This paper discusses results and insights from the 1st ReQuEST workshop, a collective effort to promote reusability, portability and reproducibility of deep learning research artifacts within the Architecture/PL/Systems communities. ReQuEST (Reproducible Quality-Efficient Systems Tournament) exploits the open-source. Collective Knowledge framework (CK) to unify benchmarking, optimization, and co-design of deep learning systems implementations and exchange results via a live multi-objective scoreboard. Systems evaluated under ReQuEST are diverse and include an FPGA-based accelerator, optimized deep learning libraries for x86 and ARM systems, and distributed inference in Amazon Cloud and over a cluster of Raspberry Pis. We finally discuss limitations to our approach, and how we plan improve upon those limitations for the upcoming SysML artifact evaluation effort.
Evaluating the computational reproducibility of data analysis pipelines has become a critical issue. It is, however, a cumbersome process for analyses that involve data from large populations of subjects, due to their computational and storage requirements. We present a method to predict the computational reproducibility of data analysis pipelines in large population studies. We formulate the problem as a collaborative filtering process, with constraints on the construction of the training set. We propose 6 different strategies to build the training set, which we evaluate on 2 datasets, a synthetic one modeling a population with a growing number of subject types, and a real one obtained with neuroinformatics pipelines. Results show that one sampling method, "Random File Numbers (Uniform)" is able to predict computational reproducibility with a good accuracy. We also analyze the relevance of including file and subject biases in the collaborative filtering model. We conclude that the proposed method is able to speedup reproducibility evaluations substantially, with a reduced accuracy loss.
Neuroscience has been carried into the domain of big data and high performance computing (HPC) on the backs of initiatives in data collection and an increasingly compute-intensive tools. While managing HPC experiments requires considerable technical acumen, platforms and standards have been developed to ease this burden on scientists. While web-portals make resources widely accessible, data organizations such as the Brain Imaging Data Structure and tool description languages such as Boutiques provide researchers with a foothold to tackle these problems using their own datasets, pipelines, and environments. While these standards lower the barrier to adoption of HPC and cloud systems for neuroscience applications, they still require the consolidation of disparate domain-specific knowledge. We present Clowdr, a lightweight tool to launch experiments on HPC systems and clouds, record rich execution records, and enable the accessible sharing of experimental summaries and results. Clowdr uniquely sits between web platforms and bare-metal applications for experiment management by preserving the flexibility of do-it-yourself solutions while providing a low barrier for developing, deploying and disseminating neuroscientific analysis.
In scientometrics, we have not yet had an intensive debate about the reproducibility of research published in our field, although concerns about a lack of reproducibility have occasionally surfaced (see e.g. Glänzel & Schöpflin 1994 and Van den Besselaar et al. 2017), and the need to improve the reproducibility is used as an important argument for open citation data (see www.issi-society.org/open-citations-letter/). We initiated a first discussion about reproducibility in scientometrics with a workshop at ISSI 2017 in Wuhan. One of the outcomes was the sense that scientific fields differ with regard to the type and pervasiveness of threats to the reproducibility of their published research, last but not least due to their differences in modes of knowledge production, such as confirmatory versus exploratory study designs, and differences in methods and empirical objects.
Methodological developments and software implementations progress in increasingly faster time-frames. The introduction and widespread acceptance of pre-print archived reports and open-source software make state-of-the-art statistical methods readily accessible to researchers. At the same time, researchers more and more emphasize that their results should be reproducible (using the same data obtaining the same results), which is a basic requirement for assessing the replicability (obtaining similar results in new data) of results. While the age of fast-paced methodology greatly facilitates reproducibility, it also undermines it in ways not often realized by researchers. The goal of this paper is to make researchers aware of these caveats. I discuss sources of limited replicability and reproducibility in both the development of novel statistical methods and their implementation in software routines. Novel methodology comes with many researcher degrees of freedom, and new understanding comes with changing standards over time. In software-development, reproducibility may be impacted due to software developing and changing over time, a problem that is greatly magnified by large dependency-trees between software-packages. The paper concludes with a list of recommendations for both developers and users of new methods to improve reproducibility of results.
Scientific reproducibility is key to the advancement of science as researchers can build on sound and validated results to design new research studies. However, recent studies in biomedical research have highlighted key challenges in scientific reproducibility as more than 70% of researchers in a survey of more than 1500 participants were not able to reproduce results from other groups and 50% of researchers were not able to reproduce their own experiments. Provenance metadata is a key component of scientific reproducibility and as part of the Provenance for Clinical and Health Research (ProvCaRe) project, we have: (1) identified and modeled important provenance terms associated with a biomedical research study in the S3 model (formalized in the ProvCaRe ontology); (2) developed a new natural language processing (NLP) workflow to identify and extract provenance metadata from published articles describing biomedical research studies; and (3) developed the ProvCaRe knowledge repository to enable users to query and explore provenance of research studies using the S3 model. However, a key challenge in this project is the automated classification of provenance metadata extracted by the NLP workflow according to the S3 model and its subsequent querying in the ProvCaRe knowledge repository. In this paper, we describe the development and comparative evaluation of deep learning techniques for multi-class classification of structured provenance metadata extracted from biomedical literature using 12 different categories of provenance terms represented in the S3 model. We describe the application of the Long Term Short Memory (LSTM) network, which has the highest classification accuracy of 86% in our evaluation, to classify more than 48 million provenance triples in the ProvCaRe knowledge repository (available at: https://provcare.case.edu/).