Search

Computational fluid dynamics (CFD) model has been widely used for prototyping data centers. Evolving it to high-fidelity {\em digital twin} is desirable for the management and operations of large-scale data centers. Manually calibrating CFD model parameters to achieve twin-class fidelity by specially trained domain expert is tedious and labor-intensive. To reduce manual efforts, existing automatic calibration approaches developed for various computational models apply heuristics to search model configurations within an empirically defined parameter bound. However, in the context of CFD, each search step requires long-lasting CFD model's iterated solving, rendering these approaches impractical with increased model complexity. This paper presents Kalibre, a knowledge-based neural surrogate approach that performs CFD model calibration by iterating four key steps of i) training a neural surrogate model based on CFD-generated data, ii) finding the optimal parameters at the moment through neural surrogate retraining based on sensor-measured data, iii) configuring the found parameters back to the CFD model, and iv) validating the CFD model using sensor-measured data as the ground truth. Thus, the parameter search is offloaded to the neural surrogate which is ultra-faster than CFD model's iterated solving. To speed up the convergence of Kalibre, we integrate prior knowledge of the twinned data center's thermophysics into the neural surrogate design to improve its learning efficiency. With about five hours computation on a 32-core processor, Kalibre achieves mean absolute errors (MAEs) of $0.81^o$C and $0.75^o$C in calibrating two CFD models for two production data halls hosting thousands of servers each while requires fewer CFD solving processes than existing baseline approaches., Accepted to ACM BuildSys 2020

Sediment core ARC5-ICE6 from the Makarov Basin was used to reconstruct circum-Arctic glaciation and surface-ocean circulation history for the Middle to Late Pleistocene, ∼0.5–0.7 Ma. Sedimentary proxies include elemental composition, coarse grain content, clay mineralogy, and radiogenic isotopes (Sr-Nd-Pb). Earlier studied records from the central Arctic Ocean were compared with ICE6 for a broader paleoceanographic context. Results indicate predominant sediment delivery from the Eurasian shelves and glaciated mainland with intermittent glacial inputs from North America. Low proxy variability prior to Marine Isotope Stage (MIS) 6 (∼0.2 Ma) suggests persistent sediment transport by sea ice via the Transpolar Drift and limited glaciation at the Eurasian margin. Since MIS 6, sediment inputs and provenance exhibit large shifts between glacial- and interglacial-type environments. The overall northern Eurasian Ice Sheet (EAIS) expansion was possibly related to climatic cooling and growth of a perennial sea ice in the Arctic Ocean. Provenance indicators from MIS 6, 5b and 4/3 reflect EAIS expansions followed by deglacial meltwater discharge from west Siberia. This provenance indicates a strong cyclonic circulation in the Arctic Ocean characterized by a significantly eastward-extended Transpolar Drift. Different glaciogenic sourcing and attendant circulation are proposed for several events within MIS 5 and 3, where proxies indicate increased sediment inputs from North America rather than Eurasian ice sheets. Overall limited glacial influence inferred for MIS 3 is consistent with relatively high sea level reconstructed for this interstadial. The Last Glacial Maximum is barely identifiable in the ICE6 record due to low glacial inputs and/or depositional hiatus.

Sediment core ARC4–BN05 collected from the Canada Basin, Arctic Ocean, covers the late to middle Quaternary (Marine Isotope Stages (MIS) 1–15, ca. 0.5–0.6 Ma) as estimated by correlation to earlier proposed Arctic Ocean stratigraphies and AMS 14C dating of the youngest sediments. Detailed examination of clay and bulk mineralogy along with grain size, content of Ca and Mn, and planktonic foraminiferal numbers in core ARC4–BN05 provides important new information about sedimentary environments and provenance. We use increased contents of coarse debris as an indicator of glacier collapse events at the margins of the western Arctic Ocean, and identify the provenance of these events from mineralogical composition. Notably, peaks of dolomite debris, including large dropstones, track the Laurentide Ice Sheet (LIS) discharge events to the Arctic Ocean. Major LIS inputs occurred during the stratigraphic intervals estimated as MIS 3, intra-MIS 5 and 7 events, MIS 8, and MIS 10. Inputs from the East Siberian Ice Sheet (ESIS) are inferred from peaks of smectite, kaolinite, and chlorite associated with coarse sediment. Major ESIS sedimentary events occurred in the intervals estimated as MIS 4, MIS 6 and MIS 12. Differences in LIS vs. ESIS inputs can be explained by ice-sheet configurations at different sea levels, sediment delivery mechanisms (iceberg rafting, suspension plumes, and debris flows), and surface circulation. A long-term change in the pattern of sediment inputs, with an apparent step change near the estimated MIS 7/8 boundary (ca. 0.25 Ma), presumably indicates an overall glacial expansion at the western Arctic margins, especially in North America.

