State-of-the-art (SOTA) models typically use large datasets for pre-training and then fine-tune on smaller datasets for better performance. However, the high computational cost can be a barrier for many researchers. There is a need to focus on data size-independent models suited for data-scarce scenarios, which is essential for tasks like fingerprint recognition and could make research more accessible and generalizable in resource-limited environments. With this aim, this paper presents a novel approach to the difficulties in contactless fingerprint recognition, particularly with scarce and poor-quality challenging dataset images due to contactless acquisition. Our proposed system uses a 'Scattering using a Shearlet Network (SSNet)' to extract fingerprint features and a score-level fusion scheme to improve authentication accuracy. In contrast to the computationally expensive and mathematically less transparent dense deep learning networks such as vision transformers, attention networks, deep learning-based hybrid approaches, etc., SSNet is an economical framework with fixed filters. The SSNet is a replacement to the Scattering Wavelet Network (SWN) that utilizes a Complex Morlet Wavelet (CMW). Our model significantly improves verification and identification accuracy over SOTA approaches, particularly with scarce and poor-quality challenging datasets.

Grammatical error correction (GEC) tools, powered by advanced generative artificial intelligence (AI), competently correct linguistic inaccuracies in user input. However, they often fall short in providing essential natural language explanations, which are crucial for learning languages and gaining a deeper understanding of the grammatical rules. There is limited exploration of these tools in low-resource languages such as Bengali. In such languages, grammatical error explanation (GEE) systems should not only correct sentences but also provide explanations for errors. This comprehensive approach can help language learners in their quest for proficiency. Our work introduces a real-world, multi-domain dataset sourced from Bengali speakers of varying proficiency levels and linguistic complexities. This dataset serves as an evaluation benchmark for GEE systems, allowing them to use context information to generate meaningful explanations and high-quality corrections. Various generative pre-trained large language models (LLMs), including GPT-4 Turbo, GPT-3.5 Turbo, Text-davinci-003, Text-babbage-001, Text-curie-001, Text-ada-001, Llama-2-7b, Llama-2-13b, and Llama-2-70b, are assessed against human experts for performance comparison. Our research underscores the limitations in the automatic deployment of current state-of-the-art generative pre-trained LLMs for Bengali GEE. Advocating for human intervention, our findings propose incorporating manual checks to address grammatical errors and improve feedback quality. This approach presents a more suitable strategy to refine the GEC tools in Bengali, emphasizing the educational aspect of language learning.

3D human pose estimation (3DHPE) in images aims at estimating 3D joint positions from images. The state-of-theart for 3DHPE is dominated by deep learning model whose accuracy is obviously affected by loss functions. The existing 3DHPE methods usually define the loss function as the error measured by Euclidean distance between the locations of the predicted joints and the ground truth of joints, which confuses two different kinds of errors: the error caused by different pose structures and the others. But in fact, the characteristics of these two kinds of errors are obviously different and should not be processed equally, and consequently decoupling these two kinds of errors and optimizing them separately is one of the ways to improve the 3DHPE accuracy. However, The existing human pose representations are not suitable to distinguish these two kinds of errors. In order to tackle this problem, we propose a novel Multi-Anchor Offset human Representation (MAOR) for human pose, which locates the position of each joint using its offsets from a group of selected high-precision joints named as Multi-Anchors. Making use of MAOR, the pose error related to the distortion of spatial structure can be measured independently from other errors, which is helpful to promote the accuracy of pose estimation. We then propose a novel MAOR based coarseto-fine diffusion model (MAOR-DiffPose) for pose estimation, which optimizes different types of errors of poses step by step. Firstly, a MAOR-based Denoising Process (MDP) is devised to explicitly optimize spatial structures of 3D poses by using MAOR to describe poses and improves the inductive learning ability of MAOR-DiffPose by extracting view-independent features. Secondly, a Joint Coordinate based Denoising Process assisted by MAOR (JCDPaM) is devised to expand the input features meaningfully by combining MAOR with the pose representation based on joint coordinate and optimize the joint coordinates of 3D poses with the assistance of MAOR. MAOR-DiffPose realizes accurate 3DHPE by iterating MDP and JCDPaM modules. Comprehensive experimental results on widely used 3DHPE benchmarks Human3.6M and MPI-INF-3DHP show that the proposed method achieves the best performance compared with the state-of-the-art methods.

