In our research, we propose HFMF, a comprehensive two-stage deepfake detection framework that leverages both hierarchical cross-modal feature fusion and multi-stream feature extraction to enhance detection performance against imagery produced by state-of-the-art generative AI models.

The work proposes a classification system based on the UNet architecture, which processes transformed spectrograms of the PCG signals. The augmented spectrograms have yielded the best results. Specifically, on the PhysioNet 2016 dataset, the proposed model has achieved an accuracy of 96.05%, specificity of 98.82%, and F1 score as 0.91.

We introduce AmCLR and xAmCLR objective functions tailored for bimodal vision-language models to further enhance the robustness of contrastive learning. AmCLR integrates diverse augmentations, including text paraphrasing and image transformations, to reinforce the alignment of contrastive representations, keeping batch size limited to a few hundred samples unlike CLIP which needs batch size of 32,768 to produce reasonable results. xAmCLR further extends this paradigm by incorporating intra-modal alignments between original and augmented modalities for richer feature learning.

This paper proposes a model HeartBeatNet (an attention UNet-based system) for heart sound classification that demonstrates comparatively better performance. The proposed system combines the strengths of attention mechanisms and the UNet architecture to effectively capture relevant features and to make accurate predictions.

The current research proposes a computationally less-expensive hybrid approach combining cluster analysis and deep neural learning with transfer learning to estimate the machine-level workload. The method implements clustering to identify the similarity patterns among the non-linear usage profiles of machines present in the input dataset.

This study utilized retrospective and prospective data from AIIMS-New Delhi, totaling 2190 and 92 samples respectively, with a 70% IS and 30% HS split. Stroke classification models were trained and then evaluated on these datasets using three calibration techniques: Platt Scaling, Histogram Binning, and Isotonic Regression, with performance measured by Expected Calibration Error (ECE).

Plant breeders and agricultural researchers can increase crop productivity by identifying desirable features, disease resistance, and nutritional content by analysing the Dry Bean dataset. This study analyses and compares different Support Vector Machine (SVM) classification algorithms, namely linear, polynomial, and radial basis function (RBF), along with other popular classification algorithms.

InDocQ is an advanced document question-answering system powered by LangChain and Large Language Models (LLMs) and hosted using StreamLit. This application enables users to upload PDF documents and engage in interactive Q&A sessions about the document's content, leveraging the power of semantic search and state-of-the-art language models.

This project provides an efficient solution to the problem of finding optimal generator matrices G to minimize the "m-height" of any analog code x. The "m-height" of a codeword c generated with Matrix G and Vector x measures the ratio of the largest and mth largest absolute elements of c. The m-height of the Generator Matrix G is the maximum m-height across all its codewords for all possible x's.to minimize the m-height of an analog code. The implementation combines genetic programming and stochastic optimization techniques to iteratively refine both G-matrices and X-vectors for improved performance.

This project involved designing and deploying an Extractive Search Engine tailored for the NCERT History textbook. The search engine aimed to provide high-school students with precise, contextually accurate answers to their queries by extracting relevant information from the textbook content. It supports over 1 million students, enhancing their learning experience through quick and accurate answers, reducing the need for manual textbook navigation.

GradFlow is a Python library that implements automatic differentiation from scratch. It provides the core building blocks for constructing and training neural networks.

(Approximate funding of $100,000 spread over a span of 3 years).

The project aimed to revolutionize the diagnostic process for Mycobacterium Tuberculosis (TB) by developing a non-invasive, rapid, and cost-effective diagnostic device. This device detects Volatile Organic Compounds (VOCs) in a patient's breath as biomarkers for TB, eliminating the need for invasive and time-consuming tests like sputum analysis or chest X-rays.