Hasil untuk "Modern"

Thongchai Winichakul

Xiaowen Zhang, Hannuo Zhang, Shin Hwei Tan

Modern agentic frameworks (e.g., CrewAI and AutoGen) have evolved into complex, autonomous multi-agent systems, introducing unique reliability challenges beyond earlier pipeline-based LLM libraries. However, existing empirical studies focus on earlier LLM libraries or task-level bugs, leaving the unique complexities of these agentic frameworks unexplored. We bridge the gap by conducting a comprehensive study of 409 fixed bugs from five representative agentic frameworks. We propose a five-layer abstraction to capture structural complexities in agentic frameworks, spanning from orchestration to infrastructure. Our study uncovers specialized symptoms, such as unexpected execution sequences and user configurations ignored, which are unique to autonomous orchestration. We further identify agent-specific root causes, including modelrelated faults, cognitive context mismanagement, and orchestration faults. Statistical analysis reveals cross-framework consistency and significant associations among these bug dimensions. Finally, our automated pattern mining identifies frequent bug-triggering patterns (e.g., model backend-ID combinations), and we show their transferability across different framework designs. Our findings facilitate cross-platform testing and improve the reliability of agentic systems.

The modern business environment depends on green innovation to achieve sustainable growth through environmentally friendly products that boost both operational effectiveness and market competencies. Research explains how external environmental pressures relate to green innovation measures alongside their effects on firm performance and continues to grow more because current studies show insufficient evidence of these interdependent connections. The study examines multiple elements affecting green innovation performance through external influences to help organizational leaders, together with policymakers, bring sustainable practices into an environmentally aware business market. This study investigated the influence of green innovation on firm performance only through the subjects of management engagement, customer demand for sustainability, and supply chain risk. Optimizing green innovation is a strategic process aimed at increasing the long-term resilience and sustainable performance of an organization. The study draws from theories of green innovation, sustainability, and firm performance, hypothesizing interactions among green practices, managerial support, consumer preferences, and supply chain dynamics. The study combines environmental policies, employee involvement, and external collaboration to enhance its theoretical framework. The quantitative method of PLS-SEM was used to analyse data gathered from 621 companies in Pakistan. This approach uses structured questions to measure constructs, which include green product innovation, green process innovation, management commitment and firm performance. Green product and process innovation drives the performance of a firm, with both management and consumers in need of sustainability. However, supply chain risks reduce the impacts. The findings imply that sustainability in central business and supply chain shortcomings are three factors that need to be considered. The research includes empirical findings related to green innovation. It can help managers develop sustainable policies that are performing well in coping with market requirements and reducing risk. It also shows companies’ contributions to the progress of environmental goals. The study reconnects determinants and impediments of green innovation; it highlights the influence on the firm and provides principal implemented indications towards the aim of sustainable exercise.

Krisztián Csák, Csilla Juhász

A modern üzleti környezetben a márkaépítés szerepe túlmutat a termékek és szolgáltatások azonosításán: alapvető tényezővé vált a vállalati siker szempontjából. A vezetői énmárka és a munkáltatói márka fontos stratégiai eszközként jelenik meg a versenyelőny elérésében, a lojalitás és az elkötelezettség növelésében. Az énmárka, mint az egyéni értékek és képességek tudatos kommunikációja, segíti a vezetőket a hitelesség és bizalom erősítésében. A vezetői énmárka a vezető személyiségének és vezetői stílusának olyan koherens összefoglalása, amely inspiráló példaként szolgál a szervezeten belül. Emellett a munkáltatói márka a vállalati kultúrát és munkahelyi értékeket tükrözi, amely vonzóvá teheti a szervezetet a tehetséges munkavállalók számára. A tanulmány célja, hogy bemutassa, miként járulhat hozzá az énmárka, ezen belül a vezetői énmárka és a munkáltatói márka együttesen a vállalat eredményesebb működéséhez, rámutatva a márkázás különböző szintjeinek szinergiájára és ezek gazdasági, társadalmi hatásaira.

