International Journal of Research in Computing

Learning-Rate Scheduling for Neural Network Training: A Scoping Review of ReduceLROnPlateau, Cosine Annealing with Warm Restarts, Cyclical Learning Rates, and Linear Warmup with Linear Decay

SPGE Illukkumbura — 2026-07-16

This scoping review presents a structured synthesis of learning-rate scheduling for neural network training. The learning rate is a consequential hyperparameter in gradient based neural network training because it mediates the trade-off between early exploration and late-stage convergence precision. The review focuses on four widely used learning-rate scheduling paradigms: (1) ReduceLROnPlateau (RLROP), a feedback-driven reactive scheduler; (2) Stochastic Gradient Descent with Warm Restarts (SGDR), a periodic cosine-shaped schedule; (3) Cyclical Learning Rates (CLR), an oscillatory triangular or exponential schedule; and (4) Linear Warmup with Linear Decay (LW+LD), a two-phase schedule widely used with transformer models. We use a common notation as a pedagogical organizing device, distinguish formal results from empirical tendencies and heuristic analogies, and summarize how each method relates to classical step-size conditions and adaptive-optimizer stability. The quantitative tables are explicitly treated as heterogeneous literature summaries rather than controlled head-to-head benchmarks; they are retained only to document source-specific evidence and practical signals. The article includes a scoping-search log, a quality-assessment rubric, caveats on benchmark comparability, hyperparameter-sensitivity diagnostics, failure-mode checks, and an AI tool usage and human contribution statement. The resulting scheduler-selection framework is intended as practical guidance, not as a statistically ranked performance claim.

ShieldLink: Retry‑Aware Authenticated Encryption for Secure and Reliable Chiplet Interconnects

Michel Nguyen — 2026-07-16

Chiplet-based systems-in-package increasingly rely on high-speed die-to-die links such as UCIe and CXL, where link- layer retry mechanisms and authenticated encryption are of- ten implemented as separate reliability and security func- tions. This separation can create a time-of-check/time-of- use risk when acknowledgments advance before authentication completes. This study introduces ShieldLink, a retry-aware authenticated-encryption protocol that integrates link-layer delivery, buffer retirement, and cryptographic verification to enable secure and reliable chiplet interconnects. Shield- Link formalizes a deliverability invariant requiring CRC, se- quence, and AEAD verification before receiver advancement. It defines per-frame authentication (Mode A) and epoch authentication (Mode B), evaluates them with a discrete- event Gilbert–Elliott burst-error model, and includes bounded safety exploration plus RTL control-plane resource sizing. Mode A removes the validity-before-verification race while improving goodput by approximately 2.4–2.5 percentage points over the secure naive baseline at representative burst prob- abilities. Mode B improves wire efficiency, reaching about 0.926 at M = 32, but its epoch-retransmission cost causes a crossover around πB ≈ 0.04 under the default β = 0.2 stress regime. The results show that authenticated delivery must be treated as part of retry semantics rather than an afterthought. ShieldLink provides an auditable design invariant and practi- cal guidance on mode selection for future secure chiplet inter- connect adapters, with relevance to SDG 9: Industry, Innovation and Infrastructure.

DOI: doi.org/10.64701/ijrc/345/9121

Differentiated Thyroid Cancer Recurrence Classification Using Machine Learning Models and Bayesian Neural Networks with Varying Priors: A SHAP-Based Interpretation of the Best Performing Model

Nadeesha Shyami Kumari Herath Mudiyanselage — 2026-07-16

Differentiated thyroid cancer (DTC) recurrence is a major public health concern, requiring classification and predictive models that are not only accurate but also interpretable and uncertainty aware. This study introduces a comprehensive framework for DTC recurrence classification using a dataset containing 383 patients and 16 clinical and pathological variables. Initially, 11 machine learning (ML) models were employed using the complete dataset, where the Support Vector Machines (SVM) model achieved the highest accuracy of 0.9481. To reduce complexity and redundancy, feature selection was carried out using the Boruta algorithm, and the same ML models were applied to the reduced dataset, where it was observed that the Logistic Regression (LR) model obtained the maximum accuracy of 0.9611. However, these ML models often lack uncertainty quantification, which is critical in clinical decision making. Therefore, to address this limitation, the Bayesian Neural Networks (BNN) with six varying prior distributions, including Normal (0,1), Normal (0,10), Laplace (0,1), Cauchy (0,1), Cauchy (0,2.5), and Horseshoe (1), were implemented on both the complete and reduced datasets. The BNN model with Normal (0,10) prior distribution exhibited maximum accuracies of 0.9740 and 0.9870 before and after feature selection, respectively. As the BNN model with N(0,10) prior distribution employed after feature selection outperformed all the other models, it was chosen as the best performing model for DTC recurrence classification. This model was further analysed using epistemic and aleatoric uncertainty, reflecting the model’s confidence in its prediction. In addition, to enhance this model’s interpretability, SHapley Additive exPlanations (SHAP) values were calculated, providing valuable insights into the contribution of key variables to the model’s output.

