The Group-By/Aggregate operation stands as a pivotal element in query processing within database management systems. Commonly, this operation is executed through two primary methods: sorting the tuples and hashing the keys. While prior research has often favored hash-based implementations for GPU acceleration, sort-based approaches remain pertinent, particularly for datasets with high cardinality. Hash-based methods can become inefficient in managing memory under such conditions. In this study, we introduce a meticulously designed implementation based on radix-hashing, optimized specifically for GPUs, with a key focus on ensuring consistent performance, especially in scenarios involving high cardinalities. Empirical evaluations showcase that our design not only maintains performance consistency across varying levels of cardinality but also outperforms the current state-of-the-art hash-based GPU implementation by up to 10 times in terms of throughput. This performance improvement is most pronounced when dealing with high-cardinality datasets. Furthermore, our implementation’s performance is highly competitive with the state-of-the-art when processing datasets with low cardinality. These findings position our implementation as an excellent choice for executing the Group-By/Aggregate operation on high-cardinality datasets. Its consistent performance also makes it a robust candidate in scenarios where cardinality is uncertain or varies.

Anna Currey, Maria Nadejde, Raghavendra Reddy Pappagari, Mia Mayer, Stanislas Lauly, Xing Niu, Benjamin Hsu, Georgiana Dinu. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

Identifying related entities and events within and across documents is fundamental to natural language understanding.We present an approach to entity and event coreference resolution utilizing contrastive representation learning.Earlier state-of-the-art methods have formulated this problem as a binary classification problem and leveraged large transformers in a cross-encoder architecture to achieve their results.For large collections of documents and corresponding set of n mentions, the necessity of performing n 2 transformer computations in these earlier approaches can be computationally intensive.We show that it is possible to reduce this burden by applying contrastive learning techniques that only require n transformer computations at inference time.Our method achieves state-of-the-art results on a number of key metrics on the ECB+ corpus and is competitive on others.

MOTIVATION: Current methods used to analyze real-time quantitative polymerase chain reaction (qPCR) data exhibit systematic deviations from the assumed model over the progression of the reaction. Slight variations in the amount of the initial target molecule or in early amplifications are likely responsible for these deviations. Commonly used 4- and 5-parameter sigmoidal models appear to be particularly susceptible to this issue, often displaying patterns of autocorrelation in the residuals. The presence of this phenomenon, even for technical replicates, suggests that these parametric models may be misspecified. Specifically, they do not account for the sequential dependent nature of the amplification process that underlies qPCR fluorescence measurements. RESULTS: We demonstrate that a Smooth Transition Autoregressive (STAR) model addresses this limitation by explicitly modeling the dependence between cycles and the gradual transition between amplification regimes. In summary, application of a STAR model to qPCR amplification data improves model fit and reduces autocorrelation in the residuals. AVAILABILITY AND IMPLEMENTATION: R scripts to reproduce all the analyses and results described in this manuscript can be found at: https://github.com/bhsu4/GAPDH.SO. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.