Similarity searches are a critical task in data mining. Nearest neighbor similarity search over geometrical shapes - polygons and trajectories - are used in various domains such as digital pathology, solar physics, and geospatial intelligence. In digital pathology for tumor diagnosis, tissues are represented as polygons and Jaccard distance - ratio of areas of intersection to union - is used for similarity comparisons. In solar physics for predicting solar flares, the query object and the dataset is made up of polygons representing solar events. In geospatial intelligence, similarity search is used to geo-locate a shape or a contour in global reference datasets. The current literature, while rich in methods for textual and image datasets, is lacking for geometric datasets. This project will develop scalable similarity search systems on polygonal and trajectory datasets. It will produce benchmark datasets of polygonal queries and responses for the research community and inform the data mining techniques which employ similarity primitives. It will help introduce student projects for courses on parallel, distributed, high performance, and data intensive computing, data mining, and spatial computing. This will also train PhD students, including those at a Hispanic Serving Institution.
Researchers: Buddhi Ashan M. K. NSF Award #2313039Micromobility vehicles, such as battery-powered e-scooters, are rapidly gaining popularity in urban communities. Yet they also present significant safety, user privacy, infrastructure, and planning challenges that must be addressed. The research community ?including computer and data scientists, engineers, and urban planners? has started responding to these challenges, but the progress has been ad hoc and slow. This can be primarily attributed to a lack of micromobility infrastructure for collecting diverse rider, mobility, and contextual data in realistic settings and environments. There is a critical need for a large-scale and easily accessible research instrument to enable such data collection. To this end, this project will design, develop, deploy, and manage ScooterLab, a community research infrastructure comprising a highly customizable fleet of micromobility vehicles. These battery-operated vehicles will be retrofitted with heterogeneous sensing and remote communication and control capabilities to crowd-sense data related to riders? mobility, context, and environment, enabling research at the confluence of (1) multiple computing disciplines, including machine learning, computer vision, image processing, high-performance computing, big data analytics, and privacy-enhancing technologies; and (2) (micro)mobility, urban planning, and transportation research. The project will provide community researchers with well-designed and usable web interfaces for requesting the deployment of customized sensing experiments and for accessing carefully curated datasets from past experiments and trials. The project will also conduct periodic community outreach and engagement activities, including workshops to promote the testbed?s use and share the outcomes of research activities enabled by the testbed data.
Researchers: Buddhi Ashan M. K., Ahmer Patel, Christina Duthie ScooterLab NSF Award #2234516Fault detection in automotive engine systems is one of the most promising research areas. Several works have been done in the field of model-based fault diagnosis. Many researchers have discovered more advanced methods and algorithms for better fault detection on the highly nonlinear dynamic engine of any automotive system. Here we propose an AI Transformer model for fault classification and fault prediction in Diesel engines. In the modern era, misdiagnosis and robustness are major concerns in any automotive vehicular system. The reason behind this is, that many manufacturers make their products and sell their vehicles all across the world. So, it is a challenging problem for manufacturers to come up with one resilient fault detection method concerning wide variations in weather, different driving styles, heavy traffic conditions, etc. Second, With the advancement of technologies, the development of autonomous vehicles/engines is a special concern
Researchers: Priyanka KumarClimate Change and global warming have impacted the Antarctic Sea region, environment, communities, and the entire ecosystem directly and indirectly. With the seasonal temperature variations, sea-ice extent and volume are increasing/decreasing in Antarctica. With the advancement of technology, the progress in machine learning (ML) and deep learning (DL) for Earth science is promising, but there is still a gap in the geoscience domain and geoscientist communities are still facing challenges that make it hard for more people to use. The key challenges include the absence of publicly available benchmark training datasets across all science disciplines, lack of interoperability among data sources and formats, limited availability of baseline pre-trained models adaptable to various Earth observation types, and difficulties in structuring label or target values, such as oceanic measurements from drifting buoys
Researchers: Priyanka KumarThe purpose of this research project is to use a Deep Q-Network (DQN) technique to optimize the performance of Optimistic Parallel Discrete Event Simulations (PDES). In order to reduce rollbacks and speed up execution overall, the main objective is to dynamically optimize crucial simulation parameters, such as the number of buckets in the calendar queue, the number of threads, and the size of the time frame. Through reinforcement learning, our own DQN model—a kind of deep neural network—interacts with the PDES environment to determine the best scheduling policy. Comprehensive simulation metrics, including the total number of rollbacks, average rollback duration, maximum load, and execution time, are included in the DQN agent's state space. In order to achieve a balance between simulation accuracy and computing efficiency, the DQN agent makes intelligent adjustments to the simulation settings based on these state observations. This method opens the door for more effective and economical PDES implementations across a range of domains by automating the parameter tuning process and offering a scalable solution for intricate, large-scale simulations.
Researchers: Shobnam Roksanaongoing
Researchers: Priyanka Rani SahaThis research investigates the creative use of binary classification graphs (BCGs) to memristor crossbar circuit design for energy efficiency. Memristors are unusual in that they can maintain resistance based on past electrical activity. This property makes them attractive for use in cutting-edge technologies like neuromorphic circuits, spintronic devices, and ultra-dense information storage. The work focuses on designing in-memory, energy-efficient, and compact memristor crossbar circuits utilizing BCGs, which rely on bit values of input characteristics instead of conventional threshold-based decision trees. By using this method, the circuits become more scalable and durable, able to withstand manufacturing faults and handle high crossbar sizes. Using SPICE models, the authors validate their designs and show that their designs have better energy efficiency and circuit size than previous approaches. Using a variety of machine learning datasets, they developed decision tree models, which they then converted into matching crossbar designs. The report ends with possible directions for future research, such as expanding the designs to include flow-based computing into more intricate neural networks and extending the designs to build random forests by merging numerous crossbar arrays. This study demonstrates the great potential of classification graphs and memristor technology in building more reliable and effective computing systems, opening the door for improvements in energy-efficient circuit design
Researchers: Shobnam RoksanaPolygonal geometric operations are fundamental in domains such as Computer Graphics, Computer-Aided Design, and Geographic Information Systems. Handling degenerate cases in such operations is important when real-world spatial data are used. The popular Greiner-Hormann (GH) clipping algorithm does not handle such cases properly without perturbing vertices leading to inaccuracies and ambiguities. In this work, we parallelize the O(n2 )-time general polygon clipping algorithm by Foster et al. which can handle degenerate cases without perturbation. Our CREW PRAM algorithm can perform clipping in O(log n) time using n + k number of processors with simple polygons, where n is the number of input edges and k is the number of edge intersections. For efficient GPU implementation, we employ three effective filters which have not been used in prior work on polygon clipping: 1) Commonminimum-bounding-rectangle filter, 2) Count-based filter, and 3) Line-segment-minimum-bounding-rectangle filter. They drastically reduce O(n2 ) candidate edge pairs comparisons by 80%- 99%, leading to significantly faster parallel execution. In our experiments, C++ CUDA-based implementation yields up to 40X speedup over real-world datasets, processing two polygons with a total of 174K vertices on an Nvidia Quadro RTX 5000 GPU compared to the sequential Foster’s algorithm running on an Intel Xeon Silver 4210R CPU.
Researchers: Buddhi Ashan M. K.