- Book Chapter
- 10.1007/978-3-032-08894-9_16
Impact of Data Augmentation Applied to Fundus Images Used in a Swin Transformer V2-Based Model for the Detection of Age-Related Macular Degeneration
- Jan 01, 2026
- Roberto Marquez Castro + 4 more +4
Publications from 2021 to 2026
Showing 10 of 26 papers
Impact of Data Augmentation Applied to Fundus Images Used in a Swin Transformer V2-Based Model for the Detection of Age-Related Macular Degeneration
Edge-Disjoint Tree Allocation for Multi-Tenant Cloud Security in Datacenter Topologies
Resource sharing with its implied mutual interference has been considered a major concern for running applications of multiple tenants in shared cloud datacenters. Besides its security benefits, the isolation of traffic might ensure a quality of service (QoS) performance guarantee avoiding interference among tenants. Traffic isolation can be achieved by dedicating the usage of link resources in the network to a single tenant preventing its sharing among others. Accordingly, tenants should be connected through an edge-disjoint tree to enable isolated communication among its hosts. In this paper, we study the problem of establishing edge-disjoint trees in common datacenter topologies. We show that the availability of such trees is highly affected by the mapping of the tenants to hosts of the topology. Specifically, with the flexibility to map tenants in the datacenter topology, we describe a mapping algorithm and an optimal tree establishment for the optimization problem. Given the mapping of the tenants, we prove the problem turns out to be NP-Hard and provide comprehensive heuristics for the problem. Finally, we conduct experiments using real workloads to examine tree availability under various scenarios.
Read moreAtt studera gymnasieelevers prestationsemotioner genom berättelser
Följande ULF-projekt (Utveckling, lärande, forskning) är ett samarbete mellan en kommunal gymnasieskola i en mellanstor stad samt Linköpings Universitet. Syftet är att öka förståelsen för hur gymnasieelever upplever prestationssituationer. Genom en narrativ ansats har 261 elevberättelser analyserats och kategoriserats. De känslor som kan knytas till erfarenheter av att bli bedömda kallas i studien för prestationsemotioner. En prestation kan definieras som en aktivitet som mäts mot någon form av standard och prestationsemotioner är emotioner som kopplas till en sådan aktivitet. Vanligt förekommande positiva prestationsemotioner är stolthet, nöje, hopp, tacksamhet, avslappning, belåtenhet och lättnad. Negativa prestationsemotioner är oro, ångest, stress, skam, tristess, hopplöshet eller besvikelse (Pekrun, 2007; 2014). I studien finns en rad positiva och negativa emotioner som eleverna tillskriver sina prestationer, samt att elevernas känslor ofta förändras i samband med viktiga prestationer. Eleverna gör känsloresor som verkar ha betydelse för deras vilja och förmåga att prestera i framtiden. En mer utförlig resultatredovisning görs i en annan artikel i Venue. Utöver dessa resultat har projektet skapat ett kollegialt lärande som lyfter och bidrar till en mer hållbar bedömningspraktik.
Read moreRequirements to Products and Processes for Software of Safety Important NPP I&C Systems
Features of software as a component of instrumentation and control (I&C) systems are analyzed. Attention is paid to the importance of functions performed by software and hazards of such software. Requirements for characteristics of software as a component of I&C systems are analyzed. Different regulatory documents are considered in order to disclose common approaches to the use of dedicated software and off-the-shelf software components. Classification of software, as well as classification of requirements, is described. Criteria of selection and structuring of requirements, as well as criteria for software verification, are defined. As long as the characteristics of software components directly depend on the quality of the processes of software development and verification, requirements for software life cycle processes are considered.
