9607 research outputs found
Sort by
Flexible I/O for Database Management Systems with xNVMe
No full textToday, NVMe SSDs cover a diverse family of devices (e.g., Zoned Namespaces, Flexible Data Placement, and Key-Value SSDs) and offer high performance (microsecond-scale latency). To leverage the capabilities of these devices, a variety of I/O paths are available (e.g., libaio, io_uring, and SPDK). On the other hand, to avoid the challenges and unpredictability that comes with writing code to target such diversity, most data systems today still rely on the conventional filesystem APIs (POSIX) and synchronous IO. While (maybe) increasing programmer productivity, this choice leads to sub-optimal utilization of the modern NVMe storage.To unify the diverse I/O storage paths and make them more accessible to a wider-scale of programmers, Samsung built xNVMe that exposes a single message-passing API with minimal overhead. This paper takes the next step and integrates xNVMe into a state-of-the-art database system, DuckDB, by creating a new filesystem extension, nvmefs, that interacts with blocks on disk instead of files. We demonstrate that xNVMe integration allows DuckDB to utilize IO Passthru, SPDK, and Flexible Data Placement. Using these modern I/O methods, compared to DuckDB’s default sync I/O, nvmefs achieves either comparable performance for non-I/O-intensive cases or up to 50% lower query times on I/O-intensive queries
Turning the AI Microscope on the Brain
Full text linkThe dominant paradigm in cognitive neuroimaging uses trial-based and tightly controlled experiments, while subsequent analysis relies on mass-univariate linear models. This approach rests on faulty assumption about neural pro-cessing and brain organisation, that are now widely recognised. Artificial Neural Networks (ANN) are in comparison powerful computational models that excel at learning nonlinear associations between relevant features to solve optimise any given objective. In this thesis, we explore how ANNs can be used as scientific tools to study electroencephalography (EEG) signal. We argue that ANNs are particularly well suited to analyse EEG signal outside of traditional experimental setups, where high variability between trials is expected. Representation spaces in ANNs trained to decode cognitive states from EEG signal can then be scrutinised using explainability techniques. We dub this approach the "AI Microscope", using the sophisticated ANN decoders to see patterns a human analyst could not have seen.The empirical work in the thesis is presented through eight papers. They form a progression from traditional approaches such event-related potentials and band power analysis, to an advancement in experimental design using naturalistic stimuli and games, and conclude with work on the necessity and analysis of ANNs trained on EEG signal. We situate the AI microscope framework in the current neuroimaging and ANN literature as an alternative methodology for cognitive neuroscience, that can applied in circumstances where traditional methods fail. Throughout, we see that cognitive neuroscience and artificial intelligence research share many ideas and approaches, and such overlaps are highlighted in present work. The aim is not to replace traditional methods in neuroscience, but to complement them using the latest advancements from AI
