Computer Science Review

Deepfake detection has advanced rapidly alongside progress in generative adversarial networks, diffusion models, neural rendering, and multimodal synthesis. However, strong performance on curated benchmark datasets does not necessarily translate into reliable performance in real-world environments, where manipulated media are affected by platform compression, reposting, post-processing, multimodal inconsistency, temporal drift, and evolving generation pipelines. This review examines deepfake detection from the perspective of in-the-wild robustness, multimodal reasoning, and cross-domain generalization. It first clarifies the conceptual foundations of in-the-wild and generalizable deepfake detection, then reviews the data ecosystem, including benchmark datasets, real-world media conditions, multimodal data gaps, annotation quality, and benchmark aging. The review further analyzes the major causes of detector failure under realistic deployment conditions, including generator shift, platform-induced degradation, modality mismatch, language and speaker variation, and broader contextual diversity. It then compares major methodological families, including artifact-based methods, deep representation-based approaches, transfer-based learning, domain adaptation, domain generalization, and multimodal detection strategies, with emphasis on their relevance to real-world robustness. Finally, the review discusses evaluation practices, external validation, realistic benchmarking, and future directions such as continually updated benchmarks, missing-modality-robust systems, uncertainty-aware detection, provenance integration, and foundation-model-based approaches. Overall, the review argues that reliable deepfake detection requires a shift from benchmark-centered optimization toward robust, multimodal, and deployment-oriented evaluation and design.

The proliferation of agentic AI is generating an energy consumption crisis that neither incremental hardware improvements nor renewable energy procurement alone can resolve. This survey argues that the most viable pathway toward sustainable autonomous intelligence lies not in making Large Language Models (LLMs) greener, but in fundamentally rethinking the unit of agency: replacing monolithic frontier models with Small Language Models (SLMs) that are purpose-designed for the constrained, repetitive, and specialized task profiles that characterize the vast majority of deployed agentic workloads. A structured review of the small-model landscape collates published evidence that 4-bit quantized models in the 1B–7B range approach LLM-scale performance on several classes of agentic sub-tasks, with energy-per-token reductions of about one order of magnitude reported by individual studies for the specific model pairs, hardware, and workloads they evaluate, while unlocking edge deployment on sub-watt neuromorphic and embedded hardware that frontier models cannot approach. Multiple application domains are surveyed, open research challenges are identified across the agentic stack, and the concept of intelligence-per-watt (IPW) is introduced as the central organizing metric for sustainable autonomous intelligence research. This comprehensive survey provides a conceptual analysis of the literature, utilizing order-of-magnitude estimations based on published data, without conducting original physical measurements.

Large language model (LLM) watermarking is a prominent approach for identifying machine-generated text, supporting provenance, protecting intellectual property, and enabling accountability in generative-AI ecosystems. Proposed mechanisms include token- and logit-based sampling, semantic and syntactic watermarking, post-hoc methods, publicly verifiable schemes, model- and service-level watermarking, and retrieval-augmented or application-specific approaches. However, their practical reliability under realistic transformations, adaptive attacks, limited model access, diverse deployment environments, and high-stakes governance settings remains uncertain. This article presents an evidence-centered, PRISMA-guided systematic literature review based on searches completed on 14 May 2026. Rather than only classifying mechanisms, we assess how convincingly existing studies support their reliability and robustness claims. We identified 1367 records, retained 1065 after duplicate removal, assessed 207 full texts, and included 172 studies. The synthesis covers six themes: watermark embedding paradigms; threat models and attacker assumptions; evaluation metrics and experimental design; robustness under post-generation transformations; deployment readiness and access assumptions; and the limited evidence on application, governance, and misuse contexts. The field is methodologically rich but evidentially fragmented: many studies report strong detection performance while relying on narrow datasets, inconsistent metrics, limited baselines, implicit threat models, unrealistic transformations, or access assumptions unsuitable for closed or API-based deployments. Recurring concerns include false-positive control, public verification, adaptive paraphrasing, mixed-origin documents, benchmark fragmentation, and reproducibility. We therefore provide a descriptive evidence-completeness assessment and a research agenda for more robust, transparent, and deployment-ready LLM watermarking systems.

