Hasil untuk "Architecture"

B. Razavi

T. Stahl, M. Völter, K. Czarnecki

P. Varaiya

Fariba Afrin Irany, Sampson Akwafuo

The increasing availability of unstructured clinical narratives in electronic health records (EHRs) has created new opportunities for automated disease characterization, cohort identification, and clinical decision support. However, modeling long, domain-specific clinical text remains challenging due to limited labeled data, severe class imbalance, and the high computational cost of adapting large pretrained language models. This study presents a GPT-based architecture for clinical text classification that adapts a pretrained decoder-only Transformer using a selective fine-tuning strategy. Rather than updating all model parameters, the majority of the GPT-2 backbone is frozen, and training is restricted to the final Transformer block, the final layer normalization, and a lightweight classification head. This approach substantially reduces the number of trainable parameters while preserving the representational capacity required to model complex clinical language. The proposed method is evaluated on radiology reports from the MIMIC-IV-Note dataset using uncertainty-aware CheXpert-style labels derived directly from report text. Experiments cover multiple problem formulations, including multi-label classification of radiographic findings, binary per-label classification under different uncertainty assumptions, and aggregate disease outcome prediction. Across varying dataset sizes, the model exhibits stable convergence behavior and strong classification performance, particularly in settings dominated by non-mention and negated findings. Overall, the results indicate that selective fine-tuning of pretrained generative language models provides an efficient and effective pathway for clinical text classification, enabling scalable adaptation to real-world EHR data while significantly reducing computational complexity.

Zizhen Liu, Fangzhiyi Wang, Mengdi Wang et al.

The growing demand for compute-intensive applications has made multi-chiplet architectures a promising alternative to monolithic designs, offering improved scalability and manufacturing flexibility. However, effectively managing the economic effectiveness remains challenging. Existing cost models either overlook the amortization of compute value over a chip's operational lifetime or fail to evaluate how redundancy strategies, which are widely adopted to enhance yield and fault tolerance, impact long-term cost efficiency. This paper presents a comprehensive cost-effectiveness framework for multi-chiplet architectures, introducing a novel Lifecycle Cost Effectiveness (LCE) metric that evaluates amortized compute costs by jointly optimizing manufacturing expenses and operational lifetime. Our approach uniquely integrates: (1) redundancy-aware cost modeling spanning both intra- and inter-chiplet levels, (2) reliability-driven lifetime estimation, and (3) quantitative analysis of how redundancy configurations on overall economic effectiveness. Extensive trade-off and multi-objective optimization studies demonstrate the effectiveness of the model and reveal essential co-optimization strategies between module and chiplet-level redundancy to achieve cost-efficient multi-chiplet architecture designs.

Jason Ho, James A. Boyle, Linshen Liu et al.

Neuromorphic systems using in-memory or event-driven computing are motivated by the need for more energy-efficient processing of artificial intelligence workloads. Emerging neuromorphic architectures aim to combine traditional digital designs with the computational efficiency of analog computing and novel device technologies. A crucial problem in the rapid exploration and co-design of such architectures is the lack of tools for fast and accurate modeling and simulation. Typical mixed-signal design tools integrate a digital simulator with an analog solver like SPICE, which is prohibitively slow for large systems. By contrast, behavioral modeling of analog components is faster, but existing approaches are fixed to specific architectures with limited energy and performance modeling. In this paper, we propose LASANA, a novel approach that leverages machine learning to derive data-driven surrogate models of analog sub-blocks in a digital backend architecture. LASANA uses SPICE-level simulations of a circuit to train ML models that predict circuit energy, performance, and behavior at analog/digital interfaces. Such models can provide energy and performance annotation on top of existing behavioral models or function as replacements to analog simulation. We apply LASANA to an analog crossbar array and a spiking neuron circuit. Running MNIST and spiking MNIST, LASANA surrogates demonstrate up to three orders of magnitude speedup over SPICE, with energy, latency, and behavioral error less than 7%, 8%, and 2%, respectively.

Kapal Dev, Sunder Ali Khowaja, Keshav Singh et al.

