Authors - Nway Nway Zaw Win, Aye Nyein Mon, Win Lelt Lelt Phyu Abstract - Generating natural language descriptions for visual content is a key task bridging Computer Vision and Natural Language Processing. Conventional CNN-based approaches often struggle to capture global contextual information, limiting semantic consistency. This paper presents a multimodal video captioning framework for Myanmar-script generation based on a Vision Transformer (ViT) encoder and a Gated Recurrent Unit (GRU) decoder. Global visual representations are derived from transformer-based self-attention, while a class-prefixing mechanism is introduced to improve semantic grounding in a low-resource language setting. Experimental results evaluated using BLEU, CHRF, and TER metrics demonstrate that the proposed ViT–GRU model outperforms CNN–RNN baselines. PCA and t-SNE visualizations further confirm the effectiveness of transformer-based visual representations.
Authors - Kaveti Nani Kartik, Tanuja Pattanshetti Abstract - The proliferation of Large Language Models (LLMs) has raised concerns about embedded social biases and violations of fairness. Previous work has explored bias detection in word embeddings, fairnessaware algorithmic interventions, and system-level auditing frameworks. However, these approaches are still scattered across datasets, evaluation strategies and implementation pipelines. In this paper, we present a comprehensive literature survey to summarize the previous work on bias detection and fairness auditing, and categorize the contributions based on multiple phases of the research. Moreover, coverage and consistency limitations on popular benchmark datasets are analyzed. To address these problems, we present a unified dataset integration pipeline and a modular bias auditing framework. Identified critical research gaps include lack of intersectional bias modeling, lack of standardized metrics, and limited scalability in real-time auditing systems.
Authors - Mohit Apte Abstract - We develop systematic pairs trading strategies exploiting price adjustment lags between commodity-exporting currencies and their underlying commodities using CME futures. Two signal generation methods are compared: a rolling Z-score with Optuna-optimized hysteresis, and walk-forward Ridge regression on fourteen engineered features. Backtests on nine currency-commodity pairs over ten years of hourly data (2016–2026) show the ungated fundamental signal achieves Sharpe 0.56 under realistic costs. Adding rolling cointegration gating improves Sharpe to 0.64 while halving maximum drawdown from 23% to 12%. The ML signal reaches Sharpe 0.92, with strongest results on INR-Gold, AUDCopper, and CAD-Copper pairs. PCA-denoised Equal Risk Contribution sizing pushes ML Sharpe above 1.0 at the cost of higher drawdowns. Results confirm a tradable but risk-sensitive commodity-currency relationship at intraday frequencies.
Authors - Sayra Islam Saki, Qaium Hossain, Nadia Jahan, Abir Sen Gupta, Md. Tafshir Jaman Takib, Rajia Sultana, S.M. Sayem Abstract - This study examines how AI Systems, Big Data Analytics, Internet of Things (IoT) and E-Procurement enhances Sustainable Supply Chain Performance (SCP), with a particular focus on Supply Chain Resilience (SCR) as mediator. Primary data were obtained from 307 respondents of manufacturing industries through structured questionnaire. Partial Least Squares Structural Equation Modelling (PLS-SEM) approach was utilized for data analysis. The findings indicate that AI Systems, Big Data Analytics, Internet of Things and Supply Chain Resilience positively influence Sustainable Supply Chain Performance. On the contrary, E-Procurement doesn’t portray any significant direct effect. In terms of indirect pathways, SCR has positive mediating relationships between AI Systems and SSCP, as well as between IoT and SSCP. The mediation effect of SCR in the links between Big Data Analytics and E-Procurement with SCP is however not significant. These results provide subtle guidance to the practitioners in the industrial contexts, highlighting the need to prioritize those technologies that will promote resilience, specifically to AI Systems and IoT and re-examine the strategic contribution of E-Procurement to Sustainable Supply Chain models.
