National School of Leadership

The rapid increase in deepfake technology is considered a serious challenge to both digital trust and media genuineness. While deep learning models have achieved high accuracy in detecting manipulated media, most operate as "black boxes," providing predictions without justification. This lack of transparency limits their reliability in critical applications like digital forensics and journalism. This paper introduces an Explainable Deepfake Detection System designed to bridge this gap. Our system utilizes a pretrained XceptionNet model for the robust classification of images and videos. We distinguish our methodology by introducing a novel dual modal explanation pipeline that synergizes visual heatmaps (Grad-CAM) with localized textual justifications (LIME), offering superior interpretability compared to prior unimodal approaches. The system is evaluated on standard benchmarks, including FaceForensics++, Deepfake Detection Challenge (DFDC), and Celeb-DF, achieving an accuracy of 92.35%, precision of 91.27%, recall of 90.88%, and an F1-score of 91.07%. Furthermore, we address critical security vulnerabilities by evaluating how well the model handles unseen deepfake generation techniques, adversarial attacks, low quality compression, and facial occlusions, ensuring its reliability in real-world forensic scenarios.

Retrieval-Augmented Generation (RAG) improves the factual accuracy of large language models by combining document retrieval with text generation. In biomedical question answering, where correctness is critical, the effect of key hyperparameters has not been studied in a systematic way. This paper presents an evaluation of RAG on the COVID-QA dataset with a focus on three retrievers (dense, BM25, hybrid), two retrieval depths (top- <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{k}=1,3$</tex>), and optional reranking with a cross encoder. We use a single biomedical prompt and measure exact match (EM), F1 score, semantic similarity, groundedness, and latency. We also report a composite score that balances lexical accuracy, semantic similarity, and efficiency. Our results on a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 0 0}$</tex>-question subset show that reranking improves grounding at the cost of extra latency, and that increasing top-k improves recall but gives smaller gains after a point. The study highlights that multiple metrics are needed to judge biomedical RAG systems reliably and that careful tuning of retrieval and reranking settings can yield practical improvements under compute constraints.

With the global population steadily increasing, the use of electronic devices is also rising, resulting in a significant accumulation of electronic waste (e-waste). This paper presents an intelligent robotic system designed to autonomously disassemble electronic devices and identify ewaste using a machine learning (ML) model trained entirely from scratch. Unlike previous disassembly robots that relied on static linebased or rule-based methods—limiting flexibility— our robot leverages a flexible ML-based model trained on diverse device types, allowing it to adapt to various forms of e-waste. Inspired by Apple’s “Daisy” robot, which disassembles iPhones and other Mac devices, our system enhances automation and sustainability in e-waste management. By deploying our robotic system, we aim to improve recycling efficiency, reduce hazardous impacts on human health, and minimize environmental damage.

As per the World Bank, health, education and employment are the pillars of economic development. In this study, parameters are chosen in a way that covers all the areas and spheres of economic growth. Several international economic policies and theories will be studied to understand whether those can be applicable in the present scenario. Special attention to be made on fiscal and monetary policy which of govt. has been taken as part of public policy. Research demonstrated the economy's extreme parameters in 3- dimensional space (nudge) and identified several public policies that can be implemented. But in reality, extreme conditions on all three dimensions generally do not sustain. Economics can be analysed in the following dimension: a. Macroeconomic factor: GDP deflator, CPI, WPI, FDI, continental labour transition, budget allocation v/s actual expenditure b. Micro-economic factor: supply and demand mechanism, fiscal policy, monetary policy Machine recommended public policy will be compared with man-recommended public policy based on parameters like accuracy, feasibility, effectiveness, efficiency, biasness, etc. This study will make a relative analysis of two mode of public policy recommendation.

More and more businesses are using computer vision systems, which has made it clear that they need strong defenses against threats from other companies. These attacks, which change raw data in small ways, can lead to big mistakes in classification, which makes AI-driven systems less reliable. There have been recent improvements in defense mechanisms for computer vision models against risks from other computers. This study paper looks at both old and new methods. The outcome documented in this study paper shows that different protection strategies work. Different levels of defense are provided by traditional techniques such as adversarial training, input preparation, and defensive distilling. Of these, adversarial training is the most effective against known threats. A lot of the time, these methods have trouble with new or complex hostile changes. More advanced methods, like verified stability and feature squeezing, work better, especially against more complicated threats. The most accurate defenses are hybrids that use antagonistic training along with feature manipulation or GAN-based cleaning. These methods offer a fuller defense. In the world of computer vision, the fight between attackers and defenders is still going on, and new attack methods are always coming out. New improvements in security systems have made computer vision models more durable, but there is still a lot of work to be done. In the future, hybrid and adaptable barriers could be very useful because they offer more flexible and strong security. To keep AI systems safe and reliable, these defenses must keep getting better. This is especially important for important uses like self-driving cars, healthcare, and spying.

