Mando Corporation (South Korea)

Transformer-based sequential recommendation (SR) models excel at modeling long-range dependencies, but suffer from high computational costs and catastrophic forgetting during continuous updates. Although continual learning has been applied to recommendation, existing methods gradually forget long-term user preferences and remain underexplored in SR. In this paper, we introduce Continual Sequential Transformer for Recommendation (CSTRec), which effectively adapt to current interests by leveraging preserved historical knowledge. Its core is Continual Sequential Attention (CSA), a linear attention tailored for continual SR, which partially retain historical knowledge without direct access to prior data. CSA features: (1) Cauchy-Schwarz Normalization to stabilize learning over time under uneven user interaction frequencies, and (2) Collaborative Interest Enrichment via shared, learnable interest pools to mitigate forgetting. We also introduce a new technique for new user adaptation by transferring historical knowledge from existing users with similar interests. Extensive experiments show CSTRec's superior performance in both knowledge retention and acquisition. Our code is available at https://github.com/Gyu-Seok0/CSTRec_WSDM26.

The experimental trial was conducted at the Oilseeds Research Institute, Agriculture Research Center (ARC), Tando Jam, during the Rabi season of 2023. Analysis of variance revealed significant differences (p < 0.01) for all traits studied, indicating substantial variation among the sunflower genotypes. The genotype SMH-1900-E demonstrated the highest germination percentage (87.66%), while SMH-1900-A had the shortest time to first flowering (62.66 days) and maturity (99.33 days). SAHRA-SUN-66 showed the largest head diameter (15.33 cm), and ACSUN-F1 recorded the highest 1000-grain weight (44.00 g). The highest seed yield plant-1 was observed in ORISUN-701 (761.3 g), while FH-829 had the highest seed yield per hectare (568.7 kg ha-1). Correlation analysis revealed significant positive relationships between days to first flowering and germination (r = 0.22*), plant height (r = 0.31*), head diameter (r = 0.35**), and 1000-seed weight (r = 0.33**). Plant height had a significant positive correlation with 1000-seed weight (r = 0.24*) and seed yield plant-1 (r = 0.53**). Head diameter was positively correlated with the number of seeds per head (r = 0.24*), seed yield plant-1 (r = 0.21*), and seed yield per hectare (r = 0.22*). Additionally, plant population showed a significant positive correlation with seed yield plant-1 (r = 0.29*). A dendrogram based on Jaccard’s similarity coefficients, generated by UPGMA cluster analysis, classified the genotypes into two main clusters. Cluster 1 included nine genotypes, while Cluster 2 comprised eleven genotypes, indicating distinct relationships between the groups. Principal Component Analysis (PCA) identified that plant height, germination rate, number of seeds per head, and seed yield (kg ha-1) were the main factors contributing to genotype variation. The highest variability was explained by PC-VI and PC-VII, accounting for 98.3% and 97.1% of the total variation, respectively.

In safety-critical domains such as industrial processes and automotive systems, accidents can have serious consequences, including the potential loss of human life. To mitigate risks associated with hazards, these systems must incorporate robust safety capabilities. International standards, such as IEC 61508, mandate the implementation of safety functions and require their verification through SIL (Safety Integrity Level) assessments. However, performing accurate SIL verification remains challenging due to the uncertainty of real-world data, which can be influenced by factors such as temperature variations (i.e., heating effects), environmental stressors, and component aging. In particular, failure rates, which serve as fundamental data for SIL verification, are typically assumed to be constant values derived under ideal conditions. But in practice, these rates vary over time due to the progressive degradation of system components, leading to potential discrepancies between theoretical safety assessments and real-world reliability. To address this issue, researchers have explored Markov models as a dynamic approach for improving SIL verification. In this paper, we propose a novel approach based on Markov; constructing one of the fault-tolerant architecture, 2oo3 (2-out-of-3), as a model and verifying SIL using probabilistic model checking. Our approach synthesizes the model to prevent systematic faults, such as human errors, and enhances verification accuracy. To validate our approach, we compare the results of the probabilistic model checking performed on our synthesized model with those obtained from a model without synthesis. The results demonstrate that the synthesized model exhibits the smallest difference rate and a more predictable variation from the target values specified in the IEC 61508 standard for SIL verification. Experimental results confirm that the proposed method enhances the reliability of SIL verification, increasing confidence in safety assessments.

Cloned code clustering identifies identical or similar code fragments, i.e., cloned code, from products of a product family, and then constructs cloned code clusters, a set of files, each of which contains the same cloned code. It is an essential step for migrating from the Clone-And-Own (CAO) approach to the Software Product Line Engineering approach for developing a family of products. This paper proposes a method for cloned code clustering based on files that share cloned code. The method proposed in this paper identifies clusters to use for constructing a product line platform at the source code level in such a way that it works regardless of cloning-in-the-small or cloning-in-the-large; it does not need to know what code is original or what code is cloned; and its results remain consistent regardless of the comparison order or the relative similarity of products of a product family. For evaluation we applied our method to Apo-Games and ArgoUML developed with the CAO approach and confirmed that our method correctly constructs code clusters.

