Poster Sessions

Session Poster-1

Poster Session 1

11:30 AM — 12:30 PM EDT
Aug 1 Sat, 11:30 AM — 12:30 PM EDT

Utilizing Different Types of Synthetic Fabrics as a Cement Reinforcement for Concrete Tile Roofing

Blessed Isaac C Conde and Daniel Jerlan S Coladilla (La Salle Green Hills, Philippines)

This study aims to alleviate the problem of concrete deterioration by introducing a material to be used as a cement reinforcement to concrete roof tiles that will decrease its water absorption percentage, without decreasing its flexural tensile strength score. Five sample groups consisting of four experimental and one control group of downgraded (modified size) flat concrete roof tiles were used for experimentation. The experimental groups were reinforced with equal amounts (15 grams) of acrylic, nylon, polyester, or rayon fabric shreds, respectively. Both experimental and control groups were uniformly treated with water to cement and cement to sand optimal volume ratio. Using the p-value method on Kruskal-Wallis H-test (? = 0.01, N = 10, df = 4) in both tests, results show that each sample group distribution had no significant difference among the other groups in terms of both test scores. Despite both p-values (p ? 0.081) falling to the null hypothesis non-rejection region of p ? 0.01, it may still be observed that all experimental groups scored lower flexural tensile strength and water absorption percentage scores. The results showed the potential of adding nylon to reduce the water absorption of concrete. Nylon-added samples have the least mean percentage of water absorbed and mean flexural tensile strength score while the control group had the highest mean scores on both tests. Consequently, adding synthetic fabrics lessened water absorption and flexural tensile strength scores. It is recommended to consider other properties under the Roofing Tile Association of Australia (RTAA) and to explore if increasing the sample size will show significant differences in the scores across all sample groups.

Efficiency of organic deodorizer in removing foul smell in shoes

Ben Angelo P. Sebastian (La Salle Green Hills, Philippines); Alejandro Jose C. Javier, N. a. (La Salle GreenHills, Philippines); Ryan Andre Mari Flores Chua and Byron Miguel De Guzman Traje (La Salle Green Hills, Philippines)

Bromodosis or foot odor affects many people around the world. Shoe deodorizers are used to prevent the malodor, but most of the products available may use harmful materials that do not fully eliminate its cause. These products contain substances that degrade the environment (Steinmann, 2016), and pose as a health hazard (Engelund, et al., 2005). The researchers from a private school in Mandaluyong City plan to utilize natural materials (activated charcoal and freeze- dried lemon) in creating a shoe deodorizer.

Previous studies have indicated that foot odor was derived from Isovaleric acid, which is produced when Staphylococcus epidermidis degrades leucine present in sweat (Ara, et al., 2006). When the bacteria in feet eat amino acids such as leucine, it produces a by-product called Isovaleric acid which causes the foul smell. Various studies have also been made regarding the use of lemon and activated charcoal to reduce foul smell (Otang & Afolavan 2016, Tada, et al., 2016). The researchers will focus on the effectiveness of freeze- dried Citrus limon, (Lisbon variety) in removing malodor.

A revised version of the Modified Kirby Bauer disc diffusion test method was done by the Industrial Technology Development Institute, Department of Science and Technology (ITDI-DOST), the prime government research agency in the Philippines. The researchers were able to verify the antibacterial properties of freeze-dried lemons. Results showed that the sample freeze -dried lemons (10mm), produced complete inhibitory activity with a Total Mean of Inhibition of 12.74mm, and with mild reactivity against the test organism Staphylococcus aureus, a common species of the normal microbiota of the skin. It may be concluded from the research that freeze-dried lemons seem to have a good potential to inhibit foul smell, and may be an effective ingredient for a shoe deodorizer.

Further studies on the antimicrobial properties of freeze-dried lemons and activated charcoal against S. aureus and other skin microorganisms should be made for a more comprehensive report. Factors such as lemon species, parts, form, may also contribute to its effectivity. These findings may support the creation of an effective and organic shoe deodorizer.

