arXiv subjects

Suman Debnath

Publications and source records attributed to Suman Debnath.

9 recordsLinked to original sources

arXiv2026

Machine Learning-Assisted Stability Boundary Determination of Multiport Autonomous Reconfigurable Solar Power Plants

The multiport autonomous reconfigurable solar power plant (MARS) is a promising solution to integrate renewable energy resources and energy storage systems into the ac power grid and HVdc links. In the MARS system, various input power sources are connected to the individual submodules (SMs) through dc-dc converters. However, the presence of external power sources can result in unbalanced capacitor voltages of SMs, thereby violating stability constraints under multiple/diverse operating conditions. This paper aims to address the research gap by accurately determining the stability boundary for the MARS system. A novel machine learning (ML)-assisted energy balancing control (EBC) criterion is proposed. In conjunction with a refined EBC, this approach ensures balanced capacitor voltages across various types of SMs, significantly enhancing the overall system efficiency. The proposed EBC criterion effectively controls EBC activation and deactivation, achieving remarkable accuracy. Both PSCAD/EMTDC simulations and control hardware-in-the-loop (cHIL) tests are conducted to validate the feasibility and efficiency of the proposed method. By combining the EBC and ML-assisted EBC criteria, efficient energy management becomes achievable for systems featuring multiple input power sources, such as MARS. This approach enables the system to fully exploit its potential across an expanded operational range while upholding high efficiency standards.

eess.SY↗

arXiv2026

Non-Bloch band theory of boundary-controlled magnon edge modes in an antiferromagnetic chain

We define a winding number within the Non-Bloch band theory framework that captures the emergence of magnon edge modes in a one-dimensional antiferromagnetic spin chain, even when the conventional Bloch winding number is trivial. Within linear spin-wave theory, magnon excitations are governed by a non-Hermitian dynamic matrix, despite the underlying Hamiltonian being Hermitian. The symmetry classification of this matrix yields a trivial bulk invariant, however, finite systems exhibit boundary-localized modes, signaling a breakdown of the conventional bulk-boundary correspondence. We further show that these edge modes can be controlled via boundary perturbations. By tuning the boundary potential, the modes can be driven into or out of the bulk spectrum. To resolve the bulk-boundary mismatch, we develop a non-Bloch framework based on a generalized Brillouin zone and a winding number that correctly predicts the presence of edge states. Our results establish boundary-controlled topological transitions that are experimentally accessible through local Zeeman fields or modified edge anisotropy in antiferromagnetic van der Waals nanostructures.

cond-mat.mes-hall↗

arXiv2025

DSBC : Data Science task Benchmarking with Context engineering

Recent advances in large language models (LLMs) have significantly impacted data science workflows, giving rise to specialized data science agents designed to automate analytical tasks. Despite rapid adoption, systematic benchmarks evaluating the efficacy and limitations of these agents remain scarce. In this paper, we introduce a comprehensive benchmark specifically crafted to reflect real-world user interactions with data science agents by observing usage of our commercial applications. We evaluate three LLMs: Claude-4.0-Sonnet, Gemini-2.5-Flash, and OpenAI-o4-Mini across three approaches: zero-shot with context engineering, multi-step with context engineering, and with SmolAgent. Our benchmark assesses performance across a diverse set of eight data science task categories, additionally exploring the sensitivity of models to common prompting issues, such as data leakage and slightly ambiguous instructions. We further investigate the influence of temperature parameters on overall and task-specific outcomes for each model and approach. Our findings reveal distinct performance disparities among the evaluated models and methodologies, highlighting critical factors that affect practical deployment. The benchmark dataset and evaluation framework introduced herein aim to provide a foundation for future research of more robust and effective data science agents.

cs.AI↗

arXiv2025

Uncovering Cultural Representation Disparities in Vision-Language Models

Vision-Language Models (VLMs) have demonstrated impressive capabilities across a range of tasks, yet concerns about their potential biases exist. This work investigates the extent to which prominent VLMs exhibit cultural biases by evaluating their performance on an image-based country identification task at a country level. Utilizing the geographically diverse Country211 dataset, we probe several large vision language models (VLMs) under various prompting strategies: open-ended questions, multiple-choice questions (MCQs) including challenging setups like multilingual and adversarial settings. Our analysis aims to uncover disparities in model accuracy across different countries and question formats, providing insights into how training data distribution and evaluation methodologies might influence cultural biases in VLMs. The findings highlight significant variations in performance, suggesting that while VLMs possess considerable visual understanding, they inherit biases from their pre-training data and scale that impact their ability to generalize uniformly across diverse global contexts.

