FFTstack: Integrating Fourier Transform and Residual Learning for Arctic Sea Ice Forecasting

Overview

Arctic sea ice plays integral roles in both polar and global environmental systems, notably ecosystems, communities, and economies. As sea ice continues to decline due to climate change, it has become imperative to accurately predict the future of sea ice extent (SIE). Using datasets of Arctic meteorological and SIE variables spa nning 1979 to 2021, we propose architectures capable of processing multivariate time series and spatiotemporal data. Our proposed framework consists of ensembled stacked Fourier Transform signals (FFTstack) and Gradient Boosting models. In FFTstack, grid s earch iteratively detects the optimal combination of representative FFT signals, a process that improves upon current FFT implementations and deseasonalizers. An optimized Gradient Boosting Regressor is then trained on the residual of the FFTstack output. Through experiment, we found that the models trained on both multivariate and spatiotemporal time series data performed either similar to or better than models in existing research. In addition, we found that integration of FFTstack improves the performanc e of current multivariate time series models. We conclude that the high flexibility and performance of this methodology have promising applications in guiding future adaptation, resilience, and mitigation efforts in response to Arctic sea ice retreat. D.I.V.A: Spatial Navigation for the Visually Impaired using Convolutional Neural Networks, Stereoscopy and Sensor Fusion Karthik Muthukkumar Urbana High School, Frederick, MD Visual impairment remains a challenging issue in biomedical sciences with a need for universally effective and affordable solutions. Severe conditions like cataracts, glaucoma, macular degeneration, and complete blindness due to visual cortex damage continue to lack economic remedies. Current solutions such as the white cane and optical surgeries are restrictive in mobility, invasive, or costly. To address this, we propose D.I.V.A, a novel wearable device to assist the visually impaired in unfamiliar enviro nments. The device utilizes a stereoscopic camera, RGB camera, gyroscope, custom PCB, proprietary computer vision software, and a novel PDMS -based microactuator grid. Through a custom U -net neural network trained with 80 object classes, the device detects and classifies obstacles in the surrounding environment of the user via semantic segmentation. Furthermore, it communicates this information through a GUI-commanded personalized spatial audio system, in conjunctio n with indicative actuation of the tactile interface that represents the front-facing environment on the palm of the user. The proposed aid is affordable, costing approximately $270, making it accessible to individuals facing financial constraints. To stat istically justify the proposed system and its performance, successful experiments were conducted in indoor environments to test the software accuracy, sensory input of the tactile interface, and viability of D.I.V.A in an unfamiliar path. The experimental results demonstrate that the proposed assistive device performs all its functions with high accuracy, allowing visually impaired people to feel safe and comfortable in an indoor environment. We envision this device significantly improving the lives of visu ally impaired individuals, especially those with limited access to expensive solutions. Michigan What’s USP with Multiple Sclerosis? Devarshi Dalal Troy High School, Troy, MI Multiple sclerosis (MS) is characterized by immune -mediated demyelination of nerve fibers, but its molecular mechanisms remain unclear. Here, we integrate bioinformatics analysis and experimental validation to investigate Ubiquitin-specific protease 33 (USP33) in MS. Using gene expression data from MS brain tissue, we identify USP33 as a significantly dysregulated gene in MS. Looking at USP33’s molecular pathways, we see it implicated in the Ubiquitin -mediated proteolysis pathway and immune activation, linking it to MS pathogenesis. Immunohistochemical staining confirms USP33 overexpression in MS brain tissue and suggests its potential as a therapeutic target. Functional annotation reveals regulatory mechanisms governing USP33 expression, including transcrip tion factors and RNAi. This approach demonstrates USP33's role in MS pathogenesis, offering insights for targeted therapies. Biomimetic Airfoil Optimization to Supplement Flight Efficiency in Unmanned Aerial Vehicles Dhruv Hegde Salem High School, Canton, MI Wing-based unmanned aerial vehicles (UAVs) are aircrafts designed to be operated remotely in critical areas for assessment mitigation and object identification. Traditionally, UAVs employ fixed airfoil systems, limiting their adaptability and efficiency, especially in dynamic flight conditions where Reynold’s number is eminently high. More specifically, the