Authors - Swe Swe Htun, Aye Nyein Mon Abstract - Text classification has become a crucial task in natural language processing, especially for low-resourced languages in which limited annotated data and linguistic resources remain main challenges. This work presents inductive graph-based approach, GraphSAGE (Graph Sample and Aggregate), for text classification applying different word embedding models for Myanmar News classification. Experiments are conducted on eight Myanmar News categories (Business, Crime, Culture&Tourism, Educa-tion&Technology, Entertainment, Health, Politics, and Sports). The experiments show the effectiveness of BiLSTM (Bidirectional Long Short-Term Memory) and GraphSAGE architectures integrated with traditional and contextual embedding meth-ods, including TF-IDF (Term Frequency–Inverse Document Frequency), MyanBERTa, and mBERT (Multilingual Bidirectional Encoder Representations from Transformers). In the proposed work, transformer-based embeddings from pre-trained language models are extracted and combined with graph neural networks to capture both semantic and structural relationships among documents. A similarity graph is built by utilizing cosine similarity and k-nearest neighbor methods, and GraphSAGE is used to aggregate neighborhood information for inductive learning. The performance of graph-based models is compared to sequential deep learning technique based on BiLSTM. Experi-mental results reveal that graph-based approaches achieve better performance than BiLSTM-based models in all embedding settings. Among the evaluated models, mBERT with GraphSAGE gets the highest classification accuracy of 63%, followed by MyanBERTa with GraphSAGE with 60%. In contrast, MyanBERTa with BiLSTM and mBERT with BiLSTM yield 47% and 52% accuracy, respectively, whereas TF-IDF with GraphSAGE obtains 57% accuracy. The findings show that combining the contex-tual transformer embeddings with graph neural networks substantially enhance text classification performance by efficiently modeling semantic and relational information.
Authors - Edgar G. Cue, Felix JR Q. Pocong, Darwin Catalan, Santa M. Faltado, Ethel Reyes-Chua, Randy Joy M. Ventayen Abstract - This qualitative study examined business leaders' use of artificial intelligence (AI) in the workplace and employees' adaptation to AI-related changes. To gain a more thorough understanding of how various leadership practices are used, the employee experience, and the organizational response to the integration of AI into their operations, the researcher employed a qualitative research design. The research included business leaders and employees from organizations that have previously implemented or are currently implementing AI technologies in their operations as the study's subject population. Participants were selected using purposive sampling based on their direct involvement or knowledge of AI integration efforts within their organizations. The researcher collected data through interviews and coded it using thematic analysis to identify recurring themes and patterns related to leadership strategy, employee adaptation, organizational challenges, and workplace transformational changes resulting from the integration of AI. Major findings of this study indicate that, in implementing AI, leaders pre-dominantly used phased, strategic methods while considering employee readiness, continuous training, ethical governance, and partnerships to successfully implement AI within their organizations. On the other hand, employees exhibited both optimism and anxiety about AI adoption, with particular concerns about job security, technological skills, and organizational support. The study also established that transformational leadership, participative decision-making, transparent communication, a supportive leadership culture, and continuous capacity development are the most effective practices for facilitating employee adaptation and successful AI integration. The study concludes that successful AI integration requires not only technology but also a high degree of human-centered leader-ship, ethical accountability, and an organizational commitment to continuous development and change management to facilitate sustainable transformational change in organizations.
Authors - Divy Awasthi, Rushil Jariwala, Pearl Patel, Dhiren Patel Abstract - Artificial intelligence is deployed at scale across high-stakes domains—healthcare, autonomous systems, finance, and critical infrastructure— yet the pace of capability development has outrun our ability to ensure these systems behave safely, transparently, and in accordance with human values. While individual aspects of AI safety have been studied in isolation, a unified treatment spanning technical vulnerabilities, ethical risks, security threats, and governance failures remains lacking. This paper addresses that gap with a structured survey of Safe AI organized around a four-layer taxonomy of challenges—data, model, system, and societal—and a corresponding set of mitigation strategies at each layer. We trace AI’s evolution across three generations of increasing capability and opacity, examine domain-specific safety risks in healthcare, autonomous vehicles, manufacturing, and large language models, analyze the alignment problem through robustness, interpretability, controllability, and ethical adherence, and consolidate ten cross-layer directives for safe deployment. We review the global regulatory landscape, including the EU AI Act, GDPR, and national AI safety initiatives across the US, UK, and India, and identify open challenges in scalable oversight, formal verification, and the governance of increasingly autonomous AI systems.
