KAKhizar Anjum

Khizar Anjum

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Khizar Anjum has a total of 23 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Markov Decision Process-based Real-time Navigation for Autonomous Aerial Vehicles Using Video

    by Khizar Anjum, Parul Pandey, Vidyasagar Sadhu, Roberto Tron, Dario Pompili

    Autonomous aerial vehicle navigation is critical for applications ranging from search and rescue operations to warehouse automation, requiring real-time decision-making under severe computational and power constraints. While traditional geometric approaches using 3D point clouds provide accurate navigation, they impose prohibitive computational overhead for resource-constrained drones, and existing computer vision algorithms for semantic navigation create a fundamental performance bottleneck where static parameter configurations fail to adapt to varying environmental conditions. This paper addresses this gap through a novel Markov Decision Process (MDP) framework that dynamically optimizes object detection parameters based on input characteristics, intelligently selecting algorithm configurations from both feature-based (HoG/Edgeboxes with SVMs) and neural network (CNN) detection pipelines while utilizing classifier confidence scores as quality indicators. Comprehensive evaluation across KUL traffic sign datasets, custom BU-RU drone-collected footage, and hardware-in-the-loop AirSim simulations demonstrates significant performance improvements: 20-70% execution time reduction while maintaining 98-100% detection accuracy compared to static approaches, real-time 30fps processing at 480 $$\times $$ × 270 resolution, and up to 18 $$\times $$ × improvement in frame processing rates during closed-loop navigation tasks. These computational efficiency gains translate to substantial energy savings, achieving 6.9W power reduction on NVIDIA Jetson TX2 hardware through proportional decreases in processing time and computational load.

  • Open Access

    Pruning 3D Convolutional Neural Networks via Channel Independence

    by Yang Sui, Khizar Anjum, Dario Pompili, Bo Yuan

    Network pruning provides a promising approach for deploying costly Deep Neural Network (DNN) models on resource-constrained devices. However, most existing pruning works focus on compressing traditional convolutional DNN models for image-related classification and object-detection tasks in 2D scenarios. Due to the complicated structure of 3D CNN models, pruning such models has not been well studied. In this paper, we analyze the different properties between 2D and 3D tasks, then propose a Filter/Depth-wise Independence Score (FDIS) to evaluate the importance of each filter. In addition, we adopt several granularity schemes to improve the performance of the proposed method. To achieve fine-grained pruning, we prune the networks gradually using an iterative pruning procedure. In addition, we experimentally show that weights with low independence scores contain less important information, enabling the removal of filters without serious accuracy degradation. Our proposed FDIS-based approach maintains high accuracy with certain FLOP reduction and practical acceleration.


  • Open Access

    Domain Specific Benchmarks for Evaluating Multimodal Large Language Models

    by Khizar Anjum, Muhammad Arbab Arshad, Kadhim Hayawi, Efstathios Polyzos, Asadullah Tariq, Mohamed Adel Serhani, Laiba Batool, Brady D. Lund, Nishith Reddy Mannuru, Ravi Varma Kumar Bevara, et al.

    Large language models (LLMs) are increasingly being deployed across disciplines due to their advanced reasoning and problem-solving capabilities. To measure their effectiveness, various benchmarks have been developed that measure aspects of LLM reasoning, comprehension, and problem-solving. While several surveys address LLM evaluation and benchmarks, a domain-specific analysis remains underexplored in the literature. This paper introduces a taxonomy of seven key disciplines, encompassing various domains and application areas where LLMs are extensively utilized. Additionally, we provide a comprehensive review of LLM benchmarks and survey papers within each domain, highlighting the unique capabilities of LLMs and the challenges faced in their application. Finally, we compile and categorize these benchmarks by domain to create an accessible resource for researchers, aiming to pave the way for advancements toward artificial general intelligence (AGI).


  • Battery-Less Implantable Continuous EEG Monitoring via Anisotropic Diffusion

    by Khizar Anjum, Dario Pompili

    In this article, we introduce a groundbreaking approach for ultra-low-power hybrid analog-digital processing of multimodal physiological data at multiple locations, emphasizing EEG signals. We propose an innovative analog Convolutional Processing Unit (CvPU) that uniquely harnesses the properties of anisotropic diffusion in electrical circuits for convolution. This novel use of anisotropic diffusion-driven convolution sets our work apart. Additionally, we present a controller architecture that allows for the sequential execution of multiple consecutive convolutional layers using the same CvPU array. The proposed neural network architecture to detect seizures using EEG signals is evaluated on a publicly available clinical dataset. Our CvPU array-based convolution’s performance and feasibility metrics have been assessed using SPICE simulation software. Furthermore, we have delved deep into studying the scalability of our approach in terms of power and space and its feasibility for battery-less and implantable applications and have compared it with both digital and hybrid analog-digital methods.