rigid structure of the airfoil prevents efficient lift generation, rapidly depletes fuel during transition periods, faces premature stall during altitud e changes, and are prone to vulnerabilities in maneuvering. This study suggests driving improvement in UAV performance through the introduction of avian features into the structure of the airfoil and air profile. Through rigorous analysis of avian exoskele tons, features, and supracoracoideus muscles, several individualized features were adapted and modeled through computer-aided design (CAD) software. The design included in this study incorporates serrated edges and vortex dividers, alula -inspired air profi le extensions, morphing airfoil modules with servo and rotary systems, and a piston -based passive flight mechanism to increase flight efficiency and retrievability of military UAVs. The paper also delves into the discussion of feasibility of implementation and material considerations in a flying prototype. Additionally, Computational Fluid Dynamics (CFD) simulations, wind tunnel testing, and mathematical modeling were employed to assess the impact of these biomimetic features and demonstrate theoretical fea sibility in subsonic, critical environments ranging in Reynold’s number from 50,000 to 1,000,000. The integration of these various structural pieces promises a synergistic effect, resulting in an estimated 35% overall improvement in aerodynamic performance from current UAV standards. KTV Transformer: A Novel Multi-scale Knowledge Transfer Vision Transformer for 3D Brain Vessel Segmentation Michael Hua Cranbrook Kingswood School, Bloomfield Hills, MI A growing body of evidence in recent studies shows that small cerebrovascular abnormalities are the cause of many brain disorders. In order to facilitate the robust and precise 3D cerebrovascular extraction and quantification from in-vivo Magnetic Resonance Imaging (MRI) data, this paper presents a novel Multi- scale Knowledge Transfer Vision Transformer (i.e., KTV-Transformer) for 3D vessel segmentation. First, it uniquely integrates convolutional embeddings with Transformer in a U -net architecture, which simultaneously responds to local receptive fields with convolution layers and global contexts with transformer encoders in a multi-scale fashion. Therefore, it intrinsically enriches local vessel feature and simultaneously promotes global vessel connectivit y and continuity for a more accurate and reliable segmentation. Furthermore, to enable using relatively low -resolution (LR) images to segment fine scale vessels, a novel knowledge transfer network is designed to explore the inter -dependencies of data and automatically transfer the knowledge gained from high-resolution (HR) data to the low-resolution handling network at multiple levels, including the multi -scale feature levels and the decision level, through an integration of multi -level loss functions. The modeling capability of fine -scale vessel data distribution, possessed by the HR image transformer network, can be transferred to the LR image transformer to enhance its knowledge for fine vessel segmentation. Based on the vessel segmentation results, quantitative metrics can be generated for computer -aided diagnosis of brain diseases and scientific discovery. Extensive experimental results on public image datasets have demonstrated that my proposed method outperforms all other state-of-the-art deep learning methods. A Deep Learning Approach to Dementia Identification Using Clock Drawing Figures Dhruti Pattabhi Canton High School, Canton, MI Dementia is the loss of cognitive functioning across multiple domains and impedes functions like memory, language skills, problem -solving and visual perception. Globally, dementia affects around 47 million people. The clock drawing test (CDT) is a neuropsychometric clinical test used to assess a variety of cognitive functions. In Dementia, CDTs are interpreted both quantitatively and qualitatively and are widely used screening tools. However, human analysis of the clock drawing tests may leave room for misinterpretation errors. Unbiased testing in healthcare is cardinal for ensuring optimal patient prognosis, and supervised machine learning can be accordingly used for objective image analysis. Here, deep learning is used to objectively analyze and categorize CDTs as either dementia positive or negative. We achieved good results and display that deep learning mechanisms are an effective way to identify dementia in patients. The model uses characteristics in the patient CDTs to diagnose individuals instead of interpreting them on a scoring system. This is u seful as there are numerous ways to score CDTs and this method is consistent across all tests. It also provides insight into how dementia may alter parts of the brain involved with CDTs. With the ease of model development, this model could easily be transf ormed into more accessible options, like an online app.