Hasil untuk "Modern"

R. Mittler, E. Blumwald

R. Baeza-Yates, B. Ribeiro-Neto

Contents Preface Acknowledgements 1 Introduction 2 User Interfaces for Search by Marti Hearst 3 Modeling 4 Retrieval Evaluation 5 Relevance Feedback and Query Expansion 6 Documents: Languages & Properties with Gonzalo Navarro and Nivio Ziviani 7 Queries: Languages & Properties with Gonzalo Navarro 8 Text Classification with Marcos Gonccalves 9 Indexing and Searching with Gonzalo Navarro 10 Parallel and Distributed IR with Eric Brown 11 Web Retrieval with Yoelle Maarek 12 Web Crawling with Carlos Castillo 13 Structured Text Retrieval with Mounia Lalmas 14 Multimedia Information Retrieval by Dulce Poncele'on and Malcolm Slaney 15 Enterprise Search by David Hawking 16 Library Systems by Edie Rasmussen 17 Digital Libraries by Marcos Gonccalves A Open Source Search Engines with Christian Middleton B Biographies Bibliography Index

R. Sutton, A. Barto

Leila Ahmed

Ian R. Macneil

Matthias Krings, A. Stone, R. Schmitz et al.

L. Ballentine

E. Yourdon

F. Glover, M. Laguna

E. A. Starke, J. T. Staley

Bullock, C. H. Fong, M. et al.

I. Borg, P. Groenen

A. Friedman

L. Grafakos

C. Campbell

J. Kooiman

J. Goodenough

This Account provides perspective on the evolution of the rechargeable battery and summarizes innovations in the development of these devices. Initially, I describe the components of a conventional rechargeable battery along with the engineering parameters that define the figures of merit for a single cell. In 1967, researchers discovered fast Na(+) conduction at 300 K in Na β,β''-alumina. Since then battery technology has evolved from a strongly acidic or alkaline aqueous electrolyte with protons as the working ion to an organic liquid-carbonate electrolyte with Li(+) as the working ion in a Li-ion battery. The invention of the sodium-sulfur and Zebra batteries stimulated consideration of framework structures as crystalline hosts for mobile guest alkali ions, and the jump in oil prices in the early 1970s prompted researchers to consider alternative room-temperature batteries with aprotic liquid electrolytes. With the existence of Li primary cells and ongoing research on the chemistry of reversible Li intercalation into layered chalcogenides, industry invested in the production of a Li/TiS2 rechargeable cell. However, on repeated recharge, dendrites grew across the electrolyte from the anode to the cathode, leading to dangerous short-circuits in the cell in the presence of the flammable organic liquid electrolyte. Because lowering the voltage of the anode would prevent cells with layered-chalcogenide cathodes from competing with cells that had an aqueous electrolyte, researchers quickly abandoned this effort. However, once it was realized that an oxide cathode could offer a larger voltage versus lithium, researchers considered the extraction of Li from the layered LiMO2 oxides with M = Co or Ni. These oxide cathodes were fabricated in a discharged state, and battery manufacturers could not conceive of assembling a cell with a discharged cathode. Meanwhile, exploration of Li intercalation into graphite showed that reversible Li insertion into carbon occurred without dendrite formation. The SONY corporation used the LiCoO2/carbon battery to power their initial cellular telephone and launched the wireless revolution. As researchers developed 3D transition-metal hosts, manufacturers introduced spinel and olivine hosts in the Lix[Mn2]O4 and LiFe(PO4) cathodes. However, current Li-ion batteries fall short of the desired specifications for electric-powered automobiles and the storage of electrical energy generated by wind and solar power. These demands are stimulating new strategies for electrochemical cells that can safely and affordably meet those challenges.

I. Dolgachev

Zhihao Shu, Md Musfiqur Rahman Sanim, Hangyu Zheng et al.

The increasing size and complexity of modern deep neural networks (DNNs) pose significant challenges for on-device inference on mobile GPUs, with limited memory and computational resources. Existing DNN acceleration frameworks primarily deploy a weight preloading strategy, where all model parameters are loaded into memory before execution on mobile GPUs. We posit that this approach is not adequate for modern DNN workloads that comprise very large model(s) and possibly execution of several distinct models in succession. In this work, we introduce FlashMem, a memory streaming framework designed to efficiently execute large-scale modern DNNs and multi-DNN workloads while minimizing memory consumption and reducing inference latency. Instead of fully preloading weights, FlashMem statically determines model loading schedules and dynamically streams them on demand, leveraging 2.5D texture memory to minimize data transformations and improve execution efficiency. Experimental results on 11 models demonstrate that FlashMem achieves 2.0x to 8.4x memory reduction and 1.7x to 75.0x speedup compared to existing frameworks, enabling efficient execution of large-scale models and multi-DNN support on resource-constrained mobile GPUs.

L. Raghavendar Raju, M. Venkata Krishna Reddy, Sridhar Reddy Surukanti et al.

Abstract Edge–cloud computing has emerged as an important paradigm for modern Internet of Things (IoT) workflow applications, enabling low latency and on-demand resource allocation. In scenarios with heterogeneous deadlines and varying workloads, SLA compliance requires efficient coordination between edge and cloud resources. However, cloud-centric scheduling and heuristic approaches tend to lack adaptability to rapidly changing system conditions and, as a result, experience long waiting times (the same applies to QoS). To tackle these issues, we present IntelliScheduler, a hybrid actor–critic deep reinforcement learning framework for adaptive task scheduling in an edge–cloud system. Our framework presents a runtime-aware state representation combined with a learning-based decision mechanism, backed by a multi-buffer experience replay architecture. Second, a learning-based optimal task scheduling (LbOTS) algorithm is developed to minimise total task execution delay by discovering optimal deployment decisions across edge and cloud computational resources using latency-aware reward modelling. We assess the proposed approach by conducting extensive simulation experiments under different workloads. We evaluate LbOTS across various experimental scenarios and report up to 13% higher normalised reward, 67% lower training loss, 52–66% lower operational cost, and 80–90% lower rejection rate compared to PSO, MBO, and MOPSObaselines, achieving approximately 15–75% better QoE. Though the current assessment is simulation-based, the adaptive learning formulation is highly relevant for application in dynamic edge–cloud scheduling scenarios.