Application gaps in Computer Science

Representative open questions

Showing 30 of 237 — one per source paper, highest-quality first.

• Federated learning for privacy-preserving skin cancer classification using deep neural networks (2026) · doi

  The proposed method outperforms state-of-the-art methods on the ISIC 2018 dataset, but its performance on other datasets (ISIC 2019 and PH2) is not compared.

• Turbulence closure in Reynolds-averaged Navier–Stokes and flow inference around a cylinder using physics-informed neural networks and sparse experimental data (2026) · doi

  The hybrid implicit/explicit coupling strategy uses the baseline Spalart–Allmaras model to provide baseline eddy viscosity while the neural network predicts residual Reynolds-force corrections; the applicability and performance of this hybrid approach with alternative baseline turbulence models (k-ε, k-ω, RSM) remains unexplored.

• FPGA-Enabled Machine Learning Applications in Earth Observation: A Systematic Review (2026) · doi

  Data compression and disaster response remain unexplored onboard use cases for FPGA-enabled ML in Earth observation, despite their relevance to the NewSpace era. Lightweight FPGA implementations of state-of-the-art compression methods for SmallSat imaging payloads generating ~640 Mbps data rates present a critical research opportunity to enable real-time alerts for wildfires and post-disaster damage assessments.

• A Machine Learning Perspective on FinTech-Driven Inclusion: Addressing Algorithm Bias in Credit Scoring Systems in Developing Economies (2026) · doi

  The model fit indices (CFI = 0.943, TLI = 0.927, RMSEA = 0.058) demonstrate theoretical framework validity, but the paper does not address how explainability techniques (LIME, SHAP, attention mechanisms) should be implemented and evaluated within credit scoring algorithms to maintain the measured fairness-trust relationships when deployed across different cultural and regulatory contexts in developing economies.

• Prediction of sedimentation concentration profiles in inclined suspension systems: A data-driven neural network framework (2026) · doi

  The study reconstructed concentration profiles for inclined suspensions using TARG (time-averaged radiography) measurements, but the framework has not been extended to predict behavior under dynamic conditions such as flow reversal, oscillatory motion, or time-varying inclination angles relevant to drilling operations.

• Ento-Linguistics: Language, Ambiguity, and Scientific Communication in Entomology: How Terminology Networks Shape Understanding of Insect Biology (And Vice-Versa) (2026) · doi

  The discursive framing network function F(D,T) and cross-domain mapping strength M_ij are introduced with mathematical notation but lack empirical case studies demonstrating how terminology networks reshape understanding across the six ento-linguistic domains; specific examples of domain-dependent ambiguity propagation are absent.

• Perineural invasion in solid tumors: biological foundations and the emerging integration of machine learning and artificial intelligence (2026) · doi

  Machine learning and artificial intelligence integration for automated perineural invasion detection in histopathology specimens is mentioned as emerging, but no specific deep learning architectures, training datasets, or validation metrics comparing AI-based detection to manual assessment across multiple tumor types (pancreatic, head and neck, prostate, cutaneous) are detailed in current literature.

• Large Language Models for Combinatorial Optimization: A Systematic Review (2026) · doi

  The hybrid integration of Linear Programming dependency graphs with LLMs for multi-robot task planning (reference [157]) has only been demonstrated for specific robotics scenarios; generalization to heterogeneous robot teams with dynamic constraints and real-time replanning needs evaluation.

• Low-sample supervised fault diagnosis for fixed-wing UAVs based on multi-scale adaptive state-aware sequence learning (2026) · doi

  While the model achieves 0.000510 seconds inference time on July 21 flight data, the relationship between sample size (90 to 1440 samples tested) and computational efficiency on resource-constrained UAV autopilot systems has not been characterized. Specific investigation is needed on inference latency, memory footprint, and real-time processing constraints for embedded deployment on fixed-wing UAV platforms.

