Search for: All records

As Cloud's adoption surges across industries, the limitations of its default scheduler, particularly on large scales or for jobs outside of its initial design scope, have become increasingly prominent. While the default schedulers in various cloud platforms were primarily engineered to focus on simple and predictable tasks, reinforcement learning (RL)-based schedulers are attracting attention as they can predict a larger and more diverse cloud environment. Nevertheless, there are practical constraints to the use of RL. Retraining for adaptation is necessary for each new environment, and exploration taken during each training may lead to unexpected performance degradation at runtime. To address these issues, this paper presents Dejavu which combines reinforcement learning with neural networks to learn and resolve scheduling problems more effectively. To tackle the extended training time and performance degradation by unexpected explorations, we apply pretraining using Demonstrations from existing heuristics. This guides the RL agent to explore in a safe and efficient manner. Furthermore, we design a robust reward function to push Dejavu to compete with and eventually outperform, the exploited heuristics and other baselines. The experimental results demonstrate the efficacy of Dejavu, showing remarkable improvements in key metrics. Compared to the default scheduler, it boosts resource utilization by 6 % and shortens scheduling time by 3% during the scheduling period. 

Abstract This paper gives an overview of Targeting Extremely Magnified Panchromatic Lensed Arcs and Their Extended Star formation (TEMPLATES), a JWST Early Release Science program that targeted four extremely bright, gravitationally lensed galaxies, two extremely dusty and two with low attenuation, as templates for galaxy evolution studies with JWST. TEMPLATES obtains a common set of spectral diagnostics for these 1.3 ≤z≤ 4.2 galaxies, in particular Hα, Paschenα, and the rest-frame optical and near-infrared continua. In addition, two of the four targets have JWST coverage of [Oiii] 5007 Å and Hβ; the other two targets have JWST coverage of polycyclic aromatic hydrocarbon 3.3μm and complementary Atacama Large Millimeter/submillimeter Array data covering the [Cii] 158μm emission line. The science goals of TEMPLATES are to demonstrate attenuation-robust diagnostics of star formation, map the distribution of star formation, compare the young and old stellar populations, and measure the physical conditions of star formation and their spatial variation across the galaxies. In addition, TEMPLATES has the technical goal to establish best practices for the integral field units within the NIRSpec and MIRI instruments, both in terms of observing strategy and in terms of data reduction. The paper describes TEMPLATES’s observing program, scientific and technical goals, data reduction methods, and deliverables, including high-level data products and data reduction cookbooks.