Prompt Optimization Strategies for Large Language Models: Insights from Random Prompting with Gemini API

OAM Jayawardana — 2026-07-16

The increasing use of large language models (LLMs) has raised concerns about efficiency regarding token usage, as excessive token consumption burdens computational resources. This study addresses the need for systematic techniques to optimize prompts, evaluating three specific strategies within the Gemini API framework aimed at reducing token usage. The research examined 480 prompts across 16 models in the Gemini family to quantify potential token savings. The three strategies assessed are structured concise formatting, which utilizes bullet lists, headings, and concise language to streamline content; verbose removal, which eliminates redundant and irrelevant information to enhance clarity and focus; and code formatting, which promotes consistent syntax while minimizing inline commentary. Token counts were tracked before and after optimization using the Gemini count Tokens API endpoint. Findings revealed an average token saving of 37.15% across most Gemini-family models, with Gemma 3 models achieving a higher average reduction of 43.55%. Among the strategies, verbose removal yielded the greatest efficiency at an average reduction of 51.98%, followed by structured concise formatting at 33.36% and code formatting at 29.71%. The maximum individual saving recorded was 126 tokens. Multimodal testing with a shared fixed prompt set demonstrated consistent tokenization behaviour within the Gemini ecosystem across Gemini 2.0, 2.5, 3 preview, latest Gemini, and Gemma 4 model groups. Gemma 3 models consistently demonstrated higher efficiency, indicating variations in tokenization behaviour influenced by model architecture. These results suggest that prompt optimization can significantly enhance token efficiency within the Gemini model family, achieving reductions of approximately 30% to over 50% depending on strategy and model type; cross-ecosystem generalizability remains to be established. This research offers valuable insights for practitioners aiming to reduce computational costs and improve inference efficiency in LLM-based systems.

Machine Learning-Based Student Academic Performance Prediction in a Nigerian University Context: A Comparative Evaluation of Classification Algorithms

FI Mamudu — 2026-07-16

Academic failure and attrition remain persistent problems in Nigerian higher education, where institutions typically respond only after a student has already failed an examination. Educational data mining has shown that machine learning can flag at risk students early, yet most published work relies on datasets and institutional practices that do not reflect the realities of resource constrained African universities, limiting practical applicability. This study develops and benchmarks a machine learning framework for predicting first-year undergraduate academic outcomes at a single private Nigerian university, comparing six classification algorithms, identifying the most informative predictors, and assessing computational feasibility and fairness for single institution deployment. Six classifiers, Random Forest, XGBoost, LightGBM, Support Vector Machine, Logistic Regression, and k-Nearest Neighbour, were trained on 1,200 student records covering demographics, study habits, socioeconomic indicators, and first-year CGPA. After preprocessing, SMOTE class balancing, and Information Gain feature selection, models were evaluated using accuracy, precision, recall, F1-score, ROC-AUC, and PRAUC, with SHAP values used to interpret feature contributions. Random Forest achieved the highest accuracy (91.3%) and AUCROC (0.947), closely followed by XGBoost (91.1%, 0.945) and LightGBM (90.8%, 0.940); all three ensemble methods substantially outperformed SVM, Logistic Regression, and k-NN. Attendance rate, hours of self-study per day, and internet access were the three most predictive features under both Information Gain and SHAP analysis. A preliminary fairness audit found modestly lower recall for low-income students (81.5%) than high-income students (87.0%). The framework runs on standard institutional hardware without GPU infrastructure, making early-warning deployment computationally feasible for resource-constrained universities. Machine learning can support academic advising as a decision-support tool, though institutional governance, ethical oversight, and further fairness auditing are needed before operational use (SDG 4: Quality Education).