  • Ultra-Low Power Analog Folded Neural Network for Cardiovascular Health Monitoring

    by Yung-Ting Hsieh, Khizar Anjum, Dario Pompili

    Wearable sensors are increasingly used for continuous health monitoring, but their small size limits battery capacity, affecting user experience and monitoring capabilities. To overcome this, we introduce an ultra-low power analog Folded Neural Network (FNN) for physiological signal processing in a batteryless fashion. Our proposed FNN, by serializing computation, provides several benefits over traditional analog implementations, such as lower space, lower power consumption, and lower peak-to-average power ratio. We evaluate our method extensively using a dataset designed for ECG-based screening and diagnosis. Our analysis considers factors such as thermal noise, spatial requirements, and power consumption. Additionally, we evaluate detection performance, investigating various parameters of the proposed FNN. This evaluation provides insights into the optimal configuration for accurate anomaly detection. We observe a good trade-off for accuracy around 6 layers and a hidden size of 30 and further demonstrate that such architecture could be implemented in a wearable device and executed in a batteryless fashion.


  • Open Access

    The Power Of Simplicity: Why Simple Linear Models Outperform Complex Machine Learning Techniques -- Case Of Breast Cancer Diagnosis

    by Muhammad Arbab Arshad, Sakib Shahriar, Khizar Anjum

    This research paper investigates the effectiveness of simple linear models versus complex machine learning techniques in breast cancer diagnosis, emphasizing the importance of interpretability and computational efficiency in the medical domain. We focus on Logistic Regression (LR), Decision Trees (DT), and Support Vector Machines (SVM) and optimize their performance using the UCI Machine Learning Repository dataset. Our findings demonstrate that the simpler linear model, LR, outperforms the more complex DT and SVM techniques, with a test score mean of 97.28%, a standard deviation of 1.62%, and a computation time of 35.56 ms. In comparison, DT achieved a test score mean of 93.73%, and SVM had a test score mean of 96.44%. The superior performance of LR can be attributed to its simplicity and interpretability, which provide a clear understanding of the relationship between input features and the outcome. This is particularly valuable in the medical domain, where interpretability is crucial for decision-making. Moreover, the computational efficiency of LR offers advantages in terms of scalability and real-world applicability. The results of this study highlight the power of simplicity in the context of breast cancer diagnosis and suggest that simpler linear models like LR can be more effective, interpretable, and computationally efficient than their complex counterparts, making them a more suitable choice for medical applications.


  • On-Board Deep-Learning-Based Unmanned Aerial Vehicle Fault Cause Detection and Classification via FPGAs

    by Vidyasagar Sadhu, Khizar Anjum, Dario Pompili

    With the increase in the use of unmanned aerial vehicles (UAVs)/drones, it is important to detect and identify causes of failure in real time for proper recovery from a potential crash-like scenario or postincident forensics analysis. The cause of crash could be either a fault in the sensor/actuator system, a physical damage/attack, or a cyber attack on the drone's software. In this work, we propose novel architectures based on deep convolutional and long short-term memory neural networks to detect (via autoencoder) and classify drone misoperations based on real-time sensor data. The proposed architectures are able to learn high-level features automatically from the raw sensor data. Empirical results show that our solution is able to detect (with over 90% accuracy) and classify various types of drone misoperations [with about 99% accuracy (simulation data) and up to 85% accuracy (experimental data)]. Furthermore, with the help of field programmable gate array-based hardware acceleration, we achieved a speedup of 40x ( $\sim\!\text{2.6 ms}$ ) for detection, while consuming half the amount of power compared with onboard GPU devices, such as NVIDIA Jetson TX2.


  • Open Access

    Demo: Deep Joint Source-Channel Coding for Underwater Image Transmission

    by Khizar Anjum, Zhuoran Qi, Dario Pompili

    Traditional methods for coding underwater acoustic communications are bound to be surpassed by methods optimizing for source-channel coding jointly. However, the complexity of joint-optimization has thwarted successful breakthroughs in this area. We, therefore, present a novel approach, where we model the coding problem as the translation problem of the input sequence to another ‘language’, depending on the estimated channel conditions.