A face é como o cartão de visita do ser humano, e o mundo estético vem se aprimorando para manter a cada dia as pessoas satisfeitas e com resultados satisfatórios nos procedimentos que se procura, seja a procura apenas da beleza ou para corrigir um trauma. Para isso existem vários preenchedores no mercado estético que ajudam os pacientes na sua busca de melhorar a auto-estima. O presente artigo se trata de uma revisão integrativa acerca de um dos preenchedores que está sendo muito usado que é o polimetilmetacrilato (PMMA), um preenchedor permanente, eficaz e de baixo custo, que vem auxiliando muito os que procuram. E demonstra alguns aspectos acerca do PMMA, no que concerne a sua composição, bem como benefícios e intercorrências que acontecem no uso desse polímero. Ainda aponta os motivos da necessidade da precaução na busca do profissional qualificado, para que não haja intercorrencias que venham a atingir a saúde do paciente.

Abstract This study aims to develop an LSM tree that uses two different kinds of probabilistic data structures (PDS). These two data structures are the Bloom Filters and the state-of-the-art Cuckoo Filter released in 2014 by Fan. Cuckoo filters are a perfect choice for saving space and also for deleting keys that are not needed in the algorithm so the filters are synchronized with the keys of all the SSTables. The implementation in this study of Cuckoo filters showed three main advantages (1) Keys that are deleted in the SSTables are immediately reflected in the Cuckoo Filters (2) When SSTables are merged, Cuckoo Filters are also merged by measuring its load factor for availability (3) The performance of Cuckoo Filters compared to Bloom filters after deleting and query shows better performance despite that cuckoo filters are also being updated. This approach is expected to reduce storage space and the amount of time for queries and deletions on the overall data structure. This new optimized LSM tree is expected to be implemented in HBase as future work to check its performance with a real-world database for more empirical results.

In this paper, we present high step-up ripple-free resonant dc/dc converter that does not use auxiliary inductors. By using the push-pull transformer with the primary-side switching modulation fixed at 0.5, the proposed converter accomplishes theoretically zero input-current ripple. By removing the input auxiliary inductors, the proposed converter achieves high power density. In addition, this converter can achieve zero-voltage switching of the primary-side switches and zero-current switching of output diodes under a wide input voltage range. A 1-kW experimental prototype was implemented and it was tested to validate the superior properties of the proposed converter.

Tenaga kerja yang berkualitas diperlukan dalam proyek konstruksi untuk mencapai hasil yang diinginkan karena produktivitas pekerja yang tinggi, khususnya dalam proyek konstruksi, memiliki dampak yang signifikan terhadap hasil proyek. Plesteran dan acian dinding adalah salah satu pekerjaan yang menuntut pekerja produktif. Tujuan dari penelitian ini adalah untuk mengetahui koefisien produktivitas pekerjaan plesteran dan membandingkannya dengan SNI 2837-2008. Penelitian ini dilakukan di lapangan melalui observasi langsung. Pengamatan ini dilakukan untuk mengumpulkan informasi primer yang dibutuhkan untuk menghitung produktivitas pekerjaan plesteran yang dilakukan pada proyek pembangunan Masjid MIBAS. 66,06 m2/hari untuk mandor, 6.60 m2/hari untuk tukang, dan 13,47 m2/hari untuk pembantu tukang merupakan hasil rata-rata dari analisis koefisien produktivitas plesteran dinding dan acian pada proyek pembangunan Masjid MIBAS. Hasil produktivitas SNI yaitu 45,45 m2/hari untuk mandor, 50 m2/hari untuk kepala tukang, 5 m2/hari untuk tukang, dan 2,5 m2/hari untuk pembantu tukang. Hasilnya, selisih bagian mandor +20,61 m2/hari, bagian kepala tukang +16,06 m2/hari, bagian tukang +4,10 m2/hari, dan bagian pembantu tukang +8,47 m2/hari. Bila ada tanda plus, berarti hasilnya lebih tinggi dari yang ditentukan oleh SNI 2837 – 2008.

<div class="section abstract"><div class="htmlview paragraph">With the spread of new trends such as autonomous driving and vehicle subscription service, drivers may pay less attention to the maintenance of the vehicle. Brake pads being safety critical components, the wear condition of all service brakes is required by regulation to be indicated by either acoustic of optical devices or a means of visually checking the degree of brake lining wear [<span class="xref">1</span>]. Current application of the wear indicator in the market uses either sound generating metal strip or wire harness based pad wear sensor. The former is not effective in generating clear alarm to the driver, and the latter is not cost effective, and there is a need for more effective and low cost solution. In this paper, a pad wear monitoring system using MOC(Motor On Caliper) EPB(Electric Parking Brake) ECU is proposed. An MOC EPB is equipped with a motor, geartrain and an ECU. The motor current when applying the parking brake is influenced by the mechanical load at the brake pad side of the system. So, by analyzing the time history of the current it is possible to measure the clearance between brake pad and disc induced by the pad wear. From the measurement of the low load interval along with the mechanical specifications of the geartrain (e.g. lead of the screw), pad wear can be calculated. A sequential procedure for pad wear measurement mode is also proposed. A HILS test bench using production MOC actuator was setup and the measurement accuracy was evaluated across various conditions such as amount of wear, supply voltage, and temperature. The result is shown and a further discussion about practical aspects of the technology is added.</div></div>

Presentamos un análisis de los protocolos de escribanos de la ciudad de Córdoba en el último cuarto del siglo XVI para reconocer las posibilidades y límites que presentan para el estudio de las movilidades de personas y bienes desde y hacia la jurisdicción en cuestión. El estudio se centra en los aspectos metodológicos del estudio de este tipo de documentación y busca destacar el valor de las escrituras públicas, en tanto fuentes seriables, para la historia social y económica a escala local como regional y/o global.