Authors - Ioannis Patias, Koutaro Hachiya Abstract - The rapid growth of compute-intensive applications has intensified the need for selecting appropriate hardware accelerators. This paper presents a workload-to-architecture framework that explains when GPUs, FPGAs, or ASICs are most suitable, based on four determinants: parallelism granularity, memory behavior, dataflow regularity, and specialization depth. We organize representative workload classes—dense linear algebra, sparse/irregular algorithms, streaming signal processing, bit-level control workloads, and deep learning (training/inference)—and discuss how each class aligns with the execution and memory models of the three accelerator families. Our analysis highlights that GPUs excel in throughput-oriented, data-regular workloads; FPGAs provide deterministic latency via spatial pipelining and customized data paths; and ASICs achieve the best performance-per-watt for stable, high-volume tasks. The resulting framework provides practical guidance for accelerator selection and motivates heterogeneous system design.
Authors - Ronky Amber-Doh, Benjamin Ghansah, Winfred Larkotey, Stephen Opoku Oppong, Ezekiel Okoe, Olivia Osei-Tutu, Emmanuel Prah, Ephrem Kwaa-Aidoo Abstract - This paper introduces a novel framework, ExplainoGraph, that integrates square loss optimization with explainable artificial intelligence methods for knowledge graph–based recommender systems. Prior studies show that knowledge graph embeddings significantly enhance recommendation accuracy; however, they suffer from limited interpretability, thereby constraining user trust and system transparency. To address this gap, we introduce ExplainoGraph, which embeds explainability directly into the recommendation process through interpretable scoring functions and feature attribution procedures that provide meaningful insights into model decisions. Again, the framework incorporates ripple set propagation to effectively model user preferences, particularly in sparse data environments where traditional methods are suboptimal. Extensive experiments conducted on multiple benchmark datasets demonstrate that Explaino-Graph consistently outperforms the state-of-the-art baselines used across key evaluation metrics, including Precision@K, Recall@K, F1-score, and normalized discounted cumulative gain (NDCG)
Authors - Sayyora Qulmatova Abstract - This study uses machine learning models such as Multi-Linear Regression (MLR) and Holt-Winters Exponential Smoothing to model and forecast agricultural production indicators in Uzbekistan. The dataset consists of key agricultural indicators such as gross agricultural output, milk production, egg production, honey production, vegetables, fruits, and livestock products (in live weight). To improve model performance and ensure comparability, various data preprocessing methods such as StandardScaler, MinMaxScaler, RobustScaler, and Normalizer were used. The forecasting accuracy of each model was evaluated using standard error metrics such as mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), and mean absolute percent- age error (MAPE). Empirical results show that the MLR model provides stable and interpretable forecasts, especially when combined with appropriate scaling methods. The Holt-Winters model exhibits strong performance for time series with consistent trends, but shows limitations when applied to variable data. The MLP model effectively captures nonlinear relationships and complex time patterns, although its performance is sensitive to data preprocessing, with MinMaxScaler generally yielding superior results. Overall, the results show that no model is universally optimal; instead, the choice of forecasting technique should be based on the characteristics of the data. The proposed modeling framework contributes to increasing the accuracy and reliability of agricultural forecasts and can support evidence-based policy planning and decision-making in the agricultural sector of Uzbekistan.
Authors - Salma Chlaikhy, Adil Chakhtouna, Abdellah Adib Abstract - We propose a neural encoding framework that predicts continuous ECoG signals from self-supervised Data2Vec speech representations using multivariate Ridge regression. Evaluated on 18 auditory cortical electrodes from nine participants, the model achieves a mean Pearson correlation of r ≈ 0.39 under pooled cross-validation and r = 0.29±0.018 under leave-one-subject-out (LOSO) evaluation, reaching approximately 57% of the noise ceiling. Results confirm that self-supervised speech representations capture stimulus-driven cortical dynamics, highlighting their promise for neural signal modeling and brain–computer interface research.
Authors - Raita Rollande Abstract - This paper addresses the research question: How can an effectively organized and managed adult education reskilling and upskilling process help bridge the skills gap in today’s labor market? In an era of rapid technological development and changing business needs, the demand for continuous workforce reskilling and up-skilling has become increasingly critical. This study examines a private sector-driven adult education model that prioritizes speed, quality, and flexibility in response to industry demands. TestDevLab is a fast-growing company specializing in software quality assurance. Due to the lack of industry-specific specialists from traditional higher education institutions, TestDevLab has developed an in-house training solution where experienced engineers train new specialists while also upskilling existing employees. To address these challenges systematically, the company established TDL School. This dedicated training institution has designed the Model for Rapid and High-Quality Reskilling and Up-skilling tailored to mid-sized businesses. This article presents the TDL School model, exploring innovative training approaches, curriculum development, and company collaboration to highlight best practices for effective workforce development. The findings demonstrate that a well-structured, professionally organized learning process enhances employability, strengthens industry competitive-ness, and fosters lifelong learning. By sharing this model, the study aims to pro-vide a scalable framework that other companies of similar size can adopt to ad-dress their workforce challenges. This research contributes to the ongoing discussion on adaptive education models and their role in bridging the skills gap in today’s labor market.
Authors - Isaac Baffour Senkyire, Benjamin Ghansah, Emmanuel Freeman Abstract - With the rapid development of deep learning, CNN-based medical im-age segmentation algorithms have been successful. However, study on the pancreas in 3D CT and MRI images is limited due to the excess use of computer memory and the complexity of the pancreas. In this paper, we present a memory-efficient cascaded 3D network for pancreas segmentation in CT and MRI. We develop a novel Lightweight 3D Bond (L3D-Bond) Layer to reduce filter size, and maintain performance while lowering memory usage, and a novel Light-weight 3D Asymmetric Corollary Atrous Spatial Pyramid Pooling Module (L3D-aCASPP) that captures multi-scale 3D context with lower computational cost. Our experiments were done using the public NIH pancreas segmentation dataset, MRI pancreas segmentation dataset, and MSD spleen segmentation dataset achieving a competitive segmentation performance of 80.12 DSC on the NIH dataset with parameters less than 0.5 million.