• Research on a strongly generalizable fault diagnosis method based on adversarial transfer learning (2026) · doi

  Cross-type and cross-power level fault diagnosis (Table 5) showed significantly degraded HDAL performance (88.642% accuracy with 1.983% standard deviation) compared to same-type transfer (91.067%), yet the paper does not analyze which fault classes or reactor parameter combinations are most challenging across different reactor types and power levels. Specific investigation into how reactor type differences and variable power levels affect feature transferability in the HDAL adversarial domain adaptation framework is needed.

• El marco regulatorio europeo de la inteligencia artificial y su impacto en el sistema judicial español. (2026) · doi

  The paper notes that secure access points (puntos de acceso seguro) enabling judicial immediacy through videoconference require technical guarantees and standardization, but the specific cybersecurity protocols, data protection measures, and real-time judicial authentication systems needed for telematics-based judicial communication are not detailed or validated across Spanish judicial organs.

• Neural Network Tools in the Arsenal of a University Teacher (2026) · doi

  While the authors recommend combining AI and traditional methodologies for literature search, no specific guidance is provided on how to strategically deploy different tools according to particular research tasks, nor are there validated protocols for when university teachers should prioritize AI-based tools versus traditional databases based on research question characteristics.

• Securing Fog-assisted IoT: An Adaptable and Efficient Threat Identification Approach (2026) · doi

  While three use-cases (smart healthcare, Industrial IoT, autonomous vehicles) are mentioned conceptually, the DEL framework has not been implemented or validated in these specific real-world fog-IoT environments. Practical deployment validation with actual patient data flows, production system traffic patterns, and vehicle sensor data streams is necessary.

• A Robust Hybrid Deep Learning Model for Multiclass Depression Classification from Speech Audio (2026) · doi

  Inference latency, computational cost, and real-time deployment feasibility were not empirically evaluated for the hybrid deep learning models; future research must benchmark these metrics to assess practical deployment scenarios for audio-based multiclass depression screening systems.

• On the interface between linguistics, computer science and psychiatry: analyzing textual key-factors affecting BERT-based classification of schizophrenia in social media texts (2026) · doi

  The generalizability of BERT-based schizophrenia classification to populations with varying literacy levels, cultural discourse norms, or non-English languages has not been tested. Cross-linguistic validation and evaluation across diverse demographic populations with different mental-health vocabulary conventions are needed to assess whether morphological and structural markers remain invariant across linguistic and cultural contexts.

• Latency-efficient edge intelligence in IoT networks using knowledge distillation (2026) · doi

  GPU utilization analysis (Figure 8) compares four methods but does not investigate how FEKD performs on edge devices without GPU acceleration or with heterogeneous compute resources (TPU, NPU, CPU-only inference) prevalent in constrained IoT edge nodes, leaving a gap in understanding applicability to truly resource-limited deployments.

• Lightweight and Explainable Neural Models for Multilingual Movie Script Certification (2026) · doi

  The lightweight neural models for multilingual movie script certification were developed and evaluated on a specific dataset, but their generalization to movie certification systems in legal and media compliance domains beyond entertainment remains unexplored. The paper mentions applicability to 'other high-stakes multilingual domains such as legal and media compliance' but does not provide empirical validation or adaptation strategies for these distinct regulatory contexts.

• Scour depth prediction using machine learning and explainable AI: assessment of bridge vulnerability (2026) · doi

  ANN-GA achieved the highest NNSE value (0.9468) among neural network models, demonstrating effectiveness of genetic algorithm-based optimization; however, the paper does not compare computational training time, convergence speed, or resource requirements between ANN-GA, ANN-PSO, and ensemble methods, which is critical for deploying real-time scour prediction systems.

• An Ontology Driven Machine Learning Framework for Early Prediction in Children with Cerebral Palsy (2026) · doi

  Continuous monitoring and ontology evolution procedures are mentioned as necessary for model longevity in the GMFCS classification framework, but specific protocols for detecting when ontological concepts require updating based on emerging clinical evidence or distributional drift in pediatric cerebral palsy populations are not defined.

