Scalability gaps in Computer Science

Representative open questions

Showing 30 of 167 — one per source paper, highest-quality first.

• Hybrid Deep Model for Pain Intensity Classification Using Fused ECG, EMG, and GSR Signals (2026) · doi

  The paper emphasizes that optimal signal sequence selection achieves performance improvements 'without increasing computational costs' but provides no computational complexity analysis, inference time comparisons, or memory footprint measurements for the BiLSTM-MHAT-CNN hybrid model versus the CNN baseline. Runtime and hardware requirements must be quantified to support claims of practical applicability in real-time automated pain recognition systems.

• 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 discrete adjoint framework from DAFoam differentiates through the fully converged steady RANS system; the computational cost, memory requirements, and scalability of this adjoint-based training approach for larger domains, unsteady flows, or three-dimensional configurations in practical CFD solvers have not been characterized.

• William & Mary: Comprehensive AI Governance & Activity Inventory (2026) · doi

  No publicly available HPC strategic plan or capacity roadmap addressing AI/ML workload growth was identified. Given the documented research portfolio including deep learning for nuclear physics, CNNs for coastal monitoring, LLM agent research, and diffusion models, the alignment between current compute capacity (600 TFLOPS combined, 12 GPUs across both campuses) and research demand requires explicit capacity planning documentation.

• Inferring High-Dimensional Dynamic Networks Changing with Multiple Covariates (2026) · doi

  The CVN methodology was applied only to the KiKme gene expression dataset with p=191 nodes and two external covariates (cancer group and radiation dose). The generalizability of the covariate-varying network inference approach to datasets with higher-dimensional gene expression spaces (p>1000) and more than two categorical or continuous covariates remains unexplored.

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

  Experimental validation was conducted at bench-scale geometry with a single fluid matrix and particle type; applicability of the ANN-based framework to field-scale directional drilling operations with varied fluid rheologies, particle size distributions, and higher solids concentrations requires investigation.

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

  The HDAL model was evaluated only on 13 operational transient conditions (fault classes) from two reactor domains with 31 image samples per class in Domain A. The generalization capability to nuclear power plants with significantly larger numbers of fault classes, different sensor configurations, or reactors from different manufacturer designs remains unexplored in adversarial transfer learning for fault diagnosis.

• A lazy and modular approach to int-blasting (2026) · doi

  The Eager algorithm implementation cannot handle 5,653 out of 25,223 SMT-LIB benchmarks due to stack-overflow errors from deeply nested bit-vector terms. The paper does not investigate alternative techniques or optimizations to handle deeply nested term structures, which represents a scalability limitation for the eager int-blasting approach.

• LLM-Powered Silent Bug Fuzzing in Deep Learning Libraries via Versatile and Controlled Bug Transfer (2026) · doi

  The window size ablation study for Bug Transfer-Guided Test Generation (Fig. 11a) identifies a plateau effect at size 10 where marginal bug discovery gains diminish beyond this threshold. However, the generalizability of this optimal window size across different cross-library transfer scenarios (PyTorch→TensorFlow, PyTorch→JAX) and varying API similarity distributions remains untested, limiting the scalability of batch testing strategy recommendations.

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

  The paper evaluates DEL models only up to 500 IoT devices; the scalability threshold and performance degradation patterns for large-scale deployments with thousands or tens of thousands of heterogeneous IoT devices connected to fog nodes remain unexplored. Specific investigation is needed on how latency, energy consumption, and detection accuracy scale beyond 500 devices.

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

  Lightweight attention-based architectures balancing predictive performance and computational efficiency have not been explored; future research should develop and evaluate streamlined attention variants (e.g., efficient Transformers, local attention) for practical deployment of audio-based multiclass depression screening.

• Artificial Intelligence (AI) Based Multi-Layered Approaches for Privacy Preservation in Federated Learning (2026) · doi

  While the G-BHO optimization algorithm achieved 35% parameter exchange reduction in the hybrid federated learning system, its applicability and performance optimization specifically for resource-constrained edge devices and IoT environments with federated learning deployments have not been investigated.

• Enhancing Breast Cancer Diagnosis through Machine Learning: A Robust Approach for Early Detection (2026) · doi

  While the web application was deployed using Streamlit and Flask frameworks, the scalability of the Random Forest breast cancer diagnosis system for real-time processing of large-scale datasets has not been evaluated. Cloud deployment optimization using Kubernetes for handling concurrent users and large mammographic image datasets in clinical environments is proposed but not yet implemented.

• Breast cancer recurrence risk prediction based on MIL (2026) · doi

  Computational efficiency is claimed for the ConvNeXt-MIL-XGBoost pipeline during inference, but the paper does not provide quantitative metrics such as inference time per WSI, memory footprint, or scalability benchmarks on datasets with varying WSI resolutions and patch counts for clinical deployment in breast cancer screening.

• A Federated Learning Framework with Metaheuristic Optimization for Heart Disease Prediction (2026) · doi

  The framework was tested on data from only 3 hospitals (Table 8); evaluation across larger federated networks with 10+ geographically distributed healthcare institutions and varying degrees of statistical heterogeneity (non-IID data distributions) is needed to validate robustness claims.

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

  The lightweight model design (measured in MB file size via ONNX format) is intended for deployment efficiency, but latency benchmarks, memory consumption during inference on edge devices, and scalability testing on full-length scripts versus script excerpts are not documented. Comparative analysis with larger transformer models (BERT) on computational constraints is absent.

