There have been many decades of work on optimizing query processing in database management systems. Recently, modern machine learning (ML), and specifically reinforcement learning (RL), has gained increased attention as a means to develop a query optimizer (QO). In this work, we take a closer look at two recent state-of-the-art (SOTA) RL-based QO methods to better understand their behavior. We find that these RL-based methods do not generalize as well as it seems at first glance. Thus, we ask a simple question:
- Award ID(s):
- 1663244
- NSF-PAR ID:
- 10168370
- Date Published:
- Journal Name:
- Mechanics research communications
- Volume:
- 103
- ISSN:
- 0093-6413
- Page Range / eLocation ID:
- 103469
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
How do SOTA RL-based QOs compare to a simple, modern, adaptive query processing approach? To answer this question, we choose two simple adaptive query processing techniques and implemented them in PostgreSQL. The first adapts an individual join operation on-the-fly and switches between a Nested Loop Join algorithm and a Hash Join algorithm to avoid sub-optimal join algorithm decisions. The second is a technique calledLookahead Information Passing (LIP), in which adaptive semijoin techniques are used to make a pipeline of join operations execute efficiently. To our surprise, we find that this simple adaptive query processing approach is not only competitive to the SOTA RL-based approaches but, in some cases, outperforms the RL-based approaches. The adaptive approach is also appealing because it does not require an expensive training step, and it is fully interpretable compared to the RL-based QO approaches. Further, the adaptive method works across complex query constructs that RL-based QO methods currently cannot optimize. -
Abstract We revisit the Haake–Lewenstein–Wilkens approach to Edwards–Anderson (EA) model of Ising spin glass (SG) (Haake
et al 1985Phys. Rev. Lett. 55 2606). This approach consists in evaluation and analysis of the probability distribution of configurations of two replicas of the system, averaged over quenched disorder. This probability distribution generates squares of thermal copies of spin variables from the two copies of the systems, averaged over disorder, that is the terms that enter the standard definition of the original EA order parameter, . We use saddle point/steepest descent (SPSD) method to calculate the average of the Gaussian disorder in higher dimensions. This approximate result suggest that at in 3D and 4D. The case of 2D seems to be a little more subtle, since in the present approach energy increase for a domain wall competes with boundary/edge effects more strongly in 2D; still our approach predicts SG order at sufficiently low temperature. We speculate, how these predictions confirm/contradict widely spread opinions that: (i) There exist only one (up to the spin flip) ground state in EA model in 2D, 3D and 4D; (ii) there is (no) SG transition in 3D and 4D (2D). This paper is dedicated to the memories of Fritz Haake and Marek Cieplak. -
Abstract The Social Approach Task is commonly used to identify sociability deficits when modeling liability factors for autism spectrum disorder (ASD) in mice. It was developed to expand upon existing assays to examine distinct aspects of social behavior in rodents and has become a standard component of mouse ASD‐relevant phenotyping pipelines. However, there is variability in the statistical analysis and interpretation of results from this task. A common analytical approach is to conduct within‐group comparisons only, and then interpret a difference in
significance levels as if it were a group difference, without any direct comparison. As an efficient shorthand, we named this approach EWOCs: Erroneous Within‐group Only Comparisons. Here, we examined the prevalence of EWOCs and used simulations to test whether this approach could produce misleading inferences. Our review of Social Approach studies of high‐confidence ASD genes revealed 45% of papers sampled used only this analytical approach. Through simulations, we then demonstrate how a lack of significant difference within one group often does not correspond to a significant difference between groups, and show this erroneous interpretation increases the rate of false positives up to 25%. Finally, we define a simple solution: use an index, like a social preference score, with direct statistical comparisons between groups to identify significant differences. We also provide power calculations to guide sample size in future studies. Overall, elimination of EWOCs and adoption of direct comparisons should result in more accurate, reliable, and reproducible data interpretations from the Social Approach Task across ASD liability models. . © 2019 International Society for Autism Research, Wiley Periodicals, Inc.Autism Res 2019, 12: 1171–1183Lay Summary The Social Approach Task is widely used to assess social behavior in mice and is frequently used in studies modeling autism. However, reviewing published studies showed nearly half do not use correct comparisons to interpret these data. Using simulated and original data, we argue the correct statistical approach is a direct comparison of scores between groups. This simple solution should reduce false positives and improve consistency of results across studies.
-
Abstract Assemblies of one-dimensional filaments appear in a wide range of physical systems: from biopolymer bundles, columnar liquid crystals, and superconductor vortex arrays; to familiar macroscopic materials, like ropes, cables, and textiles. Interactions between the constituent filaments in such systems are most sensitive to the
distance of closest approach between the central curves which approximate their configuration, subjecting these distinct assemblies to common geometric constraints. In this paper, we consider two distinct notions of constant spacing in multi-filament packings in :equidistance , where the distance of closest approach is constant along the length of filament pairs; andisometry , where the distances of closest approach between all neighboring filaments are constant and equal. We show that, although any smooth curve in permits one dimensional families of collinear equidistant curves belonging to a ruled surface, there are only two families of tangent fields with mutually equidistant integral curves in . The relative shapes and configurations of curves in these families are highly constrained: they must be either (isometric) developable domains, which can bend, but not twist; or (non-isometric) constant-pitch helical bundles, which can twist, but not bend. Thus, filament textures that are simultaneously bent and twisted, such as twisted toroids of condensed DNA plasmids or wire ropes, are doubly frustrated: twist frustrates constant neighbor spacing in the cross-section, while non-equidistance requires additional longitudinal variations of spacing along the filaments. To illustrate the consequences of the failure of equidistance, we compare spacing in three ‘almost equidistant’ ansatzes for twisted toroidal bundles and use our formulation of equidistance to construct upper bounds on the growth of longitudinal variations of spacing with bundle thickness. -
null (Ed.)We consider recognition and reconfiguration of lattice-based cellular structures by very simple robots with only basic functionality. The underlying motivation is the construction and modification of space facilities of enormous dimensions, where the combination of new materials with extremely simple robots promises structures of previously unthinkable size and flexibility; this is also closely related to the newly emerging field of programmable matter. Aiming for large-scale scalability, both in terms of the number of the cellular components of a structure, as well as the number of robots that are being deployed for construction requires simple yet robust robots and mechanisms, while also dealing with various basic constraints, such as connectivity of a structure during reconfiguration. To this end, we propose an approach that combines ultra-light, cellular building materials with extremely simple robots. We develop basic algorithmic methods that are able to detect and reconfigure arbitrary cellular structures, based on robots that have only constant-sized memory. As a proof of concept, we demonstrate the feasibility of this approach for specific cellular materials and robots that have been developed at NASA.more » « less