Limited Query Black-box Adversarial Attacks in the Real World
We study the creation of physical adversarial examples, which are robust to real-world transformations, using a limited number of queries to the target black-box neural networks. We observe that robust models tend to be especially susceptible to foreground manipulations, which motivates our novel Foreground attack. We demonstrate that gradient priors are a useful signal for black-box attacks and therefore introduce an improved version of the popular SimBA. We also propose an algorithm for transferable attacks that selects the most similar surrogates to the target model. Our black-box attacks outperform state-of-the-art approaches they are based on and support our belief that the concept of model similarity could be leveraged to build strong attacks in a limited-information setting.
Limited Query Black-box Adversarial Attacks in the Real World
We study the creation of physical adversarial examples, which are robust to real-world transformations, using a limited number of queries to the target black-box neural networks. We observe that robust models tend to be especially susceptible to foreground manipulations, which motivates our novel Foreground attack. We demonstrate that gradient priors are a useful signal for black-box attacks and therefore introduce an improved version of the popular SimBA. We also propose an algorithm for transferable attacks that selects the most similar surrogates to the target model. Our black-box attacks outperform state-of-the-art approaches they are based on and support our belief that the concept of model similarity could be leveraged to build strong attacks in a limited-information setting.
Designing a LiDAR topographic navigation system: A novel approach to aid the visually impaired
The WHO reports 2.2 billion people internationally have a form of visual impairment, with Perkins School of Blind adding that 4 to 8 percent (8.8 - 17.6 million people) solely rely on a white cane for navigation. In an interview by Stephen Yin for NPR, visually impaired interviewees claimed that a white cane was ineffective as it failed to detect moving obstacles (ex. bikes), aerial obstacles (ex. falling objects), and it became physically demanding after a prolonged period. This problem can be solved with a headset that integrates LiDAR technology and haptic feedback to provide a real-time assessment of their environment. Theoretically, the device will determine how far an object is from the user and place it into one of three conditionals based on distance (0- 290mm, 310-500mm, 510-1200mm). As the user gets closer to the object, the haptic will vibrate more frequently. The device has 11 LIDAR sensors, beetle processors, and ERM motors so that when the LiDAR detects an object, the device will send a haptic signal in that area. It not only identifies the existence of an object but it tells the user its relative position with a latency period of approximately 2 milliseconds. When testing the device, a simulated walking environment was made. Ten obstacles were included: five below the waist (72”, 28”, 35” and 8.5” tall sticks) and five above the waist (paper suspended 6”, 10”, 48” and 28” from the ceiling). The white cane detected 4.1 obstacles, whereas the device detected 7.3 on average. The LiDAR navigation system is 178% more effective at detecting objects comparatively. Visually impaired individuals no longer must rely on the white cane; rather, using this device, they can detect small, moving, and aerial objects at a much faster, and more accurate speed.
An Efficient and Accurate Super-Resolution Approach to Low-Field MRI via U-Net Architecture With Logarithmic Loss and L2 Regularization
Low-field (LF) MRI scanners have the power to revolutionize medical imaging by provid- 27 ing a portable and cheaper alternative to high-field MRI scanners. However, such scanners are usu- 28 ally significantly noisier and lower quality than their high-field counterparts. This prevents them 29 from appealing to global markets. The aim of this paper is to improve the SNR and overall image quality of low-field MRI scans (called super-resolution) to improve diagnostic capability and, as a result, make it more accessible. To address this issue, we propose a Nested U-Net neural network architecture super-resolution algorithm that outperforms previously suggested super-resolution deep learning methods with an average PSNR of 78.83 ± 0.01 and SSIM of 0.9551 ± 0.01. Our ANOVA paired t-test and Post-Hoc Tukey test demonstrate significance with a p-value < 0.0001 and no other network demonstrating significance higher than 0.1. We tested our network on artificial noisy downsampled synthetic data from 1500 T1 weighted MRI images through the dataset called the T1- mix. Four board-certified radiologists scored 25 images (100 image ratings total) on the Likert scale (1-5) assessing overall image quality, anatomical structure, and diagnostic confidence across our architecture and other published works (SR DenseNet, Generator Block, SRCNN, etc.). Our algo- rithm outperformed all other works with the highest MOS, 4.4 ± 0.3. We also introduce a new type of loss function called natural log mean squared error (NLMSE), outperforming MSE, MAE, and MSLE on this specific SR task. Additionally, we ran inference on actual Hyperfine scan images with successful qualitative results using a Generator RRDB block. In conclusion, we present a more ac- curate deep learning method for single image super-resolution applied to low-field MRI via a 45 Nested U-Net architecture.
