The convergence of artificial intelligence (AI) and blockchain is transforming the Fintech sector by offering innovative solutions in security, regulatory compliance, and operational transparency. This systematic mapping study analyzed 667 studies published between 2010 and 2024, selecting 175 relevant articles. The findings show that these technologies optimize critical processes such as Know Your Customer (KYC) and fraud prevention, though they face limitations in scalability, interoperability, and regulatory adaptation. This work provides a comprehensive perspective on their current and future impact in Fintech. Objective: Analyze the impact of AI and blockchain on Fintech platform security, evaluating benefits, limitations, and opportunities to improve critical processes and promote industry adoption. Method: This systematic literature review used five major academic databases: IEEE Xplore, ACM Digital Library, Taylor and Francis Online, Wiley Online Library, and Emerald Insight. From 667 studies, 175 met the inclusion criteria. Results: The integration of AI and blockchain significantly improves the security and efficiency of key processes such as KYC and fraud detection. However, barriers in scalability, interoperability, and regulation persist, limiting widespread implementation.
The sustainable management of Waste Electrical and Electronic Equipment (WEEE) is a critical global challenge, particularly in contexts with limited data. This study proposes a predictive model based on artificial neural networks, developed from surveys and historical records in the city of Guayaquil, with the aim of estimating WEEE generation on annual and monthly scales. The model was structured in phases of data collection, preprocessing, training, and validation, integrating sociodemographic variables and categories of discarded devices. To ensure reliability, a multi-technique validation protocol was applied, including Hold-Out, Stratified K-Fold, and Bootstrap Sampling methods. Results showed strong performance, with a coefficient of determination (R²) of 0.9125 in initial tests, an average of 0.9097 in cross-validation, and up to 0.9789 with bootstrap, significantly outperforming traditional linear regression methods. These findings confirm the model’s ability to capture non-linear relationships and produce accurate forecasts in data-scarce environments. It is concluded that neural networks represent an effective tool to support strategic planning and decision-making in sustainable WEEE management, providing a replicable framework for other regions facing similar challenges.
Activation functions are fundamental components of modern neural networks, including Large Language Models (LLMs). Nonlinear activations regulate the flow of information in most models and determine how data is processed. However, training or even fine-tuning very large and complex models with such activations, like those used in ChatGPT and DeepSeek, remains out of reach for many researchers. For this reason, traditional architectures like LSTM remain relevant for moderate or small-scale applications. In this context, the choice of appropriate functions can significantly influence a network's ability to learn complex patterns under limited computing resources. In this work, we analyze the behavior of a novel family of activation functions, called Zorro, which can serve as gating functions in LSTM architectures. We propose replacing traditional activation and gating functions in LSTMs with Zorro functions to improve model performance and convergence speed. Unlike conventional approaches, our method assigns a different function to each gate or activation, enabling the application of the same methodology to any gated architecture. We evaluate the modified LSTM models on widely used small-scale benchmark datasets, including Japanese Vowels and Human Activity Recognition for classification, as well as Chickenpox and Turbofan Degradation for regression. The results show that our method improves model accuracy by up to 10% and reduces training time by up to 15%.
Technology and communication are fundamental in modern enterprise environments, making network performance evaluation crucial. To analyze their impact on enterprise infrastructure performance, this study compares EoIP and VxLAN, two technologies used for Layer 2 network extension over IP. The PPDIOO methodology was employed to structure and optimize the network analysis and improvement process. The implementation included configuring MikroTik devices running RouterOS, playing a key role in the topologies used. Tests on bandwidth, latency, and transfer rate were conducted in simulated environments using hping, iperf, and Wireshark to assess performance. The results indicate that VXLAN provides better latency performance, whereas EoIP offers lower overhead and greater configuration simplicity. These findings highlight the importance of considering various factors when choosing between these technologies, ensuring that the implementation aligns with the specific requirements of each enterprise environment. This study provides key insights for network administrators and IT architects seeking to optimize their organization's infrastructure through advanced segmentation and network virtualization technologies.
This study explores housing price prediction in six major cities of Andalusia using a combination of traditional and machine learning models. Online property listings were compiled to build a detailed dataset incorporating structural, socioeconomic, and locational variables. A traditional hedonic model was estimated using Ordinary Least Squares (OLS), and its performance was compared with that of Random Forest, K-Nearest Neighbours, and XGBoost. Results show that linear models tend to underperform, highlighting the relevance of non-linear approaches in capturing the complex dynamics of the housing market. To enhance interpretability, SHapley Additive exPlanations (SHAP) values were applied to the XGBoost model, enabling the identification of the most influential predictors in each city. While structural variables remained important across all models, socioeconomic and locational features exhibited varying effects depending on the city. These findings underscore the potential of combining rich data sources with interpretable machine learning techniques for urban housing market analysis.