This paper investigates a range of cutting-edge technologies and architectural innovations aimed at simplifying network operations, reducing operational expenditure (OpEx), and enabling the deployment of new service models. The focus is on (i) Proposing novel, more efficient 6G architectures, with both Control and User planes enabling the seamless expansion of services, while addressing long-term 6G network evolution. (ii) Exploring advanced techniques for constrained artificial intelligence (AI) operations, particularly the design of AI agents for real-time learning, optimizing energy consumption, and the allocation of computational resources. (iii) Identifying technologies and architectures that support the orchestration of backend services using serverless computing models across multiple domains, particularly for vertical industries. (iv) Introducing optically-based, ultra-high-speed, low-latency network architectures, with fast optical switching and real-time control, replacing conventional electronic switching to reduce power consumption by an order of magnitude.

Miguel Rodrigo Castellanos, Siyun Yang, Chan-Byoung Chae et al.

Multiple-input multiple-output (MIMO) communication has led to immense enhancements in data rates and efficient spectrum management. The evolution of MIMO, though, has been accompanied by increased hardware complexity and array sizes, causing the system power consumption to increase. Despite past advances in power-efficient hybrid architectures, new solutions are needed to enable extremely large-scale MIMO deployments for 6G and beyond. In this paper, we introduce a novel architecture that integrates low-power reconfigurable antennas with both digital and analog precoding. This \emph{tri-hybrid} approach addresses key limitations in traditional and hybrid MIMO systems by improving power consumption and adds a new layer for signal processing. We provide an analysis of the proposed architecture and compare its performance with existing solutions, including fully-digital and hybrid MIMO systems. The results demonstrate significant improvements in energy efficiency, highlighting the potential of the tri-hybrid system to meet the growing demands of future wireless networks. We conclude the paper with a summary of design and implementation challenges, including the need for technological advancements in reconfigurable array hardware and tunable antenna parameters.

Tianwen Zhu, Hao Wang, Zhiwei Cao et al.

As digital twin technologies are increasingly incorporated into battery management systems to meet the growing need for transparent and lifecycle-aware operation, existing battery digital twins still suffer from fragmented operational processes and lack an architectural perspective to coordinate modeling, inference, and decision-making throughout the battery lifecycle. To this end, we develop a unified five-tier battery digital twin framework that integrates key functionalities into a coherent pipeline and facilitates a clearer architectural understanding of digital twins. The five-tier comprises geometric modeling, descriptive analytics, physics-informed prediction, prescriptive optimization, and autonomous control. In quantitative evaluation, the resulting architecture achieves high-fidelity multi-physics calibration with 0.92\% voltage and 0.18\% temperature prediction error, and provides state-of-health estimation with 1.09\% MAPE and calibrated uncertainty. As the first battery digital twin system empowered by the NVIDIA ecosystem with physics-AI technologies, our proposed five-tier framework shifts battery management from reactive protection to an interpretable, predictive, and autonomous paradigm, paving the path to develop next-generation battery management and energy management systems.

Yingying Zheng, Yun Long, Min Liu et al.

During the hydraulic performance experiment, significant vibration and noise were observed in the mixed-flow pump operating in the hump region. Cavitation occurrence in the impeller flow channels was confirmed through the transparent chamber. To analyze cavitation flow structure evolution in the mixed-flow pump, this paper integrates numerical and experimental approaches, capturing cavitation flow structures under the valley condition through high-speed photography technology. During the various stages of cavitation development, the cavitation forms are mostly vortex cavitation, cloud cavitation, and perpendicular vortex cavitation. Impeller rotation induces downstream transport of shedding cloud cavitation shedding structures. Flow blockage occurs when cavitation vortexes obstruct specific passages, accelerating cavitation growth that culminates in head reduction through energy dissipation mechanisms. Vortex evolution analysis revealed enhanced density of small-scale vortex structures with stronger localized core intensity in the impeller and diffuser. Despite larger individual vortex scales, reduced core intensity persists throughout the full flow domain. Concurrently, velocity profile characteristics across flow rates and blade sections (spanwise from tip to root) indicate heightened predisposition to flow separation, recirculation zones, and low-velocity regions during off-design operation. This study provides scientific guidance for enhancing anti-cavitation performance in the hump region.

Weiying KONG, Yizhuo LIU, Sichun DONG et al.

