ArrayQL for Linear Algebra within Umbra

Schüle, Maximilian E.; Götz, Tobias; Kemper, Alfons; Neumann, Thomas

doi:10.1145/3468791.3468838

Cited by 4 publications

(2 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Thus a union of all weight matrices forms the base case for the recursion. Within the recursive step, nested CTEs help to evaluate the model (lines 8-19), to backpropagate the loss (lines 20-33) and to compute the derivative per weight matrix (lines [34][35][36][37][38][39][40][41][42][43][44][45]. The first CTE w -just referring to the original weights-is necessary, as PostgreSQL only allows one reference to the recursive table.…”

Section: Training In Sql-92mentioning

confidence: 99%

“…However, the use of an array as a nested data type interferes with the first normal form (referring to the definition of arrays as a non-atomic data type) and requires copying the data between operations. Instead, to process data in-place of CPU registers, we suggested an array backend for codegenerating database systems [43], which stores matrices in a relational representation (cf. Figure 1).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Versioning in Main-Memory Database Systems

Schüle

Karnowski

Schmeißer

et al. 2019

Proceedings of the 31st International Conference on Scientific and Statistical Database Management

View full text Add to dashboard Cite

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, 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 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 relational representation.

show abstract

Section: Training In Sql-92mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Versioning in Main-Memory Database Systems

Schüle

Karnowski

Schmeißer

et al. 2019

Proceedings of the 31st International Conference on Scientific and Statistical Database Management

View full text Add to dashboard Cite

show abstract

Recursive SQL and GPU-support for in-database machine learning

Schüle¹,

Lang²,

Springer³

et al. 2022

Distrib Parallel Databases

Self Cite

View full text Add to dashboard Cite

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.

show abstract

Higher-Order SQL Lambda Functions

Schüle,

Hornung

2024

2024 IEEE 40th International Conference on Data Engineering (ICDE)

View full text Add to dashboard Cite

ArrayQL for Linear Algebra within Umbra

Cited by 4 publications

References 6 publications

Versioning in Main-Memory Database Systems

Versioning in Main-Memory Database Systems

Recursive SQL and GPU-support for in-database machine learning

Higher-Order SQL Lambda Functions

Contact Info

Product

Resources

About