The Journal of Computer Science and Technology(JCST),the first English-language academic journal in China's computing field,was founded in 1986 by Professor Xia Peisu at the Institute of Computing Technology(ICT),Chinese Academy of Sciences.Over the past four decades,JCST has sustained its momentum by adher-ing to its mission of publishing peer-reviewed research and development results in computer science and technol-ogy worldwide,with a particular emphasis on emerging international research frontiers and high-quality re-search from China.
Artificial intelligence (AI) is reshaping medical informatics from a discipline of data management into a science of integration, inference, and translation. As biomedical data proliferate across physiological, clinical, and molecular domains, AI functions as the integrative engine that transforms complexity into actionable understanding. In this survey, we synthesize recent advances spanning data representation, algorithmic innovation, and clinical deployment, emphasizing the transition from isolated tasks to cohesive systems that link discovery and care. We highlight how advances in medical AI algorithms across clinical data, medical imaging, and multi-omics are beginning to converge with applications in clinical diagnosis, drug discovery, precision medicine, and surgery. Looking ahead, medical AI is moving toward a self-reflective and collaborative paradigm, where progress in multi-modality, trustworthiness, human-machine synergy, and ethical reasoning may allow intelligence to be woven into the pipeline of clinical practice and fulfill its translational promise.
As conventional CPU-based security architectures struggle to scale with ever-growing network bandwidths and increasingly sophisticated cyberattacks,the data processing unit(DPU),a specialized processor for datacenter infrastructure,has emerged as a transformative foundation for secure and high-performance com-puting.Unlike prior fragmented studies,this work proposes a comprehensive security framework for DPUs by systematically investigating the DPUs'dual role in cybersecurity,serving both as an active security enforcer and as a critical component that must itself be protected.First,the framework offloads security policies onto the DPU to enable line-rate packet inspection and hardware-accelerated security processing.Second,the framework re-architects the DPU itself to defend against physical and architectural attacks,acknowledging that the DPU al-so introduces a new attack surface.We validate these two design directions through two representative case stud-ies,demonstrating the effectiveness and practicality of the proposed DPU security framework.Experimental re-sults show that the proposed framework reduces remote direct memory access(RDMA)cache side-channel detec-tion latency by up to 98.7％compared with the state-of-the-art,while enabling a trusted execution environment on field-programmable gate array(FPGA)-based DPUs with sub-100 ns overhead and less than 4％FPGA re-source consumption.
Foundation models (FMs) make possible a new software paradigm in which users express what they want to accomplish rather than how to operate an application. We call this paradigm “intent-oriented software”. Although recent FM capabilities suggest the feasibility of intent-oriented software, the concept itself has not been formally articulated or systematized. This article provides the first definition of intent-oriented software and examines how such systems can be engineered in practice. We propose two complementary architectural models for realizing intent-oriented software. 1) A plug-in model. Intent-oriented software operates as a middleware layer that wraps around an existing operation-oriented application. It interprets user intents and translates them into the application’s existing UI operations without modifying the underlying operation-oriented software. 2) A native model. FMs interpret user intents and generate the necessary backend actions (e.g., API calls, function executions) directly without going through traditional UI. Both models aim to fundamentally transform the interaction paradigm by decoupling user intents from operational procedures, enabling users to accomplish goals without requiring procedural knowledge of application interfaces, while preserving the full functional expressiveness of underlying systems. We systemize key technical challenges, including intent understanding, execution planning, and error recovery, and identify research opportunities in software engineering to advance the development of robust, trustworthy, and user-centered intent-oriented software.
Fully automated processor design has recently gained significant popularity due to its fast convergence speed and reduced human costs. However, automated design remains challenging in processor correctness and performance guarantee. In this article, we introduce a series of processor auto-design methods based on a data-driven method, Binary Speculative Diagram (BSD), emphasizing how they guarantee design correctness and improve the auto-designed processor performance. Auto-designed by BSD, QiMeng-CPU-v1, an industrial-scale RISC-V CPU, achieves up to 99.999 999 999 99% accuracy. Auto-designed by State-BSD, QiMeng-CPU-v2 is comparable to ARM Cortex A53 (2010s CPU), a human-designed superscalar processor. Finally, we discuss potential future directions for extending and improving the proposed design methods toward more generalized automated processor architectures.