The Utilization of the Feynman Technique in Paired Team Teaching Towards Enhancing Grade 10 ANHS Students' Academic Achievement in Science

Ronnel Ian A. Ambion, Rainier Santi C. De Leon, Alfonso Pio Angelo R. Mendoza and Reinier M. Navarro (La Salle Green Hills, Philippines)

Several scholars in the field of science instruction suggested that various challenges can hinder an individual in learning (Chin et al., 2016; Drinkwater et al., 2014; Huang et al., 2015). Furthermore, the help of other people enhances learning. This study applied the Feynman Technique and the practice of paired team teaching to identify the challenges of students in understanding the concept of evolution in a high school Biology class. The researchers tested 20 students from a Grade 10 Adult Night High School class of a private school in the Philippines. The control group took the class without the intervention while the experimental group was introduced to the intervention. Purposive sampling was used based on the student's class standing before the experiment. A ten-item assessment on the evolution of horses was done after the experiment. The study revealed that there seemed to be no significant difference since the t-score of both groups, 0.49, is far from the critical value, 1.75. It is recommended that the intervention should be tested in a larger population, regardless of grade level and a science subject. This is to validate what other variations of inputs can be created from the Feynman Technique.

Edison High School iSTEM Club: A Model for Educational Excellence in STEM

Sunrit Panda, Aditi Deshmukh, Gunjan Adya and Ali Ahmed (Edison High School & iSTEM Club, USA)

Currently, STEM education is a necessity for students at all levels. The iSTEM club represents a model for engaging and teaching students the necessary. Through excellent leadership, intensive programs to educate students, and opportunities to motivate students, the iSTEM club provides enhanced STEM education to students. With these methods, the iSTEM club has benefited the community through book drives, educational programs from elementary school students, and field trips to expand the student body's knowledge. By participating in this club, students are able to reinforce knowledge with hands-on experiences and build up experience in order to perform well in society. As a result, the beneficial interaction between STEM clubs and STEM education structure is revealed.

Who knows more about germs? Adults or first graders?

Rehaam Siddiqui (Pillars Preparatory Academy, USA)

With more people getting sick, it's very important to understand what germs are and how they spread. A common way to determine where germs live is to take a piece of bread and wipe it on some surface, and then measure how quickly mold grows. While this experiment is commonly done, I wanted to extend this idea to see how well people understand where germs live.

My poster includes some background information about germs and how to keep safe from getting sick. Since germs are too small to see, the bread experiment is a good way to easily see how quickly germs can grow. To figure out who knows more about germs, I made a list of questions asking where germs are likely to be.

First, I asked a group of adults where they thought germs are. I then asked my classmates (in first grade) what they thought. Then, by comparing the answers against real results that the moldy bread generates, I will answer the question of who knows more about germs: adults or first graders.

An Optimization of Computational Resources Allocation for Multi-MEC and Cloud Networks

Ally Y Du (The High School Affiliated to Renmin University of China, China)

Mobile edge computing (MEC) technology reduces network congestion and enables flexible application deployment. It runs applications and performs related processing tasks closer to mobile customer, so that the network congestion is reduced. Nowadays, MEC is often combined with cloud computing. It has been studied that an optimization of MEC with cloud computing can satisfy not only enough computing power to end-user applications but also with low latency

This paper studies a real-life MEC application on mobile video surveillance and analytics. We started from a simple practical model where the edge node (EDGE) is an AI chip enabled device handling multiple number of video cameras, and transmitting/receiving video streams and signals to a 4G or 5G base station (BS), and then the BS is communicating with the cloud center via optical fiber. In this case, if we offload some functions of EDGE to the cloud center where the computing capability is much stronger, the number of streams, that is, the number of cameras handled, can be increased, and the overall latency could also be improved.