Large language models (LLMs) are increasingly being used in network operations (NetOps) and artificial intelligence for IT operations (AIOps) for tasks ranging from telemetry retrieval and incident diagnosis to configuration planning and bounded remediation. As these systems acquire greater access to operational tools, a central question is whether operational assurance develops commensurately with the authority granted to them. This survey examines that question through a structured, evidence-stratified review of agentic NetOps and AIOps, spanning directly evaluated LLM systems, operational mechanisms, transferred evidence, and normative guidance. We organise the field around autonomy, tool scope, evidence traces, assurance controls, evaluation, security, and governance, and introduce an operational assurance contract that links each autonomy level to permitted tools, required evidence, independent gates, execution budgets, rollout and rollback duties, and audit requirements. The synthesis indicates a capability-assurance gap: evidence is comparatively strong for read-oriented assistance and tool-grounded diagnosis, but becomes substantially less complete as systems approach configuration change, bounded execution, and closed-loop operation. We therefore develop a workflow-level evaluation framework covering evidence quality, tool use, policy and invariant compliance, staged execution, recovery, calibration, cost, and human intervention, together with a threat model for prompt-borne attacks, compromised operational evidence, excessive agency, privilege boundaries, and auditability. The resulting view treats agentic NetOps and AIOps as constrained operational control, in which increasing autonomy requires correspondingly stronger evidence, independently enforced safeguards, and recovery mechanisms.

Grey sheep users (GSUs), characterized by weak or unstable alignment with the broader user population, remain a challenge in collaborative filtering-based recommender systems. Despite extensive efforts devoted to their detection and treatment, a clear and consistent understanding of GSUs is still lacking. Existing studies typically adopt model-dependent definitions and often assume that GSUs necessarily reduce recommendation quality. This paper presents a systematic and critical review of GSUs, covering their definition, detection strategies, modeling paradigms, and evaluation practices. We revisit the classical similarity-based interpretation and show that the effect of GSUs depends on model design rather than being inherently harmful. Based on this observation, we introduce a refined conceptual perspective that describes GSUs as users with consistently misaligned preference patterns, and further propose a dynamic view that distinguishes between short-term irregular behavior and stable preference differences. We then provide a comprehensive taxonomy of existing approaches, including detection-dependent methods and detection-free strategies, and analyze their underlying assumptions and limitations. We further show that the form in which a detection result is expressed constrains which handling strategies are available, and that this correspondence is rarely exploited in practice. A key finding of this study is that current research is limited by weak evaluation practices. In particular, the absence of ground truth, inconsistent definitions, and dependence on aggregate performance metrics lead to circular reasoning and unreliable conclusions. Finally, we outline future research directions toward more robust modeling frameworks and user-centered evaluation protocols, with the aim of establishing a clearer conceptual foundation and guiding the development of more reliable and inclusive recommender systems.

This paper presents a comprehensive review of the existing literature on the key phenomena influencing prehospital emergency service (PES) demand, PES demand prediction, and ambulance positioning and routing. The identification and understanding of these factors are essential for the intelligent management of PES and informed decision-making. Effective management depends on the accuracy of emergency call prediction models and the strategies used to optimize emergency vehicle routing and standby locations, both of which are examined and discussed in this paper. The findings of this study could serve as a foundation for further research and practical advancements in response to the increasing demand for PES.

Large language models (LLMs) can turn task descriptions into executable algorithms. Coupled with automatic evaluation, this creates an appealing end-user workflow: an engineer can state a computational problem and guide candidate design without first becoming an expert in the relevant algorithmic field. The practical question is what information from earlier generations should be returned to the model. We investigate this question with EASE (Effortless Algorithmic Solution Evolution), using problem-specific black-box optimizer generation as a controlled case study. Twelve strategies vary whether the prompt includes recent algorithms, high-performing algorithms, their combinations, or no earlier outputs. The strategies are evaluated with Gemini 3 Pro, GPT-5.2, and Claude Opus 4.6 on a difficult 30-dimensional GNBG-II benchmark instance. The observed ranking of individual strategies differs across model configurations, but analysis of the underlying context factors reveals a more consistent descriptive pattern. Combining the most recent candidate with the best-performing candidate yields lower mean repetition-best scores than using either component alone for all three evaluated models. In contrast, increasing the amount of recent or performance-selected context does not show a consistent additional benefit across models. The results therefore favor selective context construction over simply maximizing the amount of previous information supplied to the model. The largest improvements also tend to occur during the early generations, while substantial variation remains across repetitions and model configurations. Because the experiment is restricted to a single benchmark instance and five independent repetitions per model–strategy cell, the comparisons are interpreted descriptively and do not establish a universal context strategy or transfer across optimization problems. Within this controlled setting, however, the results indicate that recency and performance-based selection can provide complementary feedback signals, whereas larger contexts are not consistently advantageous. An interactive result explorer is available at https://tbu-ailab.github.io/PaperContext/ ; generated data and source files are available at https://github.com/TBU-AILab/PaperContext .