As one typical AI-for-Science application, machine learning interatomic potentials (MLIPs) have revolutionized the representation of potential energy surfaces. MLIPs can be categorized into specialized MLIPs, which prioritize high accuracy for specific systems, and pretrained MLIPs, which emphasize generalizability across chemical spaces. Specialized MLIPs and pretrained MLIPs differ in the dataset to be trained, model capability (parameters), the training workflow, and the workload in molecular dynamics. We review different high-performance computing (HPC) optimization techniques for training and inference that specialized MLIPs and pretrained MLIPs tend to prefer. For example, from the perspective of the training dataset, we investigate the load balance strategies, which are critical for pretrained MLIPs to enhance scalability. From the perspective of model parameters, we indicate that specialized MLIPs can benefit from curvature-aware optimization algorithms given their moderate model size. We remark that advances in HPC are not merely engineering improvements but play a key role in faster iteration of MLIPs, broader applicability, and sustained progress in MLIP development.
Privacy has rapidly become a major concern/design consideration. Homomorphic encryption (HE) and garbled circuits (GC) are privacy-preserving techniques that support computations on encrypted data. HE and GC can complement each other, as HE is more efficient for linear operations, while GC is more effective for non-linear operations. Together, they enable complex computing tasks, such as machine learning, to be performed exactly on ciphertexts. However, HE and GC introduce two major bottlenecks: an elevated computational overhead and high data transfer costs. This paper presents Privacy Preserving In-Memory Computing Engine (PPIMCE), an in-memory computing (IMC) fabric designed to mitigate both computational overhead and data transfer issues. Through the use of multiple IMC cores for high parallelism, and by leveraging in-SRAM IMC for data management, PPIMCE offers a compact, energy-efficient solution for accelerating HE and GC. PPIMCE achieves a 107x speedup against a CPU implementation of GC. Additionally, PPIMCE achieves a 1 500x and 800x speedup compared with CPU and GPU implementations of CKKS-based HE multiplications. For privacy-preserving machine learning inference, PPIMCE attains a 1 000x speedup compared with CPU and a 12x speedup against CraterLake, the state-of-art privacy preserving computation accelerator.
Despite the remarkable breakthroughs in deep neural networks (DNNs), the deployment of deep models in high-stakes, safety-critical applications remains a significant challenge. For trustworthy machine learning systems, the ability to provide reliable and well-calibrated uncertainty estimates is fundamental. However, DNNs are notoriously overconfident, often yielding miscalibrated probabilities that hinder their integration into real-world decision-making tasks. This survey provides a comprehensive review of the recent advancements in the field of uncertainty calibration in deep learning. It begins by introducing a range of advanced methods recently proposed for calibrating uncertainty in DNNs. To this end, we organize them into four primary paradigms: train-time regularization, post-hoc adjustments, Bayesian and ensemble neural networks, and hybrid approaches. Then, we transition to emerging research challenges that have gained significant attention in recent years, specifically focusing on calibration challenges in out-of-distribution (OOD) scenarios, uncertainty quantification of generative models, calibration in multimodal learning, and human-AI collaboration setting. Finally, we structure our exploration of uncertainty calibration in large language models (LLMs) around three fundamental research questions: how LLMs express uncertainty, how to evaluate their confidence, and how to effectively calibrate LLMs. By reviewing these diverse perspectives, this paper aims to serve as a holistic roadmap for researchers and practitioners aiming to bridge the gap between predictive performance and model trustworthiness.
Modern parallel and distributed computing systems are becoming increasingly complex as applications in high-performance computing (HPC) and artificial intelligence (AI) demand ever-greater levels of computation and communication efficiency. To address these demands, recent architectures integrate heterogeneous computing devices, such as CPUs, GPUs, and DPUs (or SmartNICs), within a single compute node, forming what we refer to as multi-rail heterogeneity. This trend offers substantial potential for scalability and performance but also amplifies the challenges of data movement, synchronization, and coordination across heterogeneous components. We propose Heterogeneity-Enriched Communication (HEC) as a new paradigm that embraces multi-rail heterogeneity by accurately analyzing communication primitives, adaptively composing multi-rail strategies, and scalably optimizing end-to-end pipelines. Through three representative case studies, including HCCL (collective communication), TrimEC (multi-rail erasure coding), and DPU-KV (edge data services), we demonstrate that HEC improves efficiency, scalability, and resilience in parallel and distributed systems for AI workloads. We envision HEC as a foundation for the next generation of AI infrastructure, harmonizing heterogeneous computing instruments into a symphony of scalable and efficient systems tailored for the emerging AI era.