Катерина УШКАЛО

The article analyzes the interrelation between the components of mass media (mass information and mass communication) and social networks in the context of their influence on users’ perception and trust in information. The theoretical research focuses on identifying the features of the functioning of the modern media space, where traditional mass media increasingly intertwine with digital technologies, creating new opportunities for content dissemination but also generating new risks. The purpose of the study is to analyze the impact of new media, particularly social networks, on the spread of manipulative content such as fake news, disinformation, and emotional narratives in the national information space of Ukraine. The article emphasizes that the development of digital technologies has significantly changed society’s attitude toward searching for and processing information, especially after the full-scale invasion of Russian troops into Ukraine. The popularity of social networks compared to other mass media has been analyzed, and the factors contributing to the spread of manipulative information have been identified, including: algorithm-based content promotion, the «echo chamber» effect, the mass use of digital bots, the application of artificial intelligence technologies, the spread of memes and visual narratives, as well as the growing activity of Telegram channels and alternative media platforms following the decline in trust toward traditional media. It is substantiated that despite the gradual decline in trust in social networks, users continue to actively consume news from these sources, valuing their speed, convenience, and the ability to independently select and filter information. Such a trend increases citizens’ vulnerability to informational manipulation, weakens the information resilience of society, and poses security risks to the state. The obtained results provide a deeper understanding of the mechanisms of public opinion formation in the context of digitalization and highlight the need to enhance critical thinking and media literacy at the national level.

SUN Haoran, WANG Zhihao, WU Yifan et al.

Computer education is one of the requirements of modern information society education. With the development of large language models, there has been increasing attention on applying of large language models to the computer education process. However, the hallucination problem associated with large language models poses significant challenges to their application. To solve the challenges, RAG techniques by incorporating external knowledge bases can effectively enhance the quality of responses generated by large language models. However, the traditional RAG techniques lack a fine screening mechanism for the retrieved information, which leads to the retention of a large amount of low-correlation knowledge, and the interference of irrelevant information makes the model hallucination problem not effectively solved. We collected computer-related textbooks and knowledge documents, dividing them into knowledge document blocks according to the content structure to construct an external knowledge database. On this base, we introduced the large educational models question-answering enhancement method based on re-ranking and post-retrieval reflection, which utilized a high-performance multilingual re-ranking model based on a cross-encoder to capture deep semantic information, filter the retrieval information, filter out irrelevant information to improve the retrieval quality. The proposed method applied RAG techniques for model reflection so that the model can further enhance the quality of the model's answers through self-examination, and effectively improve the accuracy of the large language model in computer question-answering. This approach significantly improves the accuracy of large language models in computer question-answering tasks. The proposed method has been tested on several popular current generative models, achieving promising results on CS-Bench, with an approximate 5% increase in accuracy for computer question-answering tasks.

А.H. Tkachuk, A.A. Humeniuk, O.O. Dobrzhansky et al.

The article examines modern approaches to improving the energy efficiency of electric drives in the context of implementing the concept of smart energy networks (Smart Grid). Particular attention is given to the integration of electric drives as active participants in the energy balance, capable not only of consuming energy but also of adaptively regulating their operating modes in accordance with network parameters, load conditions, and the state of renewable energy sources. The study emphasizes the feasibility of using intelligent control systems that ensure high-quality regulation and reduced energy consumption under dynamically changing external conditions. Methods of energy consumption optimization based on adaptive control of variable-frequency drives are analyzed. The principles of using load forecasting algorithms are considered, enabling the formation of optimal operating profiles in advance and preventing peak overloads in the grid. The potential of regenerative operating modes, which allow excess energy during braking or speed reduction to be returned to the grid or local storage systems, is highlighted. This approach improves the overall efficiency of electric drive systems and reduces power losses. The results of simulation modeling performed in MATLAB/Simulink, using adaptive regulators and load models, confirm the effectiveness of the proposed strategies. It has been established that the application of intelligent control algorithms reduces the electricity consumption of electric drives compared to traditional control methods, increases the power factor, and decreases harmonic distortion levels in the grid by 25–30 %. Additionally, it is demonstrated that the use of adaptive regulators ensures system stability even under varying motor parameters and external disturbances. The practical implementation of such solutions is feasible in a wide range of applications: industrial production lines, electric transport systems, and integrated energy complexes with renewable sources. This opens new prospects for the development of energy-efficient Smart Grid systems with a high level of flexibility, reliability, and self-recovery capability after disturbances. The proposed approaches contribute to shaping a new paradigm of electric drives focused on minimizing energy losses and enhancing the overall efficiency of modern power systems.