This research paper presents the design and implementation of an AI-Based Smart Home Security System Using Drone Surveillance. Leveraging advances in artificial intelligence and drone technology, the system aims to enhance home security through proactive threat detection and surveillance.

Phylogenetic networks are a useful model that can represent reticulate evolution and complex biological data. In recent years, mathematical and computational aspects of tree-based networks have been well studied. However, not all phylogenetic networks are tree-based, so it is meaningful to consider how close a given network is to being tree-based; Francis–Steel–Semple (2018) proposed several different indices to measure the degree of deviation of a phylogenetic network from being tree-based. One is the minimum number of leaves that need to be added to convert a given network to tree-based, and another is the number of vertices that are not included in the largest subtree covering its leaf-set. Both values are zero if and only if the network is tree-based. Both deviation indices can be computed efficiently, but the relationship between the above two is unknown, as each has been studied using different approaches. In this study, we derive a tight inequality for the values of the two measures and also give a characterisation of phylogenetic networks such that they coincide.

The application of knowledge distillation to reduce hallucination in large language models represents a novel and significant advancement in enhancing the reliability and accuracy of AI-generated content. The research presented demonstrates the efficacy of transferring knowledge from a high-capacity teacher model to a more compact student model, leading to substantial improvements in exact match accuracy and notable reductions in hallucination rates. The methodology involved the use of temperature scaling, intermediate layer matching, and a comprehensive evaluation using the MMLU benchmark, which assessed the model's performance across a diverse set of tasks. Experimental results indicated that the distilled model outperformed the baseline in generating accurate and contextually appropriate responses while maintaining computational efficiency. The findings underscore the potential of knowledge distillation as a scalable solution for improving the robustness of large language models, making them more applicable to real-world scenarios that demand high factual accuracy. Future research directions include exploring multilingual and multi-modal distillation, integrating reinforcement learning, and developing more refined evaluation metrics to further enhance model performance.

This study investigates the impact of ChatGPT as a collaborative tool in the ideation phase of human-centered design processes, involving 208 students across 37 groups. Through analysing student reports and reflections, both with and without ChatGPT usage, the study explores its impact on creative processes and its educational value. Our findings indicate that ChatGPT contributes significantly to enhancing the overall quality of creative projects and aids in problem identification, offering students broader perspectives. However, its contribution to enhancing creativity, feasibility and effectiveness of solutions was found to be marginal. We highlight the challenges of integrating ChatGPT in group work, particularly regarding its influence on group dynamics and creativity. This study highlights the need for more structured guidance on effectively using ChatGPT, balanced with encouraging student-led experimentation with the tool. We contribute insights into the nuanced application of AI tools like ChatGPT in educational contexts, especially in facilitating collaborative design tasks.

A document by Sujanavan Tiruvayipati. Click on the document to view its contents.

Breast cancer stands as a pressing health concern for women across the globe with millions of new cases worldwide each year. The diagnostic procedures often rely on expert analysis and interpretation, which can be subject to errors and limitations. This review offers an in-depth examination of the potential role of artificial intelligence as a transformative tool in the diagnosis of breast cancer. The focus is on the recent advancements within machine learning and deep learning, subsets of artificial intelligence that have shown promise in enhancing the accuracy and efficiency of breast cancer detection and classification. The paper discusses studies published in 2023, which have utilized artificial intelligence models with diagnostic medical datasets to identify, classify and predict the presence of breast cancer with increased precision. By exploring a multitude of approaches, such as federated learning, hybrid deep and machine learning models, and optimization algorithms applied to classification and predictive tasks, this review encapsulates the current state of machine learning and deep learning applications in breast cancer diagnosis.