ObjectiveIn the contemporary global context, urban areas are increasingly confronted with the dual pressures of global climate change and rapid urbanization. These pressures have led to a significant rise in urban temperature, thereby amplifying the importance of blue-green spaces in mitigating the urban heat island (UHI) effect. Blue-green spaces, which include natural water bodies, parks, green corridors, and other vegetated areas, play a crucial role in regulating urban microclimates. As cities enter an era of stock development, where the focus shifts from expansion to optimization of existing resources, the strategic configuration of these spaces has become a cornerstone for enhancing urban thermal environments. Understanding the cooling mechanisms of blue-green spaces at various spatial scales is essential for improving urban thermal comfort and guiding the planning and construction of urban blue-green infrastructure.MethodsThis research focuses on the central urban area of Xi’an, a city that has experienced substantial urban growth over the past decade. By employing a combination of spatial autocorrelation analysis and a multi-scale geographically weighted regression (MGWR) model, the research examines the change characteristics of blue-green spaces and their impact on land surface temperature from 2013 to 2023. The findings reveal the spatial heterogeneity of cooling effects and offer tailored optimization strategies for blue-green spaces across diverse urban contexts. The research methodology involves selecting six representative landscape indices to evaluate the changes in blue-green space patterns in the central urban area of Xi’an. These indices are carefully chosen to capture the nuances of spatial configuration, fragmentation, and connectivity of blue-green spaces. Spatial autocorrelation analysis is utilized to identify spatial clustering and patterns extracted from the data collected, while the MGWR model is adopted for a more granular examination of the relationship between landscape indices and land surface temperature levels. This integrated approach not only reveals the factors influencing the cooling effects of blue-green spaces but also highlights their spatial variability across the urban landscape.ResultsThe results of the research are both revealing and instructive. 1) The blue-green space patterns in the central urban area of Xi’an underwent significant changes over the research period, reflecting the dynamic interplay between urban development and environmental management. 2) The spatial distribution of land surface temperature exhibits a distinct pattern of being “high in the north and low in the south”. The central area, characterized by dense urban fabric, shows minimal fluctuations in land surface temperature, whereas low-temperature zones are predominantly concentrated in the southern part of Baqiao District. This uneven thermal distribution underscores the complexity of urban heat dynamics and the need for targeted interventions. 3) The relationship between landscape indices and land surface temperature changes displays notable spatial heterogeneity. In high-density urban areas, small and complex blue-green patches demonstrate stronger cooling effects, emphasizing the importance of intricate designs in densely built environments where space is limited but the need for effective cooling is significant. In contrast, suburban areas benefit from avoiding the aggregation of large blue-green patches, which may otherwise hinder effective cooling due to reduced air circulation and increased shading. Near large water bodies, regularly shaped and highly connected blue-green patches are found to be particularly effective in reducing land surface temperature, highlighting the synergistic effects of water and vegetation in enhancing cooling performance and suggesting that integrated blue-green networks can maximize thermal benefits.ConclusionThe research concludes that the relationship between temperature changes and blue-green space changes in the central urban area of Xi’an is significant and characterized by strong spatial heterogeneity during the period from 2013 to 2023, with the cooling effects of blue-green spaces found varying by their spatial attributes and the characteristics of the surrounding urban environment. These findings highlight the necessity for region-specific optimization strategies to maximize the cooling potential of blue-green spaces. By integrating spatial analysis and regression modeling, the research provides a detailed understanding of the cooling mechanisms of blue-green spaces across diverse urban contexts. The results emphasize the importance of tailoring blue-green space designs to local conditions, considering factors such as urban density, proximity to water bodies, and regional climatic characteristics. This approach enhances the effectiveness of blue-green spaces in mitigating the urban heat island effect and contributes to the creation of more sustainable and thermally comfortable urban environments. The research advocates a holistic and adaptive urban planning strategy, where blue-green spaces are strategically designed and managed to address the unique thermal challenges of different urban areas. This research offers valuable guidance for policymakers and urban planners aiming to optimize blue-green infrastructure and improve urban resilience in the face of climate change and urbanization.

Gebrail Bekdaş, Yaren Aydın, Umit Işıkdağ et al.

Shear wave velocity (V_s) is an important soil parameter to be known for earthquake-resistant structural design and an important parameter for determining the dynamic properties of soils such as modulus of elasticity and shear modulus. Different V_s measurement methods are available. However, these methods, which are costly and labor intensive, have led to the search for new methods for determining the V_s. This study aims to predict shear wave velocity (V_s (m/s)) using depth (m), cone resistance (q_c) (MPa), sleeve friction (f_s) (kPa), pore water pressure (u₂) (kPa), N, and unit weight (kN/m³). Since shear wave velocity varies with depth, regression studies were performed at depths up to 30 m in this study. The dataset used in this study is an open-source dataset, and the soil data are from the Taipei Basin. This dataset was extracted, and a 494-line dataset was created. In this study, using HyperNetExplorer 2024V1, V_s prediction based on depth (m), cone resistance (q_c) (MPa), shell friction (f_s), pore water pressure (u₂) (kPa), N, and unit weight (kN/m³) values could be performed with satisfactory results (R² = 0.78, MSE = 596.43). Satisfactory results were obtained in this study, in which Explainable Artificial Intelligence (XAI) models were also used.