Simon Biquard, Josquin Errard, Radek Stompor

We present a study of unbiased reconstruction of cosmic microwave background (CMB) polarization maps from data collected by modern ground-based observatories. Atmospheric emission is a major source of correlated noise in such experiments, complicating the recovery of faint cosmological signals. We consider estimators that require minimal assumptions about unpolarized atmospheric emission properties, instead exploiting hardware solutions commonly implemented in modern instruments, such as pairs of orthogonal antennas in each focal plane pixel, and polarization signal modulation via a continuously rotating half-wave plate (HWP). We focus on two techniques: (i) statistical down-weighting of low-frequency atmospheric signals, and (ii) pair-differencing (PD), which involves differencing signals collected by two detectors in the same focal plane pixel. We compare their performance against the idealized case where the atmospheric signal is perfectly known and cleanly subtracted. We show that PD can be derived from maximum likelihood principles under general assumptions about the atmospheric signal, optimizing map sensitivity. In the absence of instrumental systematics but with reasonable detector noise variations, PD yields polarized sky maps with noise levels only slightly worse than the ideal case. While down-weighting could match this performance, it requires highly accurate atmospheric models that are not readily available. PD performance is affected by instrumental systematics, particularly those leaking atmospheric signal to the difference time stream. However, effects like gain mismatch are efficiently mitigated by a rotating HWP, making PD a competitive, robust, and efficient solution for CMB polarization mapmaking without atmospheric modeling.

Natalia A. Boitsova, Anna A. Abelit, Daniil D. Stupin

Nowadays electrical impedance spectroscopy (EIS) has become an advanced experimental technique with a wide range of applications: from simple passive circuits diagnostics to semiconductor high-end device development and breakthrough technologies in bio-sensing. Although hardware for EIS today is well developed, the EIS analysis software is mainly custom, old fashioned, i.e. it is limited by features, does not utilize the progress in the modern computer science and hardware, and is usually implemented in close-source code or written on outdated programming languages, which causes slow progress in field of the EIS and complicates researchers attempts of development in EIS autonomous devices, such as implants. In this article, we introduce a free and open-sourced MatLab/GNU Octave package for EIS analysis called NELM, which provides powerful equipment tools for matching experimental impedance data with theoretical equivalent circuits. Our software has an user friendly interface and supports different formats of input data, fitting programs, and impedance models. In addition, we have developed NELM with implementation of the latest progress in computation science such as symbolic calculations, parallel computing, and artificial intelligence. The abilities of NELM were validated by its applications in the different fields of science, such as semiconductor studies, bioimpedance and electrochemestry, which demonstrated high-efficiency of the proposed software package and showed that it is a promising tool for solving actual problems in electronic industry, biosensorics, and healthcare technologies.