The Internet of Things (IoT) has become integral to modern society, yet ensuring secure, trustworthy, and accountable communication remains a critical challenge. Non-repudiation is a foundational security service that provides verifiable evidence of origin, receipt, integrity, and action, making it essential for trust establishment, auditability, dispute resolution, and forensic analysis in IoT environments. However, existing reviews have primarily focused on broader IoT security requirements, such as confidentiality, integrity, authentication, privacy, and access control, with limited systematic attention given to non-repudiation mechanisms. This paper presents a systematic literature review of non-repudiation mechanisms in IoT-based systems published between 2015 and 2026. Following PRISMA guidelines and applying rigorous inclusion and exclusion criteria, we identified 41 relevant studies. The reviewed schemes are classified into three main categories: digital signature-based schemes, blockchain-based schemes, and hybrid schemes. Digital signature-based schemes are further divided into classical signature-based mechanisms, including PKI-based, identity-based, certificateless/certificateless aggregate, and signcryption-based schemes, and post-quantum signature-based schemes. Blockchain-based schemes rely on ledger immutability, distributed consensus, and smart-contract execution to preserve tamper-evident evidence, while hybrid schemes combine digital signatures with blockchain-based evidence preservation, including blockchain with classical signature schemes and blockchain with post-quantum signature schemes. The review analyses application domains, security properties, performance evaluation practices, and key trade-offs, including scalability, resource constraints, communication and storage overhead, latency, gas cost, and centralised trust dependencies. The findings show a growing shift from standalone signer-binding mechanisms towards hybrid, auditable, and quantum-resilient evidence-preservation architectures. Future research should prioritise lightweight, formally verified, privacy-preserving, and post-quantum-resilient non-repudiation mechanisms tailored to the resource constraints, trust assumptions, and domain-specific requirements of modern IoT ecosystems.

Machine learning (ML) has become an increasingly important tool in the petroleum industry due to its ability to handle complex, high-dimensional data and support predictive decision-making. While numerous studies and reviews have examined ML applications in upstream petroleum engineering, a focused review of ML methods applied specifically to petroleum production engineering remains limited. This review examines ML applications in core production engineering workflows, including hydrocarbon production forecasting, well performance evaluation, flow assurance, artificial lift systems, and well stimulation, for both conventional and unconventional reservoirs. The surveyed literature spans more than twenty countries worldwide; however, the majority of studies originate from China and the United States, followed by Russia, Saudi Arabia, Brazil, Iran, Egypt, and Nigeria. Topics such as drilling operations, hydraulic fracturing design, carbon capture and storage, hydrogen energy, and economic or price forecasting are outside the scope of this review. The reviewed studies span a wide range of ML approaches, from traditional regression and ensemble methods to deep learning architectures and emerging physics-informed neural networks. Rather than presenting performance metrics in isolation, this review discusses where ML methods have shown practical advantages and their limitations, including overfitting, data quality constraints, limited generalizability, and challenges related to physical consistency and operational deployment. By consolidating recent advances and reported challenges, this review aims to provide a balanced perspective on the current capabilities of ML in petroleum production engineering and to identify areas where further methodological development and integration with domain knowledge are required for reliable field-scale applications.

Quantum-inspired Machine Learning (QiML) is a burgeoning field, receiving global attention from researchers for its potential to leverage principles of quantum mechanics within classical computational frameworks. However, current review literature often presents a superficial exploration of QiML, focusing instead on the broader Quantum Machine Learning (QML) field. In response to this gap, this survey provides an integrated and comprehensive examination of QiML, exploring QiML's diverse research domains including tensor network simulations, dequantized algorithms, and others, showcasing recent advancements, practical applications, and illuminating potential future research avenues. Further, a concrete definition of QiML is established by analyzing various prior interpretations of the term and their inherent ambiguities. As QiML continues to evolve, we anticipate a wealth of future developments drawing from quantum mechanics, quantum computing, and classical machine learning, enriching the field further. This survey serves as a guide for researchers and practitioners alike, providing a holistic understanding of QiML's current landscape and future directions.