Introduction.

Deploying Automatic Speech Recognition (ASR) for African languages is critically
hindered by the incompatibility of large-scale models with mobile hardware. Models like
NLLB exceed 1 GB in size and cannot load on typical devices with 2-4 GB RAM, preventing
real-time conversational ASR deployment in resource-constrained settings. This technological
barrier disproportionately affects African linguistic communities.
Methods.

We formulate ASR deployment as a constrained optimization problem: minimize
Word Error Rate (WER) subject to bounds on model size, latency, and computational complexity.
Our solution integrates three complementary techniques: (1) Knowledge Distillation,
transferring knowledge from large teacher networks to compact students; (2) Low-Rank Factorization
(W ≈ UVT ), reducing parameters from O(p) to O(rp); and (3) Post-training 8-
bit Quantization for 4× memory reduction. Theoretical analysis provides sample complexity
bounds of O(k log n) and quantifies output perturbation from compression.
Results.

On a 20M-parameter baseline (85 MB), our framework achieves 16× total compression
to approximately 5 MB. Across iSiZulu, Setswana, and Sesotho datasets, mainly South African langauges, we maintain competitive
WER with only 5-6% degradation while increasing inference throughput from 0.8× to 6× in real time
on commodity CPUs. This 7.5× latency improvement enables previously impossible mobile
deployment.
Conclusions.

This work demonstrates that algorithm design grounded in mathematical optimization
and hardware-aware compression enables scalable, practical ASR for underserved languages.
The framework provides a general blueprint for efficient edge AI deployment, directly
advancing speech technology accessibility and digital inclusion for African linguistic communities
globally.

LLM-CapGen is a lightweight framework for video caption generation using large language models. The framework focuses on effective multimodal fusion while maintaining low computational cost. It extracts spatio-temporal features from video frames to capture visual events and motion patterns. It also extracts audio features, including speech and environmental sounds. In addition, the framework constructs a semantic scene graph that represents objects, actions, and temporal relations. This graph helps the model understand what happens, when it happens, and how events are connected.

The framework fuses visual, audio, and semantic features into structured prompts and feeds them to a large language model. This process enables the model to generate context-aware, temporally consistent captions. A post-processing step further improves caption quality by reducing redundancy and ensuring grammatical correctness across video segments.

We evaluate LLM-CapGen on the EgoSchema benchmark for long-form egocentric video understanding. The framework achieves strong results with 34.2 BLEU-4, 25.8 METEOR, 49.7 ROUGE-L, and 102.3 CIDEr scores. It outperforms several strong baseline models, including VideoBERT, CLIPCap, BLIP-2, and GPT-4V in zero-shot settings. Ablation studies show that audio encoding, semantic graph reasoning, and temporal prompting each improve caption quality.

These results show that lightweight multimodal fusion with structured prompting can produce accurate and meaningful video captions. LLM-CapGen provides an efficient and adaptable solution for multimodal video understanding tasks.

Introduction: Fractional-order chaotic systems offer a more accurate representation of complex dynamical processes characterized by memory and hereditary properties. However, achieving stability in the presence of unstructured uncertainties and unknown nonlinearities remains a significant theoretical challenge in the field of non-integer calculus and control theory.

Methods: This paper proposes a robust adaptive neural network control strategy designed for a class of uncertain fractional-order chaotic systems. Utilizing the universal approximation capabilities of Radial Basis Function (RBF) neural networks, the proposed controller identifies and compensates for system uncertainties online through a deterministic adaptation law. Unlike traditional model-dependent approaches, this framework requires no a priori knowledge of system parameters. A rigorous stability analysis is conducted using the fractional-order Lyapunov direct method.

Results: The theoretical analysis proves that the closed-loop system is Mittag–Leffler stable and that the tracking errors converge to a compact neighborhood of the origin. Numerical simulations performed on benchmark chaotic attractors validate the effectiveness and robustness of the proposed neural-adaptive scheme, showing high precision in trajectory tracking even under significant external perturbations.

Conclusion: The proposed strategy provides a mathematically sound foundation for the control of complex fractional dynamics. By integrating RBF neural networks with fractional Lyapunov stability theory, this research offers a robust solution for synchronizing or controlling chaotic systems where mathematical models are partially or entirely unknown.