Read morePredicting the flowability of alumina powder during batch grinding through the establishment of a grinding kinetic model
Case Report: Cetuximab in Combination With Chemotherapy for the Treatment of Multifocal Hepatic Metastases From Colorectal Cancer Guided by Genetic Tests
Hepatic metastases were reported in up to 70% of colorectal cancer patients, among which multifocal hepatic metastasis represents one of the complications that lead to poor prognosis. The majority of the patients carrying multifocal hepatic metastases required pharmaceutical treatments to reduce the tumor size prior to surgical resection. However, the clinical responses to pharmaceutical agents were difficult to predict due to the heterogeneous nature of the multifocal tumors. Here, we report a case with multifocal hepatic metastases from colorectal cancer that was resistant to the primary chemotherapy and Bevacizumab plus chemotherapy, but responded to the combined therapy of Cetuximab and FOLFOX. Genetic tests had revealed that the tumor was highly metastatic due to the mutations of the WNT signaling pathway, and the metastatic tumors might be sensitive to Cetuximab. Consistent with the molecular characterizations, the metastatic tumors continue to emerge after chemotherapy, and rapidly relapsed in great numbers after liver resection. However, the combined therapy of Cetuximab and FOLFOX guided by the genetic tests significantly reduced the size and number of metastatic tumors. To conclude, deciphering the mutation profiles of multifocal metastatic tumors may guide the determination of treatment tactics, which may benefit the patients with non-resectable advanced carcinoma.
Read moreThe path to 1Tb/s and beyond datacenter interconnect networks: technologies, components, and subsystems
International audience
High Capacity Converged Passive Optical Network and RoF-Based 5G+ Fronthaul Using 4-PAM and NOMA-CAP Signals
Network architectural changes to satisfy all the 5G+ mobile network specifications and requirements are necessary due to the popularization of streaming and cloud applications on omnipresent portable devices. The combination of massive installation of micro-cell antenna sites with the cloud access radio network (C-RAN) architecture has recently been nominated as a promising technology for high-capacity mobile fronthaul links, albeit at a high cost. An alternative approach for next-generation fronthaul networks is to utilize the already deployed passive optical networks (PONs) where wireless and wired services may coexist in a converged manner. Non-orthogonal multiple access (NOMA) modulation with multi-band carrierless amplitude and phase modulation (NOMA-CAP) has recently been investigated as a promising 5G+ modulation format candidate to increase the capacity and flexibility of future mobile networks. Here, we experimentally demonstrate the convergence of a NOMA-CAP wireless waveform with a single-carrier wired signal in a PON scenario using radio-over-fiber (RoF) technology. Specifically, fifteen NOMA-CAP bands, with two NOMA power levels to double the capacity, transmit 15 Gb/s multiplexed with a digital 10 Gb/s four-level pulse amplitude modulation (PAM-4) signal for downlink application. Two converged system implementations have been considered, first using electrical frequency division multiplexing (EFDM) and secondly using the hybrid EFDM-wavelength division multiplexing (EFDM-WDM). Successful transmission through a 25 km span of standard single-mode fiber is achieved with negligible transmission penalty for both proposed converged solutions.
Read moreScalable Hierarchical Aggregation and Reduction Protocol (SHARP)TM Streaming-Aggregation Hardware Design and Evaluation
This paper describes the new hardware-based streaming-aggregation capability added to Mellanox’s Scalable Hierarchical Aggregation and Reduction Protocol in its HDR InfiniBand switches. For large messages, this capability is designed to achieve reduction bandwidths similar to those of point-to-point messages of the same size, and complements the latency-optimized low-latency aggregation reduction capabilities, aimed at small data reductions. MPI_Allreduce() bandwidth measured on an HDR InfiniBand based system achieves about 95% of network bandwidth. For medium and large data reduction this also improves the reduction bandwidth by a factor of 2–5 relative to host-based (e.g., software-based) reduction algorithms. Using this capability also increased DL-Poly and PyTorch application performance by as much as 4% and 18%, respectively. This paper describes SHARP Streaming-Aggregation hardware architecture and a set of synthetic and application benchmarks used to study this new reduction capability, and the range of data sizes for which Streaming-Aggregation performs better than the low-latency aggregation algorithm.
Read moreAn Efficient Stabilization Process for Analog Fractionally Spaced Equalizers
An analog fractionally spaced equalizer (FSE) is an efficient solution for high-speed communication system without a clock and data recovery circuit. However, the FSE might become unstable when a sign–sign least mean square tuning algorithm is used, especially when the transmitter pre-distortion is used. Here, an analytical model of the instability is derived, and a low-cost and efficient stabilization process based on dithering is proposed. The analytical instability model is verified in an experimental setup based on a vertical cavity surface emitting laser. Furthermore, it is experimentally demonstrated that in the case of the instability of the FSE, a clear convergence is achieved when applying the proposed stabilization scheme.
Read more