Sri Ramakrishna College of Arts and Science

Artificial Intelligence is increasingly influencing the methods through which English is taught, practised, assessed, and learned. The emergence of generative AI, intelligent tutoring systems, automated writing evaluation, adaptive learning environments, conversational agents, speech-recognition applications, and AI-supported assessment has created opportunities to move beyond uniform teacher-centred instruction towards more personalized, interactive, feedback-rich, and learner-responsive approaches. This conceptual research paper examines how Artificial Intelligence is redefining teaching methods in English language learning and how such changes may influence student performance. The study adopts an integrative literature review and conceptual analysis of recent scholarship on AI-assisted language learning, generative AI, automated feedback, academic writing, learner autonomy, personalized instruction, and digital pedagogy. Particular attention is given to changes in teaching strategies, including differentiated instruction, AI-supported conversation, adaptive language practice, automated formative feedback, process-oriented writing instruction, and data-informed assessment. The paper distinguishes between performance enhancement, in which AI improves the immediate quality or speed of task completion, and learning improvement, in which learners demonstrate transferable language competence without technological dependence. Evidence from recent empirical research indicates that structured AI-supported instruction can improve academic writing performance and support English proficiency and self-regulation. However, excessive dependence may reduce cognitive engagement, originality, and independent problem-solving, and authentic human interaction. The paper proposes the PERFORM-AI Framework, which integrates personalization, engagement, responsive feedback, formative assessment, originality, reflective learning, monitored AI use, and independent transfer. It concludes that the most effective model for English language learning is not AI replacing teachers but an instructional partnership in which teachers redesign pedagogy, AI extends opportunities for individualized practice and feedback, and students remain active agents responsible for their own learning.

This paper will examine how the random Forest, which is a powerful machine learning technique, can be used to predict employee turnover probability. The retention of employees is a very critical issue to any organization because excessive turnover may result in high expenditure and a loss of productivity. Through random forest, this research would establish a correct predictive model based on different features on employees such as demographics, performance variables, job satisfaction variables and remuneration variables. The ensemble method of random forest which involves the combination of many decision trees guarantees the high accuracy of the method since it helps in reducing the overfitting and complex correlations of the data. This is because the model can deal with both categorical and continuous variables, and hence it is highly applicable in workforce analytics. Findings of the present research point to some of the most significant aspects of employee turnover and give organizational stakeholders meaningful information on what can be done proactively in retention strategies. The tool can be used to assist the HR departments in recognizing the at-risk employees and implementing specific interventions to increase workforce stability.

The growing complexity associated with advancements in VLSI systems, along with severe restrictions on power, performance and area, have brought into light the limitations associated with traditional design optimization using isolated design layers. This is primarily due to strong interdependencies associated with architectural, logic and physical design decisions, which are typically hard to explore using traditional heuristic and task specific machine learning mechanisms. This paper proposes a new paradigm based on using cross layer co optimization using foundation models for optimal VLSI design. This work advocates the use of heterogeneous encoders so that multiple level design variables are mapped into a single space and a transformer based foundation model can capture cross layer correlations between various levels of VLSI design components. The closed loop optimization mechanism uses feedback from various stages of synthesis, placement and routing based design stages, showing promising results on advanced node benchmarks, reflecting a 15.6% improvement in power, 18.3% improvement in performance and 12.1% improvement in area using a foundation model based holistic VLSI design mechanism.

Early and accurate detection of Acute Lymphoblastic Leukemia (ALL) from peripheral blood images can significantly enhance survival rates and clinical outcomes. Conventional image segmentation techniques and initial Convolutional Neural Network (CNN)-based models have produced encouraging outcomes; however, they encounter difficulties in managing stain variations, overlapping cells, and data imbalance. This study introduces an advanced deep learning framework that combines Hybrid CNN–Transformer (HCT-Net) and Attention-driven Segmentation Networks (ASNet) to enhance the identification of leukemia-affected areas. The suggested method uses transfer learning from pretrained biomedical foundations, a dual-branch encoder–decoder architecture, and spatial-channel attention to highlight features that are specific to leukemia. Using the ALL-IDB1 and ALL-IDB2 datasets for experimental testing shows that the Jaccard Index, Tanimoto coefficient, and segmentation error are all much better than they were with CNN, Fuzzy C-Means, and K-Means methods. Keywords— Leukemia Detection; Hybrid CNN-Transformer; Attention Mechanism; Deep Learning, Medical Image Segmentation.

The heterogeneity of $\mathbf{5 G} / \mathbf{6 G}$ wireless mesh networks, including UAVs, vehicles, and edge devices, leads to frequent topology changes and intermittent links, both challenging the conventional routing protocols. This work proposes the GROF, Graph-based Routing Optimization Framework, a time-varying graph approach, fusing lightweight graph neural networks (GNN) with reinforcement routing for latencyminimizing, high-reliability path prediction in mixed-mobility meshes. GROF builds dynamic graphs from local link sensing, computes node embeddings by localized message passing, and selects the next hop using a distributed RL policy. Simulation results with 50-200 nodes, static to mobile ratio of 70:30, running on realistic mobility and traffic mixes, show improvement of PDR by $\sim 5-12 \%$ and reduction of end-to-end delay by $20-40 \%$, as compared to AODV, OLSR, HWMP, and a DRL baseline, while reducing routing overhead and energy per delivered packet. GROF's distributed graph learning is designed for deployment at the edge and promises good scalability to denser topologies.