In intelligent transportation systems (ITS), millimeter-wave (MMW) radar-camera fusion has emerged as a cost-effective and viable solution due to low sensor prices. However, deploying such a fusion system in practical far-range scenes faces significant challenges in both sensor calibration and fusion processes. To address these challenges, this paper presents a systematic study from theoretical analysis to practical system deployment. First, we review the status quo of radar-camera fusion systems, comparing existing calibration and fusion paradigms carefully. Through comparative analysis, we find that though feature-level fusion is popular in related research, the target-level fusion is more practical for roadside applications because it is computationally efficient and more robust to depth ambiguity. Second, we introduce an automatic radar-camera calibration and fusion system for real-world traffic perception. This system implements a trajectory-based calibration scheme for spatio-temporal synchronization, specifically tackling the difficulty of identifying distinguishable calibration targets in far-range environments. After calibration, this system applies a robust two-stage target-level fusion method to achieve effective radar-camera fusion in traffic scenes. Finally, we introduce the promising advancements of the proposed system and discuss several open challenges for large-scale and high-safety commercialization. We believe physics-aware self-supervised learning, cooperative perception across roadside devices, and end-to-end perception foundation models are important for future traffic perception systems.
Recent advancements in deep learning, high-fidelity simulation, and robotic hardware have propelled significant progress in Physical Artificial Intelligence (AI). This field marks a revolutionary step in the evolution of AI by combining the precision of physical laws with the adaptability of machine learning. In this paper, we review the development of Physical AI and its taxonomy by examining the relevant literature, categorizing it into three sub-domains: Physical-Informed AI, Generative Physical AI, and Embodied AI. These sub-domains primarily tackle scientific and engineering challenges, create physics-plausible scenarios, and enable robots or autonomous vehicles to interact with the physical world. This approach also addresses the questions of how to perceive, generate, and interact with the physical world by integrating physics with AI algorithms. Additionally, we discuss related benchmarks and datasets. Finally, we outline the current challenges and propose potential opportunities for future research.
Large language models (LLMs) have achieved remarkable progress in natural language processing, but their immense scale leads to significant computational and storage overheads, limiting their deployment and widespread application in resource-constrained environments. Model quantization, as an effective model compression technique, significantly reduces LLMs’ memory footprint and computational requirements by lowering the numerical precision of model parameters and/or activations, while striving to maintain minimal performance loss. This survey aims to comprehensively review the latest advancements in LLM quantization, covering various techniques from the pretraining phase to the inference phase. We will delve into state of the art quantization during pretraining, post-training quantization, and quantization-aware training in quantization fine-tuning, and various quantization methods during inference. Through in-depth analysis of these methods, this survey seeks to provide researchers and engineers with a comprehensive understanding of LLM quantization techniques to identify future research directions and offers an insight of how to generate high performance low-precision kernels in different chips.
Quantum computing is an emerging paradigm that leverages quantum mechanics to solve problems difficult for classical computing. Realizing reliable quantum advantage requires not only hardware, but also a full software stack ranging from algorithms to operating systems. This paper provides a structured review of quantum computing systems and software, examining the current state and future directions of the field. We first introduce the architectures of quantum computing using superconducting and neutral atom systems as examples. At the software level, we begin by analyzing potential quantum applications, including physical simulation, optimization, and artificial intelligence. We then review current quantum programming interfaces, including software development kits (SDKs), verification, and program repair methods. Following these interfaces, this review introduces quantum compilation passes for quantum programs, such as mapping, decomposition, and noise mitigation, that transform logical algorithms into efficient, hardware-executable instructions. At the system level, we explore quantum operating system services such as scheduling, calibration, and error correction, which directly control quantum devices. Then, we present quantitative comparisons of gate count, circuit depth, and execution latency across various software-level methods. Finally, we discuss major challenges of quantum hardware and software, including high error rates, low operation speeds, and limited scalability. By integrating research across these levels, this review provides a comprehensive overview of quantum systems and software architectures, and highlights the significance of co-design between hardware and software.
Large language models (LLMs) have demonstrated remarkable generalization capabilities across diverse domains, largely attributed to the availability of massive amounts of high-quality training data. Recently, the development paradigm of LLMs has been shifting from a model-centric to a data-centric perspective. In this paper, we provide a comprehensive survey of data preparation algorithms and workflows for LLMs, categorized into three stages: pre-training, continual pre-training, and post-training. We further summarize widely used datasets along with their associated data preparation method, offering a practical reference for researchers who may lack extensive experience in the field of data preparation. Finally, we outline potential directions for future work, highlighting open challenges and opportunities in advancing data preparation for LLMs.
This paper delves into the formation of the Church-Turing Thesis, aiming to elucidate philosophical and historical issues of the nature of computation. As a cornerstone of computer science, the Church-Turing Thesis asserts that the Turing machine is the most universal computing device. This paper explores Turing’s pivotal role in the process. The Church-Turing Thesis is a conviction supported by the fact that various computational mechanisms, including the general recursive function, λ-calculus, and the Post systems among others, have been proven mathematically equivalent. A stronger formulation of the thesis implies that “intelligence is reducible to Turing machines”, further prompting discussions on the nature of artificial intelligence. This thesis has led the theory of computation to bridge across mathematics and physics, and moreover, the most enduring dichotomy in philosophy, mind and matter.
