Efficient GPU kernels are crucial for building performant machine learning architectures, but writing them is a time-consuming challenge that requires significant expertise; therefore, we explore using language models (LMs) to automate kernel generation. We introduce KernelBench, an open-source framework for evaluating LMs’ ability to write fast and correct kernels on a suite of 250 carefully selected PyTorch ML workloads. KernelBench represents a real-world engineering environment and making progress on the introduced benchmark directly translates to faster practical kernels. We introduce a new evaluation metric $\text{fast}_p$, which measures the percentage of generated kernels that are functionally correct and offer a speedup greater than an adjustable threshold $p$ over baseline. Our experiments across various state-of-the-art models and test-time methods show that frontier reasoning models perform the best out of the box but still fall short overall, matching the PyTorch baseline in less than 20% of the cases. While we show that results can improve by leveraging execution and profiling feedback during iterative refinement, KernelBench remains a challenging benchmark, with its difficulty increasing as we raise speedup threshold $p$.

Clinical trials depend upon secure and robust data collection. Often this data is longitudinal in nature and can include clinical information captured in clinical/hospital settings as well as patient reported information. This paper describes a web application and associated mobile applications (iPhone/Android) developed to support a clinical trial focused on the application of a commercially available tropical fruit-based ointment applied to a range of skin conditions including eczema and skin rashes, cracked/dry skin on heels, sunburn and insect bites: the LPR Study. The LPR study involved 138 participants in four cohorts with treatments over different time periods (10–21 days) depending on their associated skin condition. The paper describes the solution that was developed and the practical experiences and challenges that were faced and overcome in delivery of the underpinning platform.

When can $n$ given numbers be combined using arithmetic operators from a given subset of $\{+, -, \times, ÷\}$ to obtain a given target number? We study three variations of this problem of Arithmetic Expression Construction: when the expression (1) is unconstrained; (2) has a specified pattern of parentheses and operators (and only the numbers need to be assigned to blanks); or (3) must match a specified ordering of the numbers (but the operators and parenthesization are free). For each of these variants, and many of the subsets of $\{+,-,\times,÷\}$, we prove the problem NP-complete, sometimes in the weak sense and sometimes in the strong sense. Most of these proofs make use of a "rational function framework" which proves equivalence of these problems for values in rational functions with values in positive integers.

Rui Zhang, Caitlin Westerfield, Sungrok Shim, Garrett Bingham, Alexander Fabbri, William Hu, Neha Verma, Dragomir Radev. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. 2019.

Clinical research registries need to be driven by data quality to improve the outcome of clinical trials and to provide the possibility to facilitate new research initiatives. The International Niemann-Pick Disease Registry (INPDR) is one such example of a clinical research registry. Unlike other registries where data quality is largely based around best effort manual data entry, the INPDR registry supports ongoing data completeness evaluation and feedback to those entering data. To assess the efficiency of this feedback, a research framework evaluating data completeness with different types of user feedback was applied. In this paper we consider the impact on data quality and user activity over time. Data accuracy and data completeness was measured by participating centers and combined into an overall data quality score. The results show that clinical research registries can benefit from an ongoing data quality assurance framework with targeted data quality feedback. This has implications for future clinical trials and clinical databases more generally.

Collection of clinical information can be a fraught and intimidating process in hospital environments, especially for children with neurological conditions. Gaming and mobile technology offer many potential advantages for data collection where clinically important data can be collected as a side effect of playing games. Niemann-Pick Disease (NPD) is one example of a condition where gaming technology has the potential to greatly help patients and doctors alike. NPD is a severe, often fatal neurological condition where extensive physiological and cognitive information needs to be collected by doctors, typically in hospital environments. To help healthcare providers collect far richer information on NPD patients and reduce the burden/stress of patients and their families alike, a suite of targeted mobile games was developed to capture specific clinical information on NPD patients. These were evaluated with NPD patients and the feedback received was used to tailor the games accordingly. Personalised clinical information obtained through game playing was recorded, analysed and visualised through a flexible and extensible Cloud-based solution including capabilities for tracking patients longitudinally. This work has been directly aligned with the International Niemann-Pick Disease Registry. We suggest that this approach offers the potential for improved processes for data collection and importantly, a less stressful and more cost effective approach for collecting clinical information from and for patients.

BACKGROUND: Gene expression and transcription factor (TF) binding data have been used to reveal gene transcriptional regulatory networks. Existing knowledge of gene regulation can be presented using gene connectivity networks. However, these composite connectivity networks do not specify the range of biological conditions of the activity of each link in the network. RESULTS: We present a novel method that utilizes the expression and binding patterns of the neighboring nodes of each link in existing experimentally-based, literature-derived gene transcriptional regulatory networks and extend them in silico using TF-gene binding motifs and a compendium of large expression data from Saccharomyces cerevisiae. Using this method, we predict several hundreds of new transcriptional regulatory TF-gene links, along with experimental conditions in which known and predicted links become active. This approach unravels new links in the yeast gene transcriptional regulatory network by utilizing the known transcriptional regulatory interactions, and is particularly useful for breaking down the composite transcriptional regulatory network to condition specific networks. CONCLUSION: Our methods can facilitate future binding experiments, as they can considerably help focus on the TFs that must be surveyed to understand gene regulation.(Supplemental material and the latest version of the MATLAB implementation of the United Signature Algorithm is available online at 1 or [see Additional files 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]).