Kasidis Arunruangsirilert, Jiro Katto

Many GPUs have incorporated hardware-accelerated video encoders, which allow video encoding tasks to be offloaded from the main CPU and provide higher power efficiency. Over the years, many new video codecs such as H.265/HEVC, VP9, and AV1 were added to the latest GPU boards. Recently, the rise of live video content such as VTuber, game live-streaming, and live event broadcasts, drives the demand for high-efficiency hardware encoders in the GPUs to tackle these real-time video encoding tasks, especially at higher resolutions such as 4K/8K UHD. In this paper, RD performance, encoding speed, as well as power consumption of hardware encoders in several generations of NVIDIA, Intel GPUs as well as Qualcomm Snapdragon Mobile SoCs were evaluated and compared to the software counterparts, including the latest H.266/VVC codec, using several metrics including PSNR, SSIM, and machine-learning based VMAF. The results show that modern GPU hardware encoders can match the RD performance of software encoders in real-time encoding scenarios, and while encoding speed increased in newer hardware, there is mostly negligible RD performance improvement between hardware generations. Finally, the bitrate required for each hardware encoder to match YouTube transcoding quality was also calculated.

Sahil Kale

Modern large language models integrate web search to provide real-time answers, yet it remains unclear whether they are efficiently calibrated to use search when it is actually needed. We introduce a benchmark evaluating both the necessity and effectiveness of web access across commercial models with no access to internal states or parameters. The dataset includes a static split of 783 temporally anchored questions answerable from pre-cutoff knowledge, aimed at testing whether models invoke search based on low internal confidence, and a dynamic split of 288 post-cutoff queries designed to test whether models recognise when search is required and retrieve updated information. Web access substantially improves static accuracy for GPT-5-mini and Claude Haiku 4.5, though confidence calibration worsens. On dynamic queries, both models frequently invoke search yet remain below 70 percent accuracy due to weak query formulation. Costs per accuracy-improving call remain low, but returns diminish once initial retrieval fails. Selective invocation helps, but models become overconfident and inconsistent after search. Overall, built-in web search meaningfully improves factual accuracy and can be invoked selectively, yet models remain overconfident, skip retrieval when it is essential, and falter once initial search queries underperform. Taken together, internal web search works better as a good low-latency verification layer than a reliable analytical tool, with clear room for improvement.

G. E. Granroth, M. Daum, A. A. Aczel et al.

A modern cold triple axis spectrometer is being planned for the High Flux Isotope Reactor (HFIR) at Oak Ridge National Laboratory. Here, we describe the design of an incident beamline that will put a flux of $\sim 10^8\mathrm{\frac{n}{cm^2 s}}$ on a sample with an area of 2 cm x 2 cm. It takes current physical constraints at HFIR into account and it can accommodate both single and multiplexed analyzer-detector secondary spectrometers and large superconducting magnets. The proposed incident beamline includes a multi-channel guide with horizontal focusing, a neutron velocity selector, components to facilitate an incident beam polarization option, and a double-focusing pyrolytic graphite monochromator. This work describes the process of optimizing the guide system and monochromator and summarizes the expected performance of the incident beamline for non-polarized operation.

Yiming Zhou, Zixuan Zeng, Andi Chen et al.

Exploring the capabilities of Neural Radiance Fields (NeRF) and Gaussian-based methods in the context of 3D scene reconstruction, this study contrasts these modern approaches with traditional Simultaneous Localization and Mapping (SLAM) systems. Utilizing datasets such as Replica and ScanNet, we assess performance based on tracking accuracy, mapping fidelity, and view synthesis. Findings reveal that NeRF excels in view synthesis, offering unique capabilities in generating new perspectives from existing data, albeit at slower processing speeds. Conversely, Gaussian-based methods provide rapid processing and significant expressiveness but lack comprehensive scene completion. Enhanced by global optimization and loop closure techniques, newer methods like NICE-SLAM and SplaTAM not only surpass older frameworks such as ORB-SLAM2 in terms of robustness but also demonstrate superior performance in dynamic and complex environments. This comparative analysis bridges theoretical research with practical implications, shedding light on future developments in robust 3D scene reconstruction across various real-world applications.

Sidong Feng, Suyu Ma, Han Wang et al.