• A Motion-Based Compression and Tracking System for Video Camera Trap-Based Insect Behaviour Studies (2026) · doi

  The Raspberry Pi 5 (8 GB) encountered out-of-memory errors when processing large datasets using the stand-alone H.265 codec, but the scalability limits of the proposed integrated motion-based compression approach have not been systematically characterized across different edge device memory configurations. The maximum dataset size, video resolution, and frame rate combinations that can be processed on various resource-constrained platforms (Raspberry Pi 4/5, Jetson Nano) without memory failures should be experimentally determined.

• Hepatitis C Diagnosis using Supervised Machine Learning Algorithms and Ensemble Learning Techniques (2026) · doi

  CatBoost exhibits the longest training time (17.4435 seconds) and largest model size (1103.11 KB) among all classifiers evaluated for hepatitis C diagnosis, yet achieves competitive AUC scores. The computational trade-offs between CatBoost's training overhead and its diagnostic accuracy benefits need explicit analysis to guide deployment decisions in resource-constrained clinical settings.

• Understanding the Dynamics of Trust and Engagement in E-Commerce Recommender Systems: Trends and Influences (2026) · doi

  Quantum computing applications using Quantum Approximate Optimization Algorithm (QAOA) for product recommendation optimization during high-pressure scenarios like flash sales are proposed but completely untested. Empirical validation is needed to demonstrate whether quantum optimization can actually outperform classical algorithms for large-scale e-commerce recommendation problems and handle real-time constraints.

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

  The frame-by-frame independent Gaussian point cloud representation for dynamic volumetric videos ignores extensive geometric and appearance redundancy between frames, resulting in storage and transmission costs that grow linearly with sequence length, making application to long sequences or resource-constrained mobile/XR scenarios impractical without novel cross-frame correspondence mechanisms.

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

  The RMLAA algorithm was validated only on IEEE 33-bus and 69-bus test systems; validation on larger-scale distribution networks (>100 buses) with multiple microgrids and higher penetration levels of renewable energy sources and EV charging stations is needed to assess scalability and real-world applicability.

• Sensor Data Fusion in Healthcare Monitoring System with Appropriate Rule-based Model for Error Reduction (2026) · doi

  The system demonstrated scalability for Heart Rate, Blood Pressure, Temperature, and the fused system, but the paper does not evaluate actual performance degradation or resource constraints when scaling to 50+ simultaneous patients or integrating more than five sensor types into the fusion framework.

• Criminal Face Sketch Recognition and Construction (2026) · doi

  The paper mentions integration with 3D face mapping and real-time CCTV surveillance matching as future features, but provides no methodological details on how sketch-based recognition will scale to large-scale surveillance databases or how the matching algorithm will handle temporal variations in appearance (aging, facial hair, expressions) across extended surveillance periods.

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

  The system is deployed locally with SQLite and Flask but lacks cloud scalability testing. Deployment on AWS or Azure infrastructure with containerization (Docker/Kubernetes) should be evaluated to assess performance of the hybrid XGBoost-ANN model under high-throughput real-time flight delay prediction workloads.

• ENHANCING IOT SECURITY USING LIGHTWEIGHT BLOCKCHAIN FOR DATA INTEGRITY AND TRACEABILITY (2026) · doi

  The throughput performance of 50 transactions/second was demonstrated only in a single experimental configuration. Scalability limits of the proposed lightweight blockchain model across varying numbers of IoT devices, transaction volumes, and network sizes (e.g., 100 to 10,000+ devices) are not established.

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

  The paper does not specify the computational requirements, inference latency, or energy consumption of the MobileNet model when deployed on Raspberry Pi hardware; real-time performance metrics for on-device deep learning inference in embedded systems need quantification.

• Cleansera: A Context-Aware, Algorithm-Centric Data Cleaning System with RAG-Enhanced Intelligence (2026) · doi

  Cleansera's cleaning pipeline includes seven sequential stages (Schema Validation, Duplicate Detection, Missing Value Treatment, Format Standardization, Outlier Handling, Semantic Validation, Quality Checkpoint), but the paper provides no empirical performance benchmarks or scalability analysis on datasets varying in size (from thousands to millions of records) or complexity (number of columns, data types, missing value percentages).

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

  The system demonstrated 100+ hours continuous stability without crashes, but no evaluation of performance degradation under high-volume concurrent requests (e.g., scanning 1000+ URLs/QR codes simultaneously) or with emerging phishing techniques post-deployment is documented.

• Employee Performance Classification and Monitoring using Machine zearning Models (2026) · doi

  The system's stability was validated only with repeated testing sessions on an unspecified dataset size and employee population; scalability to corporate/educational deployments with hundreds of concurrent users and varying hardware configurations (different webcam quality, network bandwidth) remains untested.

• Real-Time Bank Transaction Fraud Detection Using Kafka and Machine Learning (2026) · doi

  The system's scalability using Kubernetes and Docker containerization for managing multiple Kafka brokers and database instances has been mentioned conceptually but lacks empirical validation; performance benchmarks under varying transaction volume loads and latency constraints are absent.

• A Comparative Analysis of Machine Learning and Deep Learning Approaches to Enhanced Fake News Detection (2026) · doi

  While the paper establishes performance benchmarks for machine learning and deep learning approaches to fake news detection, it does not address how these models should handle the scalability challenge of processing greater volumes of fake news in real-time deployment scenarios across social media platforms and online news sites.