We introduce the idea and possibility of loan in a modified Dragulescu–Yakovenko model in econophysics [1]. This is the implementation of the discrete version of the modified Z-model introduced by Pomeau and Lopez-Ruiz in 2015 [2]. This new model is equipped with a parameter that provides an idea of the level of debt that the agents can reach when they trade in a free market. When this parameter is adequately modified, the total debt in the system increases and a financial bubble is created. Once a certain limit is surpassed, the bubble is oversized and the financial system becomes unstable. Simulations and calculations are displayed for this system and they are put in correlation with real data. It is shown that a critical value in the system is found when the bubble, money created in relation to real money, reaches the factor of 5.18, a value in good agreement with the real data of bubbles formed in Western countries during the 2008 financial crisis.

[1] Dragulescu A and Yakovenko VM. Statistical mechanics of money. The European Physical Journal B, 17:723-729, 2000.

[2] Pomeau Y and López-Ruiz R. Study of a model for the distribution of wealth. In Lopez-Ruiz, Fournier-Prunaret, Nishio and Gracio, editors, Nonlinear Maps and Their Applications, vol. 112, p. 1-12, Springer Int Publishing, 2015.

A posthoc model-agnostic universal explainer for all machine learning models dealing with images that works in almost real-time is a long-lasting research effort in the image processing and computer vision community. One of the most stable and accurate solution is the KernelSHAP model that fulfills all the requirements but it is computationally intensive and resource‑demanding in practice. The method is based on the computation of Shapley values, and it has emerged as a principled approach for feature attribution in machine learning models, notably instantiated in the popular SHAP (SHapley Additive exPlanations) framework. However, the exact computation of the Shapley values is exponentially expensive, which motivated during time the development of faster approximation methods.

We introduce HarmonicSHAP, a completely different approach that uses spectral Fourier representations of the model to enable computations of the near-real-time Shapley value. Our framework is based on the inversion of a Gramian matrix arising from a chosen feature function basis, which yields a closed-form solution for Shapley values. We compare HarmonicSHAP to the well-known KernelSHAP and its subsequent accelerations versions in terms of computational complexity, implementation strategy, and performance. Our analysis shows that HarmonicSHAP can drastically reduce computations, applying a one-time model decomposition cost, enabling near real-time explanations even for complex models, while fulfilling the goal of remaining model agnostic and gradient independent.

Modern science operates at a critical intersection of physics, mathematics, and computation, where many of the most fundamental open questions remain unresolved. Although landmark problem sets such as Hilbert’s problems, the Millennium Prize problems, and Smale’s list have shaped scientific progress, they are typically treated in isolation, limiting systematic and interdisciplinary analysis. This work aims to construct a unified analytical framework that organizes major unsolved problems across physics, applied mathematics, and computational theory into a coherent transdisciplinary structure.

The study conducts a structured comparative analysis of canonical open problems and foundational literature in three domains: (i) fundamental physics (including quantum gravity, dark matter, and nonlinear physical systems), (ii) mathematical theory (including complexity, topology, and nonlinear dynamics), and (iii) computation (including algorithmic complexity, simulation limits, and predictability). Using conceptual mapping and thematic clustering, the paper identifies shared structural motifs such as symmetry, emergence, nonlinearity, and computational intractability that recur across traditionally separated scientific frontiers.

The results show that many grand challenges can be classified according to a small set of underlying mathematical–computational patterns that govern both physical law and algorithmic limitation. In particular, the analysis demonstrates how computational constraints are not merely technical barriers but fundamental components shaping what can be known, simulated, and experimentally accessed. By proposing a unified classification and conceptual model of open scientific problems, this work provides a systematic lens for interpreting scientific unknowns and establishes a foundation for future interdisciplinary research strategies.

Recent studies in the Philippines on landslides have primarily focused on susceptibility mapping and the generation of hazard maps. However, research on landslide forecasting remains less explored. As artificial intelligence continues to progress, forecasting methods used have also become more advanced. This study focuses on rainfall, a primary triggering factor of landslides, examining its relationship with environmental variables such as slope, soil type, and soil moisture to predict potential landslide events. The researchers applied the random forest model to forecast landslides using a minimized yet significant set of predictors (rainfall, slope, soil moisture, slope). The data of these variables have been gathered from various sources, such as sites with real-time data and government agencies that provide public datasets. These were then added into the dataset inventory created by the researchers and then trained using the random forest model. One-way ANOVA was then conducted to assess differences in model performance under various combinations of input variables, followed by post hoc tests to determine the most effective predictive variables. The researchers found that combining rainfall and environmental variables as predictors yielded the highest accuracy at 90%, outperforming models that used only individual variable inputs. These findings demonstrate the effectiveness of the random forest model in forecasting landslides even with limited resources and data, highlighting its potential as a practical and adaptable tool for early warning systems in areas such as the mountainous regions of the Philippines.