The importance of computational modeling of mobile user interfaces (UIs) is undeniable. However, these require a high-quality UI dataset. Existing datasets are often outdated, collected years ago, and are frequently noisy with mismatches in their visual representation. This presents challenges in modeling UI understanding in the wild. This paper introduces a novel approach to automatically mine UI data from Android apps, leveraging Large Language Models (LLMs) to mimic human-like exploration. To ensure dataset quality, we employ the best practices in UI noise filtering and incorporate human annotation as a final validation step. Our results demonstrate the effectiveness of LLMs-enhanced app exploration in mining more meaningful UIs, resulting in a large dataset MUD of 18k human-annotated UIs from 3.3k apps. We highlight the usefulness of MUD in two common UI modeling tasks: element detection and UI retrieval, showcasing its potential to establish a foundation for future research into high-quality, modern UIs.

Priyanka Mandikal

LLMs have revolutionized the landscape of information retrieval and knowledge dissemination. However, their application in specialized areas is often hindered by factual inaccuracies and hallucinations, especially in long-tail knowledge distributions. We explore the potential of retrieval-augmented generation (RAG) models for long-form question answering (LFQA) in a specialized knowledge domain. We present VedantaNY-10M, a dataset curated from extensive public discourses on the ancient Indian philosophy of Advaita Vedanta. We develop and benchmark a RAG model against a standard, non-RAG LLM, focusing on transcription, retrieval, and generation performance. Human evaluations by computational linguists and domain experts show that the RAG model significantly outperforms the standard model in producing factual and comprehensive responses having fewer hallucinations. In addition, a keyword-based hybrid retriever that emphasizes unique low-frequency terms further improves results. Our study provides insights into effectively integrating modern large language models with ancient knowledge systems. Project page with dataset and code: https://sites.google.com/view/vedantany-10m

Sebastian Siegel, Ming-Jay Yang, John-Paul Strachan

Processing long temporal sequences is a key challenge in deep learning. In recent years, Transformers have become state-of-the-art for this task, but suffer from excessive memory requirements due to the need to explicitly store the sequences. To address this issue, structured state-space sequential (S4) models recently emerged, offering a fixed memory state while still enabling the processing of very long sequence contexts. The recurrent linear update of the state in these models makes them highly efficient on modern graphics processing units (GPU) by unrolling the recurrence into a convolution. However, this approach demands significant memory and massively parallel computation, which is only available on the latest GPUs. In this work, we aim to bring the power of S4 models to edge hardware by significantly reducing the size and computational demand of an S4D model through quantization-aware training, even achieving ternary weights for a simple real-world task. To this end, we extend conventional quantization-aware training to tailor it for analog in-memory compute hardware. We then demonstrate the deployment of recurrent S4D kernels on memrisitve crossbar arrays, enabling their computation in an in-memory compute fashion. To our knowledge, this is the first implementation of S4 kernels on in-memory compute hardware.

Praveen Srinivasa Varadhan, Amogh Gulati, Ashwin Sankar et al.

Despite rapid advancements in TTS models, a consistent and robust human evaluation framework is still lacking. For example, MOS tests fail to differentiate between similar models, and CMOS's pairwise comparisons are time-intensive. The MUSHRA test is a promising alternative for evaluating multiple TTS systems simultaneously, but in this work we show that its reliance on matching human reference speech unduly penalises the scores of modern TTS systems that can exceed human speech quality. More specifically, we conduct a comprehensive assessment of the MUSHRA test, focusing on its sensitivity to factors such as rater variability, listener fatigue, and reference bias. Based on our extensive evaluation involving 492 human listeners across Hindi and Tamil we identify two primary shortcomings: (i) reference-matching bias, where raters are unduly influenced by the human reference, and (ii) judgement ambiguity, arising from a lack of clear fine-grained guidelines. To address these issues, we propose two refined variants of the MUSHRA test. The first variant enables fairer ratings for synthesized samples that surpass human reference quality. The second variant reduces ambiguity, as indicated by the relatively lower variance across raters. By combining these approaches, we achieve both more reliable and more fine-grained assessments. We also release MANGO, a massive dataset of 246,000 human ratings, the first-of-its-kind collection for Indian languages, aiding in analyzing human preferences and developing automatic metrics for evaluating TTS systems.