This paper addresses the issue of robust tracking control of Coaxial Octorotor Unmanned Aerial Vehicles (COUAVs). First, a novel Multiple-Input-Multiple-Output (MIMO) super-twisting Sliding Mode Controller (SMC) is proposed for a wide range of systems, in which uncertainties and disturbances are estimated through a Multilayer Perceptron Neural Network (MLPNN). Lyapunov stability theory is used to guarantee the stability of the closed loop system. The real-time update laws for the MLPNN weights are formulated within the stability analysis. The effectiveness of the proposed controller is demonstrated through its application to control uncertain COUAVs in the presence of disturbances. A control structure employing an inner-outer loop framework is adopted to handle the underactuated COUAV dynamics. Simulations involving a commercial heavy-lift COUAV are presented, illustrating the robustness and efficacy of the proposed method.

This paper describes our submission to the SemEval-2023 for Task 6 on LegalEval: Understanding Legal Texts. Our submission concentrated on three subtasks: Legal Named Entity Recognition (L-NER) for Task-B, Legal Judgment Prediction (LJP) for Task-C1, and Court Judgment Prediction with Explanation (CJPE) for Task-C2. We conducted various experiments on these subtasks and presented the results in detail, including data statistics and methodology. It is worth noting that legal tasks, such as those tackled in this research, have been gaining importance due to the increasing need to automate legal analysis and support. Our team obtained competitive rankings of 15th , 11th , and 1st in Task-B, Task-C1, and Task-C2, respectively, as reported on the leaderboard.

To enhance the efficiency of language models, it would involve optimizing their training and inference processes to reduce computational demands while maintaining high performance. The research focuses on the application of model compression, quantization, and hardware acceleration techniques to the Llama model. Pruning and knowledge distillation methods effectively reduce the model size, resulting in faster training times and lower resource consumption. Quantization techniques, including 8-bit and 4-bit representations, significantly decrease memory usage and improve computational speed without substantial accuracy loss. The integration of GPUs and TPUs further accelerates the training and inference processes, demonstrating the crucial role of hardware in optimizing large-scale models. The study highlights the practical implications of those techniques, paving the way for more sustainable and scalable AI solutions.

Hyperdimensional Computing (HDC) is a bioinspired learning paradigm, that models neural pattern activities using high-dimensional distributed representations. HDC leverages parallel and simple vector arithmetic operations to combine and compare different concepts, enabling cognitive and reasoning tasks. The computational efficiency and parallelism of this approach make it particularly suited for hardware implementations, especially as a lightweight, energy-efficient solution for performing learning tasks on resource-constrained edge devices. The HDC pipeline, including encoding, training, and comparison stages, has been extensively explored with various approaches in the literature. However, while these techniques are mainly oriented to improve the model accuracy, their influence on hardware parameters remains largely unexplored. This work presents AeneasHDC, an automatic and open-source platform for the streamlined deployment of HDC models in both software and hardware for classification, regression and clustering tasks. AeneasHDC supports an extensive range of techniques commonly adopted in literature, automates the design of flexible hardware accelerators for HDC, and empowers users to easily assess the impact of different design choices on model accuracy, memory usage, execution time, power consumption, and area requirements.

Marketing Mix Modeling (MMM) traditionally employs statistical metrics such as R-squared, Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), and Akaike Information Criterion (AIC) for model calibration and evaluation. These metrics, while insightful, often fall short in addressing the complexities of real-world scenarios. This report explores advanced analytical techniques, focusing on the Probability Integral Transform (PIT) residuals and Kullback-Leibler (KL) divergence, to enhance the calibration of MMM. Our findings indicate significant deviations from uniformity in the PIT residuals for both optimal and suboptimal models, with the best model demonstrating lower KL divergence, suggesting a closer fit to the expected uniform distribution. This study underscores the value of incorporating advanced metrics for a more nuanced understanding of MMM calibration, beyond conventional evaluation methods.

