK-COBURG relies on decision trees to classify e-mails into topics. As the data is stored within relational database systems, this paper investigates the performance improvement of in-database evaluation of decision trees. The paper first summarises the theory behind binary decision trees, then describes the implementation as SQL case-when statements. The evaluation shows the benefit of the in-database application of decision trees by eliminating expensive extraction-transform-load time. In the future, we intend to optimise the training as well as inference of decision trees in SQL to fully eliminate the need for data extraction. ER - TY - JOUR TI - Editorial AU - Schüle, Maximilian E. AU - Klettke, Meike AU - Störl, Uta PY - 2024 JO - Datenbank-Spektrum VL - 24 IS - 3 SP - 183-186 PB - Springer Science and Business Media LLC SN - 1610-1995 DO - 10.1007/s13222-024-00493-2 UR - https://fis.uni-bamberg.de/handle/uniba/106796 LA - deu ER - TY - JOUR TI - The Duck’s Brain : Training and Inference of Neural Networks within Database Engines AU - Schüle, Maximilian AU - Neumann, Thomas AU - Kemper, Alfons PY - 2024 JO - Datenbank-Spektrum VL - 24 IS - 3 SP - 209-221 PB - Springer CY - Berlin ; Heidelberg SN - 1610-1995 DO - 10.1007/s13222-024-00485-2 UR - https://fis.uni-bamberg.de/handle/uniba/106797 LA - eng AB - Although database systems perform well in data access and manipulation, their relational model hinders data scientists from formulating machine learning algorithms in SQL. Nevertheless, we argue that modern database systems perform well for machine learning algorithms expressed in relational algebra. To overcome the barrier of the relational model, this paper shows how to transform data into a coordinate relational representation for training neural networks in SQL: We first describe building blocks for data transformation, model training and inference in SQL-92 and their counterparts using an extended array data type. Then, we compare the implementation for model training and inference using array data types to the one using a coordinate relational representation in SQL-92 only. The evaluation in terms of runtime and memory consumption proves the suitability of modern database systems for matrix algebra, although specialised array data types perform better than matrices in coordinate relational representation. KW - SQL-92 KW - Neural Networks KW - Automatic Differentiation KW - In-Memory Database Systems KW - Machine Learning ER - TY - CPAPER TI - Give a JIT on GPUs : NVRTC for Code-Generating Database Systems AU - Sachnov, Anton AU - Merzljak, Leonard von AU - Schüle, Maximilian E. PY - 2024 T2 - 2024 IEEE 40th International Conference on Data Engineering Workshops (ICDEW) SP - 384-387 PB - IEEE DO - 10.1109/icdew61823.2024.00061 UR - https://fis.uni-bamberg.de/handle/uniba/106798 LA - eng ER - TY - CPAPER TI - Higher-Order SQL Lambda Functions AU - Schüle, Maximilian E. AU - Hornung, Jakob PY - 2024 T2 - 2024 IEEE 40th International Conference on Data Engineering (ICDE) SP - 5622-5628 PB - IEEE DO - 10.1109/icde60146.2024.00450 UR - https://fis.uni-bamberg.de/handle/uniba/106800 LA - eng ER - TY - CPAPER TI - Teaching Blue Elephants the Maths for Machine Learning AU - Ruck, Clemens AU - Schüle, Maximilian Emanuel PY - 2023 T2 - DEEM '23: Proceedings of the Seventh Workshop on Data Management for End-to-End Machine Learning SP - 1-4 PB - ACM SN - 979-8-4007-0204-4 DO - 10.1145/3595360.3595852 UR - https://fis.uni-bamberg.de/handle/uniba/94682 LA - eng ER - TY - CPAPER TI - Recursive SQL and GPU-Support for In-Database Machine Learning AU - Schüle, Maximilian PY - 2023 SP - 931 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/59173 LA - eng AB - But to retrieve the latest data from a database, time-consuming extraction is necessary as database systems have rarely been used for operations such as matrix algebra and gradient descent.In this work, we demonstrate that SQL with recursive tables makes it possible to express a complete machine learning pipeline out of data preprocessing, model training and its validation.To facilitate the specification of loss functions, we extend the code-generating database system Umbra by an operator for automatic differentiation for use within recursive tables:With the loss function expressed in SQL as a lambda function, Umbra generates machine code for each partial derivative.We further use automatic differentiation for a dedicated gradient descent operator, which generates LLVM code to train a user-specified model on GPUs.We fine-tune GPU kernels at hardware level to allow a higher throughput and propose non-blocking synchronisation of multiple units.In our evaluation, automatic differentiation accelerated the runtime by the number of cached subexpressions