With this expectation, we investigate real mobile video surveillance systems used for public traffic and public security, and propose a mathematical system model based on the real mobile video surveillance systems to quantify the relations among the overall system processing delay, the MEC/Cloud offload ratio and the communication channel signal-noise ratio (SNR). We make calculations and obtain the following findings using MATLAB. 1) For any offload ratio, there is a point at which the latency reaches its floor, and thereafter, the effect of increasing SNR is negligible. 2) When SNR is small, the maximum latency decreases as the offload ratio increases, but it requires a much larger SNR to reach the floor point. 3) There is an inflection point which is the minimum overall latency for each given SNR. Overall, to ensure a given latency, offloading more computations up to the cloud center requires less SNR, and thus consumes less energy. These findings suggest that we can design and reach different maximum affordable number of video streams for edge node in a given situation to make the full use of resources. It is possible to improve the system performance in terms of energy saving and efficiency improving by balancing the workload between MEC and the cloud. For future work,

Developing an Efficient Remotely-Operated Vehicle to Address Current Marine Issues

Mia Ladolcetta, Suhani Balachandran and Nishtha Dandriyal (Rogue Robotics, USA); Nivedha Srinivasan (nonE, USA)

Rogue Robotics is a nonprofit marine technology development program that gives its members an opportunity to explore and spread knowledge about STEM. The objective of the team is to develop the underwater vehicles that will address the challenging marine issues that society faces today. Every year, the team designs, builds, and tests a remotely-operated vehicle (ROV) to accomplish the underwater missions set forth by the MATE ROV competition.

The missions for the 2019 competition tasked ROVs with inspecting dams, maintaining river water quality, and preserving historical artifacts. Using the competition prompt and video of the mission flythrough as references, the team brainstormed and researched together, then produced 3D computer models of several possible designs. For the frame itself, several materials were researched, and finally, HDPE was chosen because of its high strength, insolubility in water, and high buoyancy. Buoyancy, thruster configuration, and electronics organization were some key elements to consider while designing the setup of the ROV beyond the basic frame. Initial thruster configuration and buoyancy were determined by calculations and then modified with detailed field testing. In addition, different components had to be designed to accomplish the various tasks, such as end effectors to lift objects underwater and a storage bin to carry mission-specific props, which were all first modeled using CAD, printed, and then modified after tests.

Equipped with six thrusters, two cameras, and mission-specific devices such as hooks, a metal detection system, and a grout and trout storage box, the ROV was able to accomplish many of the MATE competition tasks. The team scored 105 out of 270 possible points to score, a 10 point increase from the previous year, and completed 10 tasks in the 15 minutes given during the product demonstration. The team was especially proud of the compact design of the ROV and the organization board incorporated into the electronics enclosure, which both helped them earn the maximum score for size and weight constraints. Upon reflection, the team determined some future improvements to be made for the next MATE competition season, like creating a more modular frame and interchangeable mission-specific parts.

As of January 2020, the Rogue Robotics ROV for the 2020 competition is still in the first stages of design. Although not all components have to be redesigned, as budgeting and thoughtful design to enable reusability have always been core parts of the challenge for the team, new mission-specific parts of the ROV are currently in the works.

From July 2019 onward, Rogue Robotics has changed into an all-girls team, designed to promote girls working in STEM fields. Advocacy has always been an integral part of the team's missions, and the team is proud to be able to reach out to young girls more effectively and encourage their participation in science. The team members are all girls of color and feel that underrepresentation in STEM fields is a major issue, and it is their passion to design innovative solutions to serve the marine world while inspiring the next generation of scientists.

Relationship between Musical Scale, Cello String Length, and Math

Neo Cheng (Clakrsville Elementary School, USA)

I play the cello but my intonation is bad because I do not know where to put my fingers. Usually I use a tuner to help me decide where to press the strings, and then I put stickers to mark the locations. In this project, I want to know how to determine the sticker locations using only math without a tuner. First, I used my tuner to determine where to press the string for C, C#, D, D#, E, F, F#, G, G#, A, A#, B, and C. Next, I measured the pressed string length with a tape ruler. Then I calculated the pressed string length ratio to the whole string. For math, I knew there are 12 half-notes within an octave and the length is halved (50%) for each octave. I just needed to find a multiplier that divides the length between 100% and 50% equally. In other words, I needed to find a number M such that M*M*M*M*M*M*M*M*M*M*M*M=0.5. By using a scientific calculator, I found the magic number, M, to be 0.944! This means that for each half-note, the string needs to be reduced to 94.4%, and for each whole-note, the string needs to be reduced to 89.1%.