This study presents a comprehensive analysis of three bipolar-junction-transistor (BJT) amplifier topologies common-emitter, common-collector, and common-base amplifiers-under transient conditions across low and high frequencies. Utilizing MATLAB-Simulink simulations, the response of each amplifier topology to varying frequency inputs is investigated, with a particular emphasis on state space modeling. Beginning with low-frequency models characterized by specific voltage or current sources, the study delves into the steady-state dynamics of output variables, employing state space equations to elucidate amplifier behavior. Notably, the common-emitter amplifier topology exhibits high gains, as corroborated by established literature, underlining its potential for high-performance applications. Conversely, the common-collector and common-base amplifier topologies demonstrate lower gains, reflecting distinct design characteristics. Transitioning to high-frequency operation, the impact of Miller capacitance on gain reduction is explored, especially pronounced in the common-emitter amplifier topology. Despite this, all amplifier configurations remain functional, with state space models validated through meticulous simulations. Simulation parameters are meticulously adjusted to accurately represent real-world scenarios, enhancing the reliability of the findings. These insights underscore the importance of comprehending amplifier behavior across different frequency domains, with state space modeling offering a powerful framework for analysis. By providing detailed insights into BJT amplifier performance under transient conditions, this study facilitates informed decisions regarding amplifier configuration selection and design, tailored to specific application requirements and frequency considerations.

Stability analysis of power electronicsdominated grids is challenging due to the need of manufacturers to protect the intellectual property of converters control systems. Black-box dq impedance models have been proposed to overcome this challenge in grid stability analysis. However, such models are limited to small-signal analyses and are accurate only around a predefined and pre-computed equilibrium point. Conversely, the recently proposed nonlinear Differential-Algebraic Equations (DAE) models of converter-dominated grids can be used for stability analyses in every operating conditions and additionally for generalized power flow analysis and transient time-domain simulations. Despite these advantages, nonlinear state-space converters models require the full knowledge of their control system. To overcome this limitation, this paper proposes a nonlinear grey-box modeling strategy which uses structured perturbations to represent the converters' control system uncertainties. The stability boundaries and control loops sensitivity are analyzed in a multi-vendor converterdominated grid at different operating points through robust control theory. Hardware-In-the-Loop results are provided.

Noninvasive augmented-reality (AR) brain-computer interfaces (BCIs) that use steady-state visually evoked potentials (SSVEPs) typically adopt a fully autonomous goal-selection framework to control a robot, where automation is used to compensate for the low information transfer rate of the BCI. This scheme improves task performance but users may prefer direct control (DC) of robot motion. To provide users with a balance of autonomous assistance and manual control, we developed a shared control (SC) system for continuous control of robot translation using an SSVEP AR-BCI, which we tested in a 3D reaching task. The SC system used the BCI input and robot sensor data to continuously predict which object the user wanted to reach, generated an assistance signal, and regulated the level of assistance based on prediction confidence. Eighteen healthy participants took part in our study and each completed 24 reaching trials using DC and SC. Compared to DC, SC significantly improved (paired two-tailed t-test, Holm-corrected α<0.05) mean task success rate (p<0.0001, µ=36.1%, 95% CI [25.3%, 46.9%]), normalised reaching trajectory length (p<0.0001, µ=-26.8%, 95% CI [-36.0%,-17.7%]), and participant workload (p<0.02, µ=-11.6, 95% CI [-21.1,-2.0]) measured with the NASA Task Load Index. Therefore, users of SC can control the robot effectively, while experiencing increased agency. Our system can personalise assistive technology by providing users with the ability to select their preferred level of autonomous assistance.

Malware detection has become increasingly challenging due to the sophisticated evasion techniques employed by modern threats. A novel situation-aware malware detection framework is introduced, integrating machine learning models with environmental information to enhance detection capabilities. By leveraging contextual data, including network activity, file system changes, user interactions, and system state variables, the framework provides a holistic understanding of system behavior. The detection system achieves significant improvements in accuracy, precision, and recall, outperforming traditional methods. Comparative analyses reveal that incorporating environmental information enhances detection accuracy by 6.4%, while significantly reducing false positives and false negatives. The proposed framework continuously adapts to new threats, ensuring robust defense against a wide range of malware variants. Experimental results highlight the effectiveness of the approach, validated through rigorous testing and comparisons with baseline methods. The study advances the field of cybersecurity by offering an adaptive, context-aware solution that addresses the limitations of existing detection techniques. Future research directions include extending the framework to other operating systems, integrating advanced machine learning techniques, and exploring privacypreserving data collection methods.