Techno International Newtown, formerly Known as Techno India College Of Technology(TICT), is an engineering college in West Bengal, India. It is the seventh and youngest college established by the Techno India Group. The college is located on 5 acres (20,000 m2) of land in the Megacity (New Town) area of Rajarhat, Kolkata. It is affiliated with Maulana Abul Kalam Azad University of Technology(formerly known as West Bengal University of Technology) and its courses are approved by All India Council of Technical Education.
Automatic generation of radiology reports seeks to reduce clinician workload while improving documentation consistency. Existing methods that adopt encoder-decoder or retrieval-augmented pipelines achieve progress in fluency but remain vulnerable to visual-linguistic biases, factual inconsistency, and lack of explicit multi-hop clinical reasoning. We present NeuroSymb-MRG, a unified framework that integrates NeuroSymbolic abductive reasoning with active uncertainty minimization to produce structured, clinically grounded reports. The system maps image features to probabilistic clinical concepts, composes differentiable logic-based reasoning chains, decodes those chains into templated clauses, and refines the textual output via retrieval and constrained language-model editing. An active sampling loop driven by rule-level uncertainty and diversity guides clinician-in-the-loop adjudication and promptbook refinement. Experiments on standard benchmarks demonstrate consistent improvements in factual consistency and standard language metrics compared to representative baselines.
In this paper, we extend the classical Hastings–Powell tri-trophic food chain model by incorporating two ecologically realistic mechanisms: predator-induced fear in prey and a weak Allee effect. A rigorous mathematical analysis is carried out, including positivity, boundedness, dissipativeness, and persistence of solutions, followed by a detailed investigation of all equilibrium points and their local and global stability properties. Through bifurcation analysis, we identify that govern system transitions. Sensitivity analysis using Partial Rank Correlation Coefficients (PRCC) highlights the most influential parameters for prey, predator, and top predator dynamics. Numerical simulations, conducted via MATLAB and MATCONT, further reveal complex dynamical regimes. In particular, variation of the prey-handling time parameter [Formula: see text] drives the system into chaos, as confirmed by a positive maximum Lyapunov exponent (MLE). Importantly, we demonstrate that chaos can be suppressed by strengthening the weak Allee effect ([Formula: see text]) and predator-induced fear ([Formula: see text]). Two-parameter bifurcation diagrams for [Formula: see text], [Formula: see text], and [Formula: see text] provide deeper insights into the combined ecological effects of these parameters. The results, supported with ecological interpretations, show how behavioral responses and Allee effects can regulate chaotic oscillations and promote system stability, offering useful theoretical insights for ecological management and conservation.
High-dimensional gene expression data pose substantial challenges for predictive modeling due to noise, redundancy, and the curse of dimensionality. This work introduces a hybrid machine learning framework that integrates a Variational Autoencoder (VAE) for nonlinear dimensionality reduction with an Explainable Boosting Machine (EBM) for transparent multiclass cancer classification. The VAE compresses over 20,000 transcriptomic features into a 64-dimensional latent space while preserving biologically meaningful structure, yielding improved separability across tumor types. These latent representations are then classified using an EBM, enabling inherently interpretable predictions supported by global feature importance and SHAPbased local explanations. Experiments on the TCGA Pan-Cancer dataset demonstrate the effectiveness of the proposed pipeline, achieving 0.98 accuracy, 0.99 macro-precision, 0.97 macro-recall, and 0.98 macro-F1, with per-class F1-scores ranging from 0.95 to 1.00. The results highlight that the VAE-EBM framework provides both high predictive performance and clinical interpretability, offering a promising direction for precision oncology.
Relying on a manual system can be time-consuming for a doctor when managing multiple patients simultaneously, making it challenging to deliver improved service in a shorter time frame. Moreover, the possibility of human error remains an undeniable factor in this situation. In this context, an automated doctor-patient interaction system can mitigate such issues effectively. To achieve this, an automated system, PulmoConnect that enables doctors to deliver enhanced services to patients in significantly less time has been developed. Additionally, patients can directly seek assistance from doctors through this platform. The goal of the application is to enhance communication and effectively manage patient records, enabling coordinators and doctors to deliver prompt support and care. This system accommodates the development of two distinct mobile applications. One application designed for coordinators and doctors, known as the Hospital Staff App, includes interfaces for both the registration desk and the doctors and coordinators. Another application is designed for patients, allowing them to access their information and available services. In designing the system's front end, some of the latest technologies, like React Native Framework and Expo Development Platform have been used. For the backend, we have utilized Express.js and Node.js, with MongoDB serving as our database. This reflects the current status of the development so far. The remaining development tasks involve enabling coordinators to update patient information during revisits, edit existing details, and generate APK files for the applications.
This study investigates sentiment analysis on code-mixed Bengali, English, and Hindi text, evaluating both transformer-based architectures and traditional machinelearning models. Experimental results show that multilingual transformers significantly outperform classical approaches in capturing complex multilingual and contextual patterns. The mBERT model achieved the highest weighted F1 score of 0.75, followed by XLM RoBERTa with 0.73, highlighting the effectiveness of fine-tuned multilingual Pre trained Language Models (PLMs) for code-mixed data. Among classical baselines, Support Vector Machine performed best with an F1 score of 0.677, demonstrating that TF IDF and n-gram features remain competitive for mixed-language tasks. These findings emphasize that multilingual pre training and cross lingual alignment are essential for robust sentiment analysis in code-mixed environments.