Tengfei Song, Yu Liu, Xuefei Zhang et al.

About ten years ago, we established the first coronagraph that has been continuously operating on the high plateau of western China. This coronagraph is an internal occulting, 10 cm aperture instrument, installed at Lijiang Station through a collaboration with the Norikura Station of the National Astronomical Observatory of Japan. To ensure high efficiency in current and future coronal observations, developing integrated observation systems is essential for reliable, autonomous, and remote operation of coronagraphs. This paper introduces an advanced integrated observation and control system, based on the Lijiang 10 cm coronagraph. The coronagraph focuses on the observations for the solar inner corona, capturing the coronal green-line emission within a field range from <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>1.03</mn><msub><mi>R</mi><mo>⨀</mo></msub></mrow></semantics></math></inline-formula> to <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>2.5</mn><msub><mi>R</mi><mo>⨀</mo></msub></mrow></semantics></math></inline-formula>. To enhance the observational precision and efficiency, a comprehensive integrated system has been designed, incorporating various subsystems, including precise pointing and tracking mechanisms, a multi-band filter system, a protective dome system, and a robust data storage infrastructure. This paper details the hardware architecture and software frameworks supporting each subsystem. Results from extended operational testing confirm the stability of the system, its capacity for autonomous and remote observations, and significant improvements in the automation and efficiency of coronal imaging. The automated observation system will be further improved and used for our future coronagraphs to be developed for coronal magnetism diagnosis.

Ivona Nedevska Trajkova, Zlatko Zafirovski

Understanding the rate of track degradation is essential for effective railway infrastructure management, particularly in mountainous and geotechnically unstable regions. This paper presents a comprehensive analysis of the track geometry degradation on the Kolašin–Podgorica railway section in Montenegro, using the Track Quality Index (TQI) as the primary indicator. TQI values from three consecutive inspection periods (2017–2019, 2019–2022, and 2022–2024) were analyzed to compute degradation rates (ΔTQI) across all track segments. The results were visualized through spatially distributed line graphs, enabling the identification of segments with progressive geometric deterioration. The analysis reveals a recurring pattern: several sections demonstrate improvement following tamping interventions, yet degrade again within a short period, indicating deeper structural or geotechnical issues. Particular attention is given to sections located on bridges, in tunnels, and near stations—areas associated with increased dynamic loads and limited substructure resilience. An overlay of maintenance data and structural object locations further strengthens the causal interpretation. The findings support the prioritization of high-risk segments for targeted interventions beyond routine maintenance. This degradation-based evaluation framework contributes to data-driven decision-making for long-term railway asset management, combining infrastructure condition assessment with spatial engineering analytics.

Peter Gersing, Mark Doll, Joerg Huschke et al.

Integrating sensing functionality into 6G communication networks requires some changes to existing components as well as new entities processing the radar sensing signals received by the communication antennas. This whitepaper provides a comprehensive overview of the 6G design proposal for ISaC (Integrated Sensing and Communication). The whitepaper has been created by the architecture group of the KOMSENS-6G project with the intend to serve as a basis for further discussions and alignment across innovative 6G projects.

Abdullah Arafat Miah, Yu Bi

Deep neural networks (DNNs) have long been recognized as vulnerable to backdoor attacks. By providing poisoned training data in the fine-tuning process, the attacker can implant a backdoor into the victim model. This enables input samples meeting specific textual trigger patterns to be classified as target labels of the attacker's choice. While such black-box attacks have been well explored in both computer vision and natural language processing (NLP), backdoor attacks relying on white-box attack philosophy have hardly been thoroughly investigated. In this paper, we take the first step to introduce a new type of backdoor attack that conceals itself within the underlying model architecture. Specifically, we propose to design separate backdoor modules consisting of two functions: trigger detection and noise injection. The add-on modules of model architecture layers can detect the presence of input trigger tokens and modify layer weights using Gaussian noise to disturb the feature distribution of the baseline model. We conduct extensive experiments to evaluate our attack methods using two model architecture settings on five different large language datasets. We demonstrate that the training-free architectural backdoor on a large language model poses a genuine threat. Unlike the-state-of-art work, it can survive the rigorous fine-tuning and retraining process, as well as evade output probability-based defense methods (i.e. BDDR). All the code and data is available https://github.com/SiSL-URI/Arch_Backdoor_LLM.