Model-based reinforcement learning (MBRL) has been proposed as a promising alternative solution to tackle the high sampling cost challenge in the canonical RL, by leveraging a system dynamics model to generate synthetic data for policy training purpose. The MBRL framework, nevertheless, is inherently limited by the convoluted process of jointly optimizing control policy, learning system dynamics, and sampling data from two sources controlled by complicated hyperparameters. As such, the training process involves overwhelmingly manual tuning and is prohibitively costly. In this research, we propose a "reinforcement on reinforcement" (RoR) architecture to decompose the convoluted tasks into two decoupled layers of RL. The inner layer is the canonical MBRL training process which is formulated as a Markov decision process, called training process environment (TPE). The outer layer serves as an RL agent, called intelligent trainer, to learn an optimal hyperparameter configuration for the inner TPE. This decomposition approach provides much-needed flexibility to implement different trainer designs, referred to "train the trainer." In our research, we propose and optimize two alternative trainer designs: 1) an unihead trainer and 2) a multihead trainer. Our proposed RoR framework is evaluated for five tasks in the OpenAI gym. Compared with three other baseline methods, our proposed intelligent trainer methods have a competitive performance in autotuning capability, with up to 56% expected sampling cost saving without knowing the best parameter configurations in advance. The proposed trainer framework can be easily extended to tasks that require costly hyperparameter tuning. Energy Market Authority (EMA) Info-communications Media Development Authority (IMDA) National Research Foundation (NRF) This work was supported in part by the Energy Program, Nation Research Foundation, Prime Minister’s Office, Singapore, administrated by the Energy Market Authority of Singapore, under Award NRF2017EWT-EP003-023, in part by the Green Data Centre Research administrated by the Info-communications Media Development Authority, under Award NRF2015ENC-GDCR01001-003, and in part by the Behavioral Studies in the Energy, Water, Waste and Transportation Sector under Award BSEWWT2017_2_06.

Image hashing is critical for large-scale image analytic-based applications, such as image retrieval. Although there have been dozens of hashing approaches, few of them take the hierarchical structure of the image categories into consideration. In this paper, we propose to incorporate the category taxonomy information in a deep reinforcement learning (DRL) model for image hashing. In particular, we learn an agent to predict the hashing codes sequentially under the DRL theme. Each coordinate of the hashing function can take the errors incurred by previous ones into consideration and hence more reliable hashing codes can be obtained than learning them independently. Besides, we design a novel level-specific reward function to gradually refine the hashing function according to the taxonomy information. Extensive experiments on two popular datasets demonstrate effectiveness of the proposed method.

We investigated the petrographic composition of coarse-grained clasts (>250 μm) in 117 core-top sediment samples from the western Arctic Ocean (the Amerasian Basin, Chukchi Sea, Alaskan/Beaufort margin and East Siberian Sea). Combined with analyses of bulk sediment mineralogy and grain size, we use the results to elucidate the terrigenous clastic sediment distribution patterns and transport modes in four different regions: (1) In the Chukchi Sea, the Bering Strait Inflow (BSI) contributes most of the sediments, with an additional terrigenous clastic contribution from Northern Alaska and Eastern Siberia. The highest coarse clastic content and a northward-fining of the grain-size distributions are attributed to the winnowing by the BSI. (2) In the Alaskan margin, terrigenous clasts are enriched in sea ice/anchor ice-transported quartz-rich sediments from Northern Alaska, and in addition the Alaska Coastal Current (ACC) contributes sediments originating from the Yukon River. Coarse-grained sediments accumulate near offshore Alaska with the finer fractions being transported further into the Chukchi Sea Shelf. (3) In the East Siberian Sea, coarse clasts are sparse and even absent in most study sites. Combined with the results of previous studies, we conclude that the terrigenous components are derived from the Laptev Sea and the adjacent Siberian Platform. They are transported by the Siberian Coastal Current and sea ice. (4) In the Canada Basin, terrigenous clasts are characterized by carbonate and quartz-rich sediments derived from the Canadian Arctic Archipelago (CAA). Sediments incorporated in sea ice/anchor ice drift from the CAA to the Beaufort Sea, where they combine with sediments from the Mackenzie River and are then transported into the Canada Basin by the Beaufort Gyre. Sediments and grain size in northern and southern Canada Basin also involve the early Holocene iceberg-entrained IRD from CAA and winnowing/nepheloid transported mica from marginal areas, respectively.