The field experiment entitled “Effect of splitting of nitrogen and harvesting time on yield and economics of fodder maize (Zea mays L.)” was conducted during the kharif season of 2023-24 at Dr. Sharadchandra Pawar College of Agriculture, Baramati. The field research was laid out in factorial randomized block design (FRBD) with three replications. There were nine treatment combinations consisting of three nitrogen splitting treatments i.e., N1: 50 % RDN at (basal) + 50 % RDN at 30 DAS, N2: 75 % RDN at (basal) + 12.5 % RDN at 15 DAS and 12.5 RDN at 30 DAS, N3: 50 % RDN at (basal) + 25 % RDN at 15 DAS and 25 % RDN at 30 DAS and three harvesting times i.e., C1: Harvesting at flowering stage, C2: Harvesting at milk stage, C3: Harvesting at physiological maturity stage. Among different nitrogen splitting, applications of 50 % RDN at (basal) + 25 % RDN at 15 DAS and 25 % RDN at 30 DAS recorded significantly higher yield parameters and economics studies such as green fodder yield (581.96 q ha-1), dry matter content (21.61%) dry matter yield (126 q ha-1), gross monetary returns (₹ 192045 ha-1), net monetary returns (₹ 117740 ha-1) and B: C ratio (2.58) than other treatments. Whereas, harvesting time of fodder maize harvesting at physiological maturity obtained significantly higher yield parameters and economics studies viz., green fodder yield (580.58 q ha-1), dry matter content (21.79%), dry matter yield (126.95 q ha-1), gross monetary returns (₹ 191591 ha-1), net monetary returns (₹ 113787 ha-1) and B: C ratio (2.47) as compared to other remaining harvesting stages.

Pitch control of an aviation system presents a plethora of control complications, such as nonlinear effects, parameter uncertainties, disturbances, etc. In the face of parametric uncertainties and disturbances, as well as communication limits in an auditory medium, it is difficult to create efficient control algorithms that can accomplish precise path-following tasks and effective resilient control actions. This paper presents the Second Order Sliding Mode Control method for regulating the pitch angle in an aero-pendulum mechatronic system. The fundamental purpose of this study is to develop a reliable controller capable of minimising the system error between the regulated signal and the planned signal that corresponds to the pitch angle of the vehicle. By employing the proposed approach, both the prevalence of chattering and the control efficiency can be greatly improved. A comparison is given between the proposed model and the classical technique now in use, such as the classical Sliding Mode Controller.

Extreme weather events such as droughts are catastrophic and can have serious consequences for people and the environment. Drought may be managed if measures are taken in advance. The success of this endeavor depends on a number of factors, not the least of which is accurate descriptions and measurements of drought conditions. Reducing the negative consequences of droughts requires an early forecast of drought conditions. The primary objective of this research is, hence, to establish a process for the assessment and prediction of drought. The drought evaluation was carried out using the standards established by the SPI and the Indian Meteorological Department. Maps of drought severity were generated using severe drought data. Thirty years’ worth of SPI readings was analyzed. Fuzzy-based drought forecasting model parameters were determined during a 25-year period, and the model was validated throughout the remaining years. The findings of this study can be used by the community to help combat the drought. Before the drought worsens, the local government can implement lifesaving mitigating measures.

The quantum walk is a subset of classical random walks. Considerable research has lately been conducted on quantum walks in relation to quantum computing. The walk may be thought of as a type of quantum cellular automata. In a broader sense, there is a one-to-one link between them. Furthermore, we look at another type of quantum cellular automata that may be thought of as a quantum version of stochastic cellular automata. The quantum walk formalism is a widely used and quite practical paradigm for modelling quantum systems such as Dirac equation simulations, discrete dynamics in both the low and high energy regimes, and the development of a multitude of quantum algorithms. The walker in a QW moves in a quantum superposition of pathways, and the ensuing interference serves as the foundation for a broad range of quantum algorithms, including quantum search, graph isomorphism issues, and rating nodes in a network.