Cenlin Duan, Jianlei Yang, Yiou Wang et al.

Bit-level sparsity in neural network models harbors immense untapped potential. Eliminating redundant calculations of randomly distributed zero-bits significantly boosts computational efficiency. Yet, traditional digital SRAM-PIM architecture, limited by rigid crossbar architecture, struggles to effectively exploit this unstructured sparsity. To address this challenge, we propose Dyadic Block PIM (DB-PIM), a groundbreaking algorithm-architecture co-design framework. First, we propose an algorithm coupled with a distinctive sparsity pattern, termed a dyadic block (DB), that preserves the random distribution of non-zero bits to maintain accuracy while restricting the number of these bits in each weight to improve regularity. Architecturally, we develop a custom PIM macro that includes dyadic block multiplication units (DBMUs) and Canonical Signed Digit (CSD)-based adder trees, specifically tailored for Multiply-Accumulate (MAC) operations. An input pre-processing unit (IPU) further refines performance and efficiency by capitalizing on block-wise input sparsity. Results show that our proposed co-design framework achieves a remarkable speedup of up to 7.69x and energy savings of 83.43%.

Zerui Wang, Yan Liu, Jun Huang

This article presents the design of an open-API-based explainable AI (XAI) service to provide feature contribution explanations for cloud AI services. Cloud AI services are widely used to develop domain-specific applications with precise learning metrics. However, the underlying cloud AI services remain opaque on how the model produces the prediction. We argue that XAI operations are accessible as open APIs to enable the consolidation of the XAI operations into the cloud AI services assessment. We propose a design using a microservice architecture that offers feature contribution explanations for cloud AI services without unfolding the network structure of the cloud models. We can also utilize this architecture to evaluate the model performance and XAI consistency metrics showing cloud AI services trustworthiness. We collect provenance data from operational pipelines to enable reproducibility within the XAI service. Furthermore, we present the discovery scenarios for the experimental tests regarding model performance and XAI consistency metrics for the leading cloud vision AI services. The results confirm that the architecture, based on open APIs, is cloud-agnostic. Additionally, data augmentations result in measurable improvements in XAI consistency metrics for cloud AI services.

María Teresa Gómez-Villarino, María del Mar Barbero-Barrera, Ignacio Cañas et al.

The wine industry requires a considerable amount of energy, with an important fraction corresponding to the cooling and ventilation of above-ground aging warehouses. The large investments made in aging facilities can compromise the viability and competitiveness of wineries if their design is not optimized. The objective of this study was to provide guidance for the efficient design of new above-ground warehouses. To this end, multiple construction solutions (structure, envelopes, levels of integration, etc.) were characterized, and their costs and the resulting interior environments were analyzed. The results offer a comprehensive view of potential construction solutions and benchmark price ranges for viable and profitable designs. With a total cost of 300 EUR/m², an average damping of 98% per day can be achieved. Increasing the costs does not imply better effectiveness. A double enclosure with internal insulation—with or without an air chamber—can achieve excellent results. Greater integration as a result of several enclosures being in contact with other rooms and/or the terrain allows for a high effectiveness to be achieved without air conditioning. Perimeter glazing and ventilation holes can reduce the effectiveness of the construction, resulting in greater instability and a lower damping capacity.

Yichen Yang, Jingtao Li, Nishil Talati et al.

The irregular nature of memory accesses of graph workloads makes their performance poor on modern computing platforms. On manycore reconfigurable architectures (MRAs), in particular, even state-of-the-art graph prefetchers do not work well (only 3% speedup), since they are designed for traditional CPUs. This is because caches in MRAs are typically not large enough to host a large quantity of prefetched data, and many employs shared caches that such prefetchers simply do not support. This paper studies the design of a data prefetcher for an MRA called Transmuter. The prefetcher is built on top of Prodigy, the current best-performing data prefetcher for CPUs. The key design elements that adapt the prefetcher to the MRA include fused prefetcher status handling registers and a prefetch handshake protocol to support run-time reconfiguration, in addition, a redesign of the cache structure in Transmuter. An evaluation of popular graph workloads shows that synergistic integration of these architectures outperforms a baseline without prefetcher by 1.27x on average and by as much as 2.72x on some workloads.