cs.CV↗

arXiv2025

A New Hybrid Quantum-Classical Algorithm for Solving the Unit Commitment Problem

Solving problems related to planning and operations of large-scale power systems is challenging on classical computers due to their inherent nature as mixed-integer and nonlinear problems. Quantum computing provides new avenues to approach these problems. We develop a hybrid quantum-classical algorithm for the Unit Commitment (UC) problem in power systems which aims at minimizing the total cost while optimally allocating generating units to meet the hourly demand of the power loads. The hybrid algorithm combines a variational quantum algorithm (VQA) with a classical Bender's type heuristic. The resulting algorithm computes approximate solutions to UC in three stages: i) a collection of UC vectors capable meeting the power demand with lowest possible operating costs is generated based on VQA; ii) a classical sequential least squares programming (SLSQP) routine is leveraged to find the optimal power level corresponding to a predetermined number of candidate vectors; iii) in the last stage, the approximate solution of UC along with generating units power level combination is given. To demonstrate the effectiveness of the presented method, three different systems with 3 generating units, 10 generating units, and 26 generating units were tested for different time periods. In addition, convergence of the hybrid quantum-classical algorithm for select time periods is proven out on IonQ's Forte system.

quant-ph↗

arXiv2025

Robust and Fine-Grained Detection of AI Generated Texts

An ideal detection system for machine generated content is supposed to work well on any generator as many more advanced LLMs come into existence day by day. Existing systems often struggle with accurately identifying AI-generated content over shorter texts. Further, not all texts might be entirely authored by a human or LLM, hence we focused more over partial cases i.e human-LLM co-authored texts. Our paper introduces a set of models built for the task of token classification which are trained on an extensive collection of human-machine co-authored texts, which performed well over texts of unseen domains, unseen generators, texts by non-native speakers and those with adversarial inputs. We also introduce a new dataset of over 2.4M such texts mostly co-authored by several popular proprietary LLMs over 23 languages. We also present findings of our models' performance over each texts of each domain and generator. Additional findings include comparison of performance against each adversarial method, length of input texts and characteristics of generated texts compared to the original human authored texts.

cs.CL↗

arXiv2025

Improving Multilingual Capabilities with Cultural and Local Knowledge in Large Language Models While Enhancing Native Performance

Large Language Models (LLMs) have shown remarkable capabilities, but their development has primarily focused on English and other high-resource languages, leaving many languages underserved. We present our latest Hindi-English bi-lingual LLM \textbf{Mantra-14B} with ~3\% average improvement in benchmark scores over both languages, outperforming models twice its size. Using a curated dataset composed of English and Hindi instruction data of 485K samples, we instruction tuned models such as Qwen-2.5-14B-Instruct and Phi-4 to improve performance over both English and Hindi. Our experiments encompassing seven different LLMs of varying parameter sizes and over 140 training attempts with varying English-Hindi training data ratios demonstrated that it is possible to significantly improve multilingual performance without compromising native performance. Further, our approach avoids resource-intensive techniques like vocabulary expansion or architectural modifications, thus keeping the model size small. Our results indicate that modest fine-tuning with culturally and locally informed data can bridge performance gaps without incurring significant computational overhead. We release our training code, datasets, and models under mit and apache licenses to aid further research towards under-represented and low-resource languages.

cs.CL↗

arXiv2020

A Hybrid DC Fault Primary Protection Algorithm for Multi-Terminal HVdc Systems

Protection against dc faults is one of the main technical hurdles faced when operating converter-based HVdc systems. Protection becomes even more challenging for multi-terminal dc (MTdc) systems with more than two terminals/converter stations. In this paper, a hybrid primary fault detection algorithm for MTdc systems is proposed to detect a broad range of failures. Sensor measurements, i.e., line currents and dc reactor voltages measured at local terminals, are first processed by a top-level context clustering algorithm. For each cluster, the best fault detector is selected among a detector pool according to a rule resulting from a learning algorithm. The detector pool consists of several existing detection algorithms, each performing differently across fault scenarios. The proposed hybrid primary detection algorithm: i) detects all major fault types including pole-to-pole (P2P), pole-to-ground (P2G), and external dc fault; ii) provides a wide detection region covering faults with various fault locations and impedances; iii) is more robust to noisy sensor measurements compared to the existing methods. Performance and effectiveness of the proposed algorithm are evaluated and verified based on time-domain simulations in the PSCAD/EMTDC software environment. The results confirm satisfactory operation, accuracy, and detection speed of the proposed algorithm under various fault scenarios.

eess.SY↗

arXiv2020

Hybrid EMT-TS Simulation Strategies to Study High Bandwidth MMC-Based HVdc Systems

Modular multilevel converters (MMCs) are widely used in the design of modern high-voltage direct current (HVdc) transmission system. High-fidelity dynamic models of MMCs-based HVdc system require small simulation time step and can be accurately modeled in electro-magnetic transient (EMT) simulation programs. The EMT program exhibits slow simulation speed and limitation on the size of the model and brings certain challenges to test the high-fidelity HVdc model in system-level simulations. This paper presents the design and implementation of a hybrid simulation framework, which enables the co-simulation of the EMT model of Atlanta-Orlando HVdc line and the transient stability (TS) model of the entire Eastern Interconnection system. This paper also introduces the implementation of two high-fidelity HVdc line models simulated at different time steps and discusses a dedicated method for sizing the buffer areas on both sides of the HVdc line. The simulation results of the two HVdc models with different sizes of buffer areas are presented and compared.

eess.SY↗