BeeMind AI: Development of an AI-Based System to Assess Honeybee Health, Behavior, and Nutrient Effects on Learning and Memory
Due to their pollination services, honeybees are one of the most ecologically vital animals, being singlehandedly responsible for nearly 80% of global agricultural pollination [1]. However, in recent years, they have experienced large declines in populations, and as a survey reported roughly 50% of beekeepers in the US lost their honeybee colonies [2]. These losses are experienced globally due to a combination of many factors, including but not limited to habitat loss, pesticides, climate change, and other invasive species [3, 4]. One of the biggest factors attributed to the decline of honeybee colonies is the usage of pesticides, specifically neonicotinoids [3-6]. Neonicotinoid compounds have been used globally since their introduction in the early 1990s [4]. Studies have shown that neonicotinoids can have both sublethal and lethal effects on honeybees, depending on the dosages that they are exposed to, as neonicotinoids bind to nervous system receptors of honeybees [7]. These effects can range from behavior changes to altered motor functions [7-9]. Among the reported effects, one of the more significant ones is the effect of neonicotinoids on honeybee learning and memory [10, 11]. Additionally, there is a lack of availability for methods of monitoring of honeybee hives, essentially meaning that the only methods to track honeybee health are through obtrusive physical methods of inspection. This paper aims to develop a novel AI-based honeybee health assessment system, able to monitor beehives using the following functions: continuous temperature and humidity monitoring both inside and outside the hive, as well as video and audio recording to assess honeybee health as well as population. In addition, this system can be used for honeybee-related studies such as nutrition effects and evaluation on health, learning, and memory. To do this, four types of nutrition have been studied and their effects have been analyzed by a deep learning approach.
Decoding Climate Resilience: Functional Profiling of Protein Phosphatase 2C Family Genes for Abiotic Stress Tolerance in Rice
Problem • Rice is the primary cereal crop consumed by nearly half the population worldwide • By 2050, there will be a 50% increase in demand for rice • The world’s poor populations depend more on rice, both for income and consumption, than any other food. Rice is the single-largest source of employment and income for rural people • Worldwide, 51–82% of agricultural crop yield is lost annually due to abiotic stress due to climate change • Climate change causes extreme temperatures, erratic rainfall, dangerous droughts, and increased salinity from rising sea levels Solution • To adapt to abiotic stress, rice has intricate signaling pathways, particularly those mediated by the phytohormone abscisic acid (ABA), that cause an increase in stress tolerance • Clade A genes of the Protein Phosphatase 2C (PP2C) gene family are known to be negative regulators of the ABA signaling pathway. • “Deleting” these genes activates the ABA pathway and increases stress tolerance in rice without inducing stress CRISPR gene editing technology is the ideal solution Research Goal • While the role of PP2C genes in stress response is recognized, there is a gap in understanding the specific genes within this family that contribute significantly to stress signaling. Furthermore, there is a need for a detailed investigation into the effects of targeted CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) genome editing on rice stress response pathways.
Strict Inequalities for the n-crossing Inequality
In2013,Adamsintroducedthen-crossingnumberofaknotK,denoted by cn(K).Inequalities between the 2-, 3-, 4-, and 5-crossing numbers have been previously established.We prove c9(K)≤c3(K)−2 for all knots Kthat are not the trivial, trefoil, or figure-eightknot.Weshowthisinequalityisoptimalandobtainpreviouslyunknownvalues for c9(K).
Quantitative environmental DNA metabarcoding for the enumeration of Pacific salmon (Oncorhynchus spp.)
Understanding species abundance is critical to managing and conserving planetary biodiversity. Pacific salmon (Oncorhynchus spp.) are keystone species of cultural, economic, and ecological importance in Alaska and especially Southwest Alaska. Traditional methods of enumerating salmon such as weirs and visual surveys are often costly, time-intensive, and reliant on taxonomic expertise. Environmental DNA (eDNA), which identifies and quantifies species based on DNA they shed in their habitats, is a potential cost- and time- saving alternative. The relative ease of collecting eDNA samples also enables citizen scientist involvement, expanding research coverage. Currently, more research is required to define eDNA’s potential and limits. This project investigates whether quantitative eDNA metabarcoding can accurately quantify the abundances of six fish species: the five Pacific salmon species plus rainbow trout. Water samples were collected from eight creeks in the Wood River watershed of Southwest Alaska. eDNA metabarcoding and subsequent bioinformatics processing produced a read count for each species. These were compared to visual survey counts, taken to be the true counts for the purposes of this study. Data analyses showed a positive, linear relationship between visual survey count and eDNA count for sockeye salmon. The regressions were significant for both the early (p = 0.089) and late (p = 0.030) sampling dates when 𝛼 = 0.10. eDNA detections of non-sockeye species generally corresponded to visual survey observations of species presence or absence. Overall, the results of this study support eDNA’s potential to be an alternative or supplement to standard methods for the enumeration of fish species.