compared to compiling each derivative separately.Our GPU kernels with independent models allowed maximal throughput even for small batch sizes, making machine learning pipelines within SQL more competitive. KW - SQL KW - Machine Learning ER - TY - CPAPER TI - NN2SQL : Let SQL Think for Neural Networks AU - Schüle, Maximilian AU - Kemper, Alfons AU - Neumann, Thomas PY - 2023 SP - 183-194 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/59172 LA - eng AB - Although database systems perform well in data access and manipulation, their relational model hinders data scientists from formulating machine learning algorithms in SQL. Nevertheless, we argue that modern database systems perform well for machine learning algorithms expressed in relational algebra. To overcome the barrier of the relational model, this paper shows how to transform data into a relational representation for training neural networks in SQL: We first describe building blocks for data transformation in SQL. Then, we compare an implementation for model training using array data types to the one using a relational representation in SQL-92 only. The evaluation proves the suitability of modern database systems for matrix algebra, although specialised array data types perform better than matrices in relational representation. KW - SQL-92 KW - Neural Networks KW - Automatic Differentiation ER - TY - CPAPER TI - Recursive SQL and GPU-Support for In-Database Machine Learning AU - Schüle, Maximilian PY - 2023 T2 - Lecture Notes in Informatics (LNI) : Proceedings ; P-331 SP - 931 PB - Gesellschaft für Informatik e.V. CY - Bonn SN - 978-3-88579-725-8 DO - 10.18420/BTW2023-62 UR - https://fis.uni-bamberg.de/handle/uniba/58982 LA - eng AB - In machine learning, continuously retraining a model guarantees accurate predictions based on the latest data as training input. But to retrieve the latest data from a database, time-consuming extraction is necessary as database systems have rarely been used for operations such as matrix algebra and gradient descent.In this work, we demonstrate that SQL with recursive tables makes it possible to express a complete machine learning pipeline out of data preprocessing, model training and its validation.To facilitate the specification of loss functions, we extend the code-generating database system Umbra by an operator for automatic differentiation for use within recursive tables:With the loss function expressed in SQL as a lambda function, Umbra generates machine code for each partial derivative.We further use automatic differentiation for a dedicated gradient descent operator, which generates LLVM code to train a user-specified model on GPUs.We fine-tune GPU kernels at hardware level to allow a higher throughput and propose non-blocking synchronisation of multiple units.In our evaluation, automatic differentiation accelerated the runtime by the number of cached subexpressions compared to compiling each derivative separately.Our GPU kernels with independent models allowed maximal throughput even for small batch sizes, making machine learning pipelines within SQL more competitive. KW - SQL KW - Machine Learning ER - TY - CPAPER TI - NN2SQL : Let SQL Think for Neural Networks AU - Schüle, Maximilian AU - Kemper, Alfons AU - Neumann, Thomas PY - 2023 T2 - Lecture Notes in Informatics (LNI) : Proceedings ; P-331 SP - 183-194 PB - Gesellschaft für Informatik e.V. CY - Bonn SN - 978-3-88579-725-8 DO - 10.18420/BTW2023-09 UR - https://fis.uni-bamberg.de/handle/uniba/58981 LA - eng AB - Although database systems perform well in data access and manipulation, their relational model hinders data scientists from formulating machine learning algorithms in SQL. Nevertheless, we argue that modern database systems perform well for machine learning algorithms expressed in relational algebra. To overcome the barrier of the relational model, this paper shows how to transform data into a relational representation for training neural networks in SQL: We first describe building blocks for data transformation in SQL. Then, we compare an implementation for model training using array data types to the one using a relational representation in SQL-92 only. The evaluation proves the suitability of modern database systems for matrix algebra, although specialised array data types perform better than matrices in relational representation. KW - SQL-92 KW - Neural Networks KW - Automatic Differentiation ER - TY - CPAPER TI - Blue Elephants Inspecting Pandas: Inspection and Execution of Machine Learning Pipelines in SQL AU - Schüle, Maximilian E. AU - Scalerandi, Luca AU - Kemper, Alfons AU - Neumann, Thomas PY - 2023 PB - OpenProceedings.org DO - 10.48786/EDBT.2023.04 UR - https://fis.uni-bamberg.de/handle/uniba/58983 LA - eng ER -