Object Recognition using TensorFlow

Nahuel E Albayrak (Chesapeake Science Point High School, USA)

Computers can apply vision technologies using cameras and artificial intelligence software to achieve image recognition and identify objects, places, and people. The objective of this project is to capture the image of an automobile as it drives by, identify its model and color, and determine its location, travel direction, and speed. This system can be used to assist law enforcement with vehicle identification in an emergency such as an Amber alert or to detect traffic infractions.
For this purpose, we constructed, trained, and applied an object detection model using TensorFlow. First, an image capturing system was built using camera lenses (Raspberry Pi Camera V2-8) and Raspberry Pi (Raspberry Pi 4) small computers. Next, the computers were set up with a software application called TensorFlow. The system was trained to recognize an automobile's model and color by processing a variety of car images. Pictures of different cars were uploaded from Google images and resized highlighting the features of the vehicle. Finally, code was developed in Python to create a universal clock for each camera that recorded the detection time.
Five trials were conducted using 2 automobiles available for testing. The cars were recognized by the model with 87 percent certainty in each of the 5 trials. That information was recorded on a table together with the time of capture and the location of the camera. The information from the table was used to successfully identify a specific car's location and speed, with a few limitations. Because of budget restrictions only two cameras were built and two models were used for training. The information from the cameras was not transmitted in real time because wifi or LTE capability are not available at this time. An extension of this research will include multiple cameras, multiple models and real time data transmission.

Rapid and Novel Thickness Identification Methodology For Two-Dimensional MoS? and In?Se? Nanosheets Using Optical Microscopy

Darren Y Wu (Charter School of Wilmington, USA); Tingyi Gu (Columbia University, USA)

From the first isolation of graphene in 2004, it has been well documented that the physical, optical, electrical, and thermal properties of two-dimensional materials are highly correlated to their thickness. For instance, material groups such as the transition metal chalcogenides (TMDs) evolve into direct bandgap semiconductors in monolayer form, transitioning to indirect bandgap conductors when their thicknesses are more than two layers, allowing for expanded applications such as phototransistors and photovoltaic devices due to the tunable properties possessed. Therefore, the development of a novel and accurate thickness identification methodology is imperative for the continued study and potential commercialization of two-dimensional materials. In this investigation, an effective, straightforward, and reliable methodology for the thickness identification of MoS? and In?Se? nanosheets on 300nm Si/SiO? under optical microscopy from approximately single to decuple layer (1L-10L) has been produced. The optical contrast difference values of the atomically-thin nanostructures were collected throughout experimentation and arranged into a valuable standard reference index which was correlated to height number in nanometers. Using this method, the thickness of a substance could be simply and accurately determined without the use of complex instrumentation, experimental setup, and calculation, while also saving time and monetary costs. The method illustrated in the investigation will streamline the research, manufacturing, and application of all two-dimensional nanostructures and further facilitate the advancement of two-dimensional materials towards industrial commercialization.

Study of sample efficiency improvements for reinforcement learning algorithms

Tianyue Cao (Princeton International School of Mathematics and Science, USA)

Machine learning is the study of how programmed algorithms can learn useful knowledge from data automatically. As a sub-field of machine learning, reinforcement learning (RL) focuses on problems that require sequential decision making. In particular, it is about interacting with the environment and taking action according to the environment information sequentially to maximizing some rewards. Reinforcement learning attracts many interests due to its recent successes in robotics as well as playing video games, GO, and poker. However, the fundamental challenges in reinforcement learning still limit its applications to real-world, cost and risk sensitive applications. One major challenge is relatively low sample efficiency in most systems. Sample efficiency is a term used to describe how well the samples are used to train the model. Because of low sample efficiency, it requires a huge number of samples to reach a certain level of performance. In most algorithms of reinforcement learning, methods such as experience replay are used to increase the sample efficiency. In the experience replay, a certain number of samples are saved in a buffer and new data will replace the oldest data in the set. When training, data will be randomly selected from the buffer. However, this will generate the problem of distribution mismatch, as the data chosen this way may not match the current model. In my research, methods are designed so that the samples collected from the past can reflect the current model. That will allow the model to use the data more effectively and thus increase its training efficiency.