Ryleigh Landstra, Ian Winkelstern

Large concentrically laminated carbonate grains (here referred to as pisoids) have been observed sporadically throughout the geological record and in modern environments. Explanations for how these grains form have varied widely in different settings, although microbial effects are often involved. In Ore Lake, a ~1 km2 flow-through lake in southeast Michigan, one to four centimeter oblong calcite pisoids are observed in both lake bottom shallows and mounded as a small spit near the primary outflow. In section mm-scale light and more porous along with and dark and more dense concentric laminations are apparent. Here we use field observations, petrography, water chemistry, and stable isotopes to understand their formation. Measurements of pisoid calcite δ18O and lake water δ18O indicate that precipitation occurs in waters between roughly 19 – 28°C. These warm temperatures imply that pisoid growth happens almost entirely within the summer, contrary to prior work that suggested wintertime precipitation was important. Pisoid δ18O values largely overlap with coexisting lake bivalve values, suggesting that pisoid precipitation is in equilibrium. In contrast, pisoid δ13C is as much as 8 ‰ more positive than bivalve δ13C due to photosynthetic effects. We propose that the laminations in these pisoids arise from different rates of formation within the warm months, rather than large seasonal differences. A decline in lake alkalinity beginning in late spring likely coincides with more rapid growth, with slower growth mediated by cyanobacteria continuing through the summer. This range of observations enables the use of Ore Lake as a potential model for understanding pisoid formation throughout the geological record.

Eduardo Baena, Sergio Fortes, Francisco Muro et al.

The management of cellular networks, particularly within the environment rapidly advancing to 6G, presents considerable challenges due to the highly dynamic radio environment. Traditional tools such as Radio Environment Maps (REMs) have proven inadequate for real-time network changes, underlining the need for more sophisticated solutions. In response to these challenges, this work introduces a novel approach that harnesses the unprecedented power of state-of-the-art image classifiers for network management. This method involves the generation of Network Synthetic Images (NSIs), which are enriched heat maps that precisely reflect varying cellular network operating states. Created from user location traces linked with Key Performance Indicators (KPIs), NSIs are strategically designed to meet the intricate demands of 6G networks. This research delves deep into a comprehensive analysis of the diverse factors that could potentially impact the successful application of this methodology in the realm of 6G. The results from this investigation, coupled with a comparative assessment against traditional REM usage, emphasize the superior performance of this innovative method. Additionally, a case study involving an automatic network diagnosis scenario validates the effectiveness of this approach. The findings reveal that a generic Convolutional Neural Network (CNN), one of the most powerful tools in the arsenal of modern image classifiers, delivers enhanced performance, even with a reduced demand for positioning accuracy. This contributes significantly to the real-time, robust management of cellular networks as we transition into the era of 6G.

Frank Uhlig

We assess the situation of our elementary Linear Algebra classes in the US holistically and through personal history recollections. Possible remedies for our elementary Linear Algebra's teaching problems are discussed and a change from abstract algebraic taught classes to a concrete matrix based first course is considered. The challenges of such modernization attempts for this course are laid out in light of our increased after-Covid use of e-books and e-primers. We specifically address the useless and needless, but ubiquitous use of determinants, characteristic polynomials and polynomial root finding methods that are propagated in our elementary text books and are used in the majority of our elementary Linear Algebra classes for the matrix eigenvalue problem but that have no practical use whatsoever and offer no solution for finding matrix eigenvalues. This paper challenges all mathematicians as we have misinformed and miseducated our students badly for decades in elementary Linear Algebra now and urges a switch to a new, fully matrix theoretical approach that covers all classical subjects in a practical and computable way.