Sediment cores from the Northwind Ridge north of the Arctic Chukchi-Alaskan margin were analyzed for a suite of stratigraphic and provenance proxies. Based on the correlation to previously studied regional records, cores under study present sedimentary environments of Marine Isotope Stages (MIS) 1–11 (∼0.4 Ma) on millennial time scales. Contrasting glacial-interglacial conditions and provenance were characterized by means of clay-mineral and geochemical (elemental and isotopic) proxies along with grain size. Interglacial environments are interpreted as sea-ice dominated, with pre-Holocene sediment transported mostly from the East Siberian to Chukchi-Alaskan margin. In comparison with the recent conditions, this pattern suggests a stronger Transpolar Drift expanding into the Canada Basin at the expense of the Beaufort Gyre, or stronger along-shelf currents. Glacial intervals prior to MIS4 have a predominantly North American signature with carbonaceous or terrigenous source rocks. Carbonaceous inputs marked by Ca peaks are tracked to the western Canadian Arctic Archipelago eroded by the Laurentide Ice Sheet (LIS). Terrigenous sources (Zr peaks) are likely related to areas impacted by the Innuitian Ice Sheet and/or the LIS Mackenzie ice stream. MIS 4 and 2 intervals have a different composition with the Siberian/Chukchi rather than North American provenance. A stronger winnowing indicated by grain size in the glacial intervals suggests a stronger mid-depth circulation. Study results provide new boundary conditions for modeling paleocirculation and glaciations in the western Arctic Ocean.

Adaptive bitrate (ABR) streaming is the de facto solution for achieving smooth viewing experiences under unstable network conditions. However, most of the existing rate adaptation approaches for ABR are content-agnostic, without considering the semantic information of the video content. Nevertheless, semantic information largely determines the informativeness and interestingness of the video content, and consequently affects the QoE for video streaming. One common case is that the user may expect higher quality for the parts of video content that are more interesting or informative so as to reduce overall subjective quality loss. This creates two main challenges for such a problem: First, how to determine which parts of the video content are more interesting? Second, how to allocate bitrate budgets for different parts of the video content with different significances? To address these challenges, we propose a Content-of-Interest (CoI) based rate adaptation scheme for ABR. We first design a deep learning approach for recognizing the interestingness of the video content, and then design a Deep Q-Network (DQN) approach for rate adaptation by incorporating video interestingness information. The experimental results show that our method can recognize video interestingness precisely, and the bitrate allocation for ABR can be aligned with the interestingness of video content while not compromising the performances on objective QoE metrics. This project is partially funded by Microsoft Research Asia.

To analyze the distribution and sources of polycyclic aromatic hydrocarbons (PAHs) and evaluate their potential ecological risks, the concentrations of 16 PAHs were measured in 43 surface sediment samples from the Bering Sea and western Arctic Ocean. Total PAH (tPAH) concentrations ranged from 36.95 to 150.21 ng/g (dry weight). In descending order, the surface sediment tPAH concentrations were as follows: Canada Basin>northern Chukchi Sea>Chukchi Basin>southern Chukchi Sea>Aleutian Basin>Makarov Basin>Bering Sea shelf. The Bering Sea and western Arctic Ocean mainly received PAHs of pyrogenic origin due to pollution caused by the incomplete combustion of fossil fuels. The concentrations of PAHs in the sediments of the study areas did not exceed effects range low (ERL) values.