• A general framework for Gaussian Splatting-based human-centric volumetric videos (2026) · doi

  Relighting and material editing techniques for 3D Gaussian Splatting remain underdeveloped, preventing realistic compositing of reconstructed human-centric subjects into large-scale or virtual environments, and blocking downstream applications requiring photorealistic material and lighting manipulation in volumetric videos.

• Adaptive distribution network reconfiguration with renewable energy and EV integration using reverse-multiverse learning archimedes algorithm (2026) · doi

  Comparative analysis shows RMLAA computation time improvements (5.93-6.96 seconds for IEEE test systems) but lacks evaluation on real-time reconfiguration requirements, communication delays in remote terminal units, and the minimum reconfiguration interval needed between successive switch operations in practical distribution systems with dynamic EV loads.

• Artificial intelligence and sustainable human resource management: Technologies, applications, and future directions (2026) · doi

  Federated Learning demonstrates superior privacy and algorithmic fairness scores (97 and 92 respectively) but exhibits a significant technology-readiness gap with adoption rates of only 42 and cost efficiency at 55. Research is needed to identify and test specific implementation strategies that can bridge this adoption-cost-efficiency gap for federated learning in sustainable HRM contexts.

• Uncertainty Assessment in Deep Learning-based Plant Trait Retrievals from Hyperspectral data (2026) · doi

  The spectral saturation problem affecting high LAI value predictions represents a fundamental data-intrinsic limitation that cannot be overcome by distance-based uncertainty estimation alone. More sophisticated sensing strategies and multi-modal approaches beyond purely optical methods are needed to address this physics-based canopy reflectance constraint in hyperspectral plant trait retrieval systems.

• Generalized Metric Subregularity with Applications to High-Order Regularized Newton Methods (2026) · doi

  The paper proves that under generalized metric subregularity (equation 74), the algorithm achieves convergence rate defined by τ(t), but does not provide concrete characterizations of the admissible function ψ for standard nonconvex problem classes (e.g., sums-of-squares polynomials, neural network training objectives) where metric subregularity can be verified.

• Smart Prediction of Weather-Induced Flight Delays Applying Deep Learning (2026) · doi

  User session authentication does not persist credentials to the SQLite database, reducing multi-session continuity in the Flask-based flight delay prediction system. Database schema and authentication middleware must be redesigned to support stateful user sessions and historical prediction tracking.

• International Journal of Intelligent Data and Machine Learning (2026) · doi

  The paper applies probability adjustment techniques to two automotive use cases (SUV purchase and brake service response), but does not explore whether the adjustment formula remains valid across different industry verticals or product categories with fundamentally different response rate distributions and imbalance ratios beyond the 10% and 40% examples provided.

• Deep Learning Based Fish Species and Freshness Detection Using Convolutional Neural Networks (2026) · doi

  No evaluation of the Tamil text-to-speech module's accuracy, intelligibility, or performance in high-noise fish market environments is provided; the effectiveness of this accessibility feature for non-technical users requires field testing and user satisfaction metrics.

• Artificial Intelligence and Multi-Omics for Anticancer Drug Development and Repurposing (2026) · doi

  Generative models for anticancer drug discovery have not been thoroughly integrated into standard laboratory workflows for practical application. The specific computational architectures (GANs, generative models) required to augment sparse multi-omics datasets and improve drug repositioning predictions need empirical validation in clinical oncology settings.

• Unified URL and QR Based Phishing Detection Framework (2026) · doi

  The web-based implementation uses Flask/Streamlit, but the paper does not address how the framework integrates with existing security infrastructure (email gateways, browser extensions, corporate proxies) or handles browser-based QR scanning with varying camera quality and lighting conditions.

• Microscopic Image-Based TB Detection Using an Enhanced Bi-LSTM Model Optimized by Firefly Algorithm (2026) · doi

  The paper claims suitability for low-resource clinical environments but provides no evaluation of inference time, computational memory requirements, or model size for deployment on resource-constrained devices (e.g., field microscopy workstations with limited GPU/CPU capacity).