The rapid depletion of fossil fuel resources and the increasing pressure to mitigate environmental impacts have accelerated the large-scale integration of renewable energy resources into modern power systems. In this context, this paper presents a comprehensive tri-objective optimization framework for the dynamic optimal power flow (OPF) of a short-term hydro-thermal scheduling problem incorporating wind energy, solar photovoltaic (PV) generation, and battery energy storage systems (BESSs). The proposed framework addresses the coordinated operation of four progressively complex system configurations: thermal-only, thermal–wind, thermal–wind–solar PV, and thermal–wind–BESSs. The formulated problem simultaneously minimizes total generation cost, atmospheric emissions, and voltage deviation under a wide range of nonlinear operational constraints, including valve-point loading effects, hydro reservoir dynamics, renewable generation uncertainty, transmission losses, and battery charging–discharging behavior.

To effectively solve this highly nonconvex, large-scale, and constrained optimization problem, a hybrid Adaptive Encoding Learning-based Artificial Bee Colony algorithm integrated with the Non-Dominated Sorting Genetic Algorithm II (AEL-ABC–NSGA-II) is developed. The adaptive encoding mechanism enhances population diversity and convergence speed, while NSGA-II ensures robust Pareto-based tri-objective optimization. The effectiveness of the proposed approach is validated on the IEEE 39-bus test system under 24-hour dynamic load conditions. Comprehensive performance evaluations, including convergence analysis, Pareto front assessment, and statistical validation using ANOVA and box-plot analysis, demonstrate that the proposed method achieves superior solution quality, robust stability, and significant reductions in cost and emissions while maintaining an improved voltage profile. The results confirm the proposed framework as a reliable and efficient tool for next-generation sustainable power system operation.

Reinforcement learning (RL) has emerged as a promising paradigm for autonomous control tasks, where agents must learn sequential decision-making in dynamic and uncertain environments. Among RL algorithms, Deep Q-Networks (DQN) and Advantage Actor-Critic (A2C) represent two widely used approaches—value-based and policy-based, respectively—each with distinct strengths and limitations. This study presents a comparative analysis of DQN and A2C applied to the challenging CarRacing-v3 environment, where agents must handle continuous dynamics such as steering, acceleration, and braking.

Both agents were trained using preprocessed input frames, which involved grayscale conversion, cropping, resizing, normalization, and frame stacking to capture temporal dependencies. DQN was implemented with experience replay, target networks, and Double DQN extensions, while A2C employed a shared convolutional encoder for actor and critic networks with entropy regularization to encourage exploration. Training progress was measured using average return, stability, and computational efficiency.

Results revealed that neither DQN nor A2C achieved consistently stable driving policies. DQN struggled with continuous control due to its discretized action formulation, resulting in fluctuating average returns and poor convergence (final average return ≈ –71.5). A2C, while better suited for continuous actions, also stagnated with limited learning progression (average return ≈ –72.4), suggesting inefficiencies in exploration and sensitivity to hyperparameters.

In conclusion, this study highlights the challenges of applying classical RL algorithms to high-dimensional autonomous driving tasks. The findings provide empirical insights into their trade-offs and point toward the need for hybrid or advanced methods—such as DDPG, TD3, or SAC—that combine stability, efficiency, and adaptability for real-world autonomous driving applications.

This thesis examines the interaction between students' awareness of cybersecurity and quantitative mathematics modeling applications in gamified teaching environments. The primary reason is to utilize statistical fundamentals, mathematical logic, and quantitative modeling to analyze the knowledge, perceptions, and how students are prepared to identify and manage cyber threats. It presents a mathematical view on how information and knowledge can be used to create a model of sense and ability to make students create those models and systems, and to incorporate probability analysis, matrices, and logical algorithms to assess the level of sensitivity about cybersecurity. All the studies conducted in the thesis demonstrate a view and the student concept to know and find practical information from peers, to show the mathematical models, and to offer solutions on how to teach computational thinking. Also, the study will show the way students think that this computational science applies to, to understand if this method should be applied earlier in their school time, or if it will be perfectly suited for them in the high school curriculum only. This view integrates the mathematical concept and artificial intelligence to create a gamified environment in which students can practice critical thinking and database analysis, to make education about cybersecurity more effective. The outcome provides a strong base for strategy development and combines strong analytical skills, mathematical logic, and awareness about cybersecurity.