A Futuristic Kitchen Assistant - Powered by Artificial Intelligence and Robotics

Riya J. Roy (Ridge High School, USA)

My project is about building a prototype of a futuristic kitchen assistant that is powered by Artificial Intelligence and Robotics. Using Cozmo (an AI-powered robot made by Anki) and Calypso (a language developed by Professor David Touretzky at Carnegie Mellon University for programming intelligent robots), I have built a proof-of-concept futuristic kitchen assistant that shows how the food identification and serving process can be automated.

I accomplished this by learning Calypso's rule-based language and its five fundamental laws of computation. Using Calypso's various programming features such as perception, teleoperation, pursue and consume, conflict resolution, speech and hearing, landmark-based navigation, and path planning, I learned how to make Cozmo move around and do intelligent activities, which are demonstrated in my prototype.

I designed a model kitchen using a cardboard box. I used the wall templates that had special symbols called "ArUco markers" to help Cozmo recognize kitchen walls and door openings and plan his path accordingly. Once I had the physical model of the kitchen ready, I created a new Calypso program that simulated the model kitchen along with the walls, door openings, the Cozmo robot, and three cubes that represented three different types of food. The program enabled Cozmo to recognize my voice instructions to get a particular food, go to the kitchen through the door opening, pick up the cube that represented the correct food, bring it to the dining room, and then drop it on a plate in front of me.

I faced several challenges such as how to make Cozmo recognize my voice, identify the door openings correctly, and move around without hitting obstacles. Eventually, after a lot of testing and debugging, I was able to get the kitchen assistant working and was able to prove that using a robot programming language such as Calypso, a robot can be programmed to perform highly complicated tasks such as listening to voice commands from human beings, navigate from one room to another (i.e., from the dining room to the kitchen), pick up an object (i.e., food), and then navigate and bring the object to another room (i.e., from the kitchen back to the dining room).

In the future, I plan to add more intelligence to the kitchen assistant such as providing the ability for a person to select a dish/recipe on a smartphone app, making the kitchen assistant go to the kitchen, find the right ingredients, follow the instructions in the recipe, make the food, and then serve it to the person.

Session Chair

To Be Determined

Session Poster-2

Poster Session 2

4:30 PM — 6:20 PM EDT
Aug 1 Sat, 4:30 PM — 6:20 PM EDT

Development of stock correlation networks

Lixin Huang (Princeton International School of Mathematics and Science, USA)

How to depict the relationships between stocks has always been a focus for scholars. Knowing the relationship between two stocks means that we can adjust the investment plans based on the correlation between the stocks. We are able to lower the risk of the portfolio while maintaining its expected return if we know the correlations between the stocks in the portfolio, assuming that information carries on through time
In this study, we establish a method to depict the relationship between two stocks in a more generalized way, as to provide a new approach to find the relationship between two stocks other than correlation. The following four categories are taken into account: the correlation between the stocks, how close the stocks are in case of the category of the companies that issue the stocks, how frequently that the two stocks are mentioned together, and possible transaction in the business between the two companies that issue the stocks.
To determine the relationships between stocks, an algorithm is initiated to generate a score between 0 and 1 for all four categories described above. Typically, a higher score indicates a more significant relationship between the two stocks. The data of the stocks are imported from the wind database, including the price and category of the stocks. The business transactions between the companies have been taken from D&B Hoovers. Primary and secondary sources about the stocks will be considered as textual evidence.
On the basis of the algorithm, the following 4-step analyses have been conducted. First, the correlation between the two stocks is calculated using the covariance matrix from the DCC-GARCH model. We assume the score of the correlation section equals the correlation between the two stocks. Second, if the fields of the two companies that issued the stocks are closer, the score for this section will be higher. Third, the score for the business transaction between the companies is determined by the proportion of transactions between the two companies. Last but not least, the score for textual evidence will be calculated using the equation below.
Where s_t is the score for textual evidence and n is the number of articles that mentioned both stocks. The final score between the two stocks is calculated by the weighted average of the scores for the four categories. After the score between each pair of stocks in the market is determined, an app is developed to display the top ten correlated stocks with the user's search to facilitate and optimize their selections in the stock exchange market.
In the future, this research could be conducted in the following three aspects. To begin with, the weight for the score for each part can be adjusted with more stock examples in order to depict the relationship between two stocks more accurately. Furthermore, the graph can be more user-friendly to increase the engagement of the users. Last but not least, a more sophisticated, multi-level categorization can be developed to optimize the categorization of the stocks given.

Uncovering the Genetics behind Alzheimer's Disease and Sleep: A Co-expression and Evolutionary Analysis

Sihan Fei (Princeton International School of Mathematics and Science, USA); Zeqing Li (Princeton International School of Math and Science, USA)

Alzheimer's disease is a fatal form of dementia, and it caused over 120 thousand deaths in the U.S. alone in 2017. Recent studies indicate that sleep deprivation is both a cause and an effect of Alzheimer's disease. Our research investigates the correlation between sleep deprivation and Alzheimer's disease through two stages. In the first stage of the research, the genetic coexpression of Alzheimer's disease-related genes (A-genes) and sleep-related genes (S-genes) across different stages of human development is explored. A general correlation between the expression of these two sets of genes is confirmed and strongly correlated A-gene and S-gene pairs are located, including GATA1 & ALAS2, TF & MOG, etc. In the second stage of the research, the expressions of A-genes and S-genes across different species are compared. Genes with unusual expression patterns in humans compared to those in other primate species are identified, hinting at possible genetic pathways key to solving the mystery of Alzheimer's disease.

Developing a Respiration Sensor for Babies

Ingrid J. Cruz (Oxon Hill Middle School & Southern University, USA); Michelle Soriano, Sarah Christie, Jazlynne Pichinte and Peter Chura-Borda (Oxon Hill Middle School, USA)

We designed and built a wearable technology that quickly and accurately measures the respiration rate of an infant. Specifically, a low-cost sensing device that can accurately measure the respiratory rate of a baby. An accurate respiration measurement is critical because an elevated Respiratory Rate is a marker of serious respiratory illness and is the main indicator for childhood pneumonia which is the leading cause of death in children aged 0 to 5 years worldwide.

We started the project by researching existing sensors in the market. We first narrowed down the 2 main ways of measuring an infant's respiration rate. The first way is contact-based while the second way was non-contact based. We decided contact-based was the best option since non-contact ways were more difficult to try. From contact-based, we figured out that there were 4 ways of measuring the respiration rate. These 4 were the acoustic method, the Co2 method, the airflow method, and the chest and abdominal method. The acoustic method needed a microphone and we imagined it would be hard to get a microphone for a baby. The Co2 method was also expensive to afford. We couldn't find a device to go along the airflow method. The last option was chest and abdominal movement and we looked into it and it seemed like the best choice if we use a flex sensor with it. Coding took the most time on the project because we had to learn how to do it from scratch. Our code was based on previously published codes we found online that we combined together in order to make our respiration sensor work.

The final prototype has the following design features:
-We used an expandable waistband with velcro lock to make it adjustable and allow appropriate fit on different infants while at the same time ensuring comfort for the baby. We also made it double layered in order to hold the sensor in place and catch every movement of the chest.
-We also added a little cut in the waistband to allow the flex sensor to be removable, making the waistband washable
- We made a pink waistband and a blue waistband for if the parents wanted a certain color for their child.
-We also made the code have a special feature in which if the baby stops breathing then the code will assume the baby is 'not breathing' and a buzzer on the device will sound.
-To top all of that, we added an LCD screen to show the readings and the whole system connects to a battery making our design very portable.

Aerogel Composites: Historical and Novel Synthesis Methods and Applications

Michael Chen (Hillsborough High School, USA); Akindu Dasanayake (Hillsborough High School, USA); Rohan Deb ((Hillsborough High School); Varun Deb (Hillsborough High School); and Evan Zhao (Dunlap High School, IL)

Our poster seeks to provide a broad overview of the many wondrous properties of aerogel composites (which include hydrophobicity, low density, blue light scattering, and superb noise, heat, and electrical insulation), detail chronologically various methods of the material's synthesis (such as sol-gel polymerization and seed growth method), and offer insight into its vast array of present and future applications (in fields such as fashion, aerospace, and construction). Research on our project began in May 2019 and is expected to be completed in March 2020. During the span of our research, we analyzed a wide selection of scientific papers detailing the fabrication processes, chemical and physical properties, and history of aerogel. To supplement our research, we performed live experiments (with the assistance of our AP Chemistry teacher) on a small monolith sample we purchased online. In one test, we placed the aerogel on a metal ring fixed to a stand above a Bunsen burner, lighted the burner, and then placed a match on top of the aerogel (the match remained unlit, demonstrating the material's extraordinary heat-insulating capability). Though our poster seeks, in part, to exhibit our experience in the classroom in research and experimental design in the field of materials science, it most importantly seeks to instill within the viewer a curiosity and awe for the highly promising future of this novel material.

Case Study of Asymmetries in Polar Rain Aurora

Dennis M Herschbach (Reservoir HS, USA); Yongliang Zhang (Johns Hopkins University Applied Physics Laboratory, USA)

Electrons and ions from the solar wind can directly enter Earth's polar upper atmosphere on both closed and open magnetic field lines. This occurs via the magnetosphere, where these particles are energized. When they then hit the neutral atmosphere, they ionize/excite molecules and atoms. Excited neutrals subsequently emit photons when they return to their previous ground state, which can have different wavelengths and are often visible to the naked eye such as in aurora Australis or aurora Borealis. Because they originate from the solar wind, auroral observations can reveal some of the physical processes that occur in the space that surrounds the Earth.

A special kind of aurora, polar rain aurora (PRA), is a phenomenon caused by solar wind electrons that enter the polar atmosphere directly on open field lines. Precipitating electrons, which are not energized/accelerated by the magnetosphere, often have low energy flux and don't create visible aurora. However, satellite-based ultraviolet imagers have higher sensitivities and are able to detect lower energies. PRA events were obtained through a manual search of auroral images from the Global UltraViolet Imager (GUVI) on the Thermosphere Ionosphere Mesosphere Energetics and Dynamics (TIMED) satellite and the Special Sensor Ultraviolet Spectrographic Imager (SSUSI) on the Defense Meteorological Satellite Program (DMSP) satellites. While PRA often appears in symmetrical and homogenous shapes, we present multiple events that exhibit unique spatial variations and structures such as shifts, tilts, or gaps. These features are likely due to structures in the solar wind energetic electrons, the magnitude and orientation of the interplanetary magnetic field (IMF), magnetic field reconnection, magnetic variations on the high latitude magnetopause, and/or a combination of the four processes above. In order to fully understand PRA variations and structures, a comprehensive statistical study as well as global magnetosphere simulation is required.

Nanoservice Infrastructure Notation (NINo) and the ASPIRE Interns

Chancellor T. Pascale (Johns Hopkins University & JHU APL, USA); Maria Rice (W. T. Woodson High School, USA); Shiva Sharma (Hammond High School, USA)

NINo is a future DevOps / Data Science pipeline tool that is being developed by JHU APL and two ASPIRE interns. The goal of this capability is to expose function-level capabilities, via either a simple application or configuration file, for use in Docker[1], Serverless Architectures[2], or data science/analytic pipelines. The goal is similar to efforts such as Metaparticle[3] and Source-to-Image[4] that aim to lower the barrier to horizontal scaling of data processing and analysis capabilities.
In previous years ASPIRE interns have developed tools to ease the acceptance of DevOps principles in JHU APL. They have created a web application, Harmonia, that asked users a few simple questions and supplied the scaffolding for a software project with artifacts to support sound software engineering processes. The lack of user interest has driven us to a more focused objective. NINo will focus on easing deployment to cloud environments.
Ideally, any person could develop cloud-based data science services.
The team and its work has been organized in an asynchronous and agile manner. There have been three members working on three subsystems: configuration, framework/integration, and artifact generation. An incremental and prototype-driven approach has allowed for creation of increasingly more functional software as internship has proceeded.
Interns have been given extensive control over their development processes and have investigated the programming frameworks used.
While the initial stages have not resulted in a complete system, the interns have improved their programming skills and complete common coding challenges. The team is close to integration testing and initial demonstration.
As the academic year closes, team members will work on design improvement, refactoring, and generation of future feature requests from prospective users. One summer intern will focus on developing a user interface for configuring and observing results.

Development of a System of Cerium(IV) Oxide Nanoparticles In Maximizing its Antioxidant Ability

Ziqi Jiao (Princeton Int'l School of Maths & Science, USA)

My research project incorporates the usage of Cerium(IV) Oxide nanostructures (nanoceria) to neutralize highly oxidative free radicals from an aqueous environment. The ultimate goal of this project is to modify and functionalize the nanoceria to maximize this antioxidant effect. The project is divided into 3 main stages: first, the nanoceria would be synthesized via hydrothermal methods, altering the reaction conditions each time to achieve controlled size and morphology of the nanoparticles. Secondly, the nanoparticles would have to be functionalized with smaller, biocompatible molecules such as ethylene glycol, poly vinyl alcohol, etc. Then, the antioxidant ability of the nanoceria would be tested quantitatively by mixing it with a free-radical generator and a dye that would discolor once oxidized by the free radicals (Photometric Evaluation). Future implements of the nanoceria would include testing its antioxidant ability in a in vitro environment and evaluating its cytotoxicity.
Over the past 5 months, I have read countless online sources about the synthesis of Cerium(IV) Oxide nanoparticles and their biomedical applications. I gathered numerous methods of synthesis and conducted experiments for each of them. In total, I have acquired 4 samples of CeO2 nanorods and 2 samples of CeO2 nanospheres. These nanostructures have been sampled under a Transmission Electron Microscope (TEM) and produced representative images of their morphology and size. Furthermore, these nanoparticles have been confirmed of their CeO2 content using FT-IR spectroscopy. One sample of CeO2 nanorods has also been surface-modified using polyvinyl alcohol (PVA) and FT-IR results show that the polymer molecules are firmly attached to the nanoparticles and don't come off even after violent disturbing and washing using a high-power sonicator probe. These results show that the synthesis methods viable in out laboratory turns out successful and with good yield.
I have also discovered the complex method utilized by other researchers of coating the nanoceria on silicon dioxide nanoparticles, which should increase the number of active sites and make the nanoparticles more efficient. I have tried synthesizing a coating layer of SiO2 on ferromagnetic nanoparticles to make the system easier to maintain, and further results are still pending.
A number of photometric evaluations of the antioxidant ability of nanoceria have also been conducted. In brief, CeO2 nanoparticles are mixed in an environment containing a hydroxyl radical source and a dye vulnerable to radical oxidation. The results show that CeO2 nanoparticles have a significant effect in neutralizing part of the hydroxyl radicals generated, but the results are not as prominent as stated in previous research. Finer tunes to the experimental conditions have also taken place, and results are still pending.

Effect of Roundup on Planarian Locomotion

Neelofar F Tamboli (Princeton International School of Mathematics and Science, USA)

Over time, the use of Roundup has been plummeting because of increasing glyphosate-resistant weeds. The main ingredient of Roundup is a toxic chemical called Glyphosate. Glyphosate inhibits a step in the Shikimic acid pathway. It prevents the plants from making proteins needed for plant growth, which has allowed it to become an efficient weed-killer. The purpose of this study is to investigate the effect of a widely used controversial herbicide called Roundup on flatworms called Planaria. The specific Planaria used in this experiment is called Dugesia dorotocephala. Planaria have an extraordinary ability to regenerate after they've been cut into two new individuals. They also have eyespots that act as photoreceptors and they tend to move away from light. The rationale behind using Planaria is that it is a good bioindicator since it occurs in the freshwater and testing effects on it will let us know about the effects on the entire ecosystem. Additionally, we should be very worried about the amount of Glyphosate contamination in our food supply since it has been on the rise. The experiment was conducted on the locomotion of Planaria and was evaluated by counting the number of 1mm by 1mm grid lines passed in a total of 2 minutes. The concentrations of Roundup that locomotion and regeneration were evaluated at were 0mg/L(control), 7mg/L, and 15mg/L. Our results indicate that Roundup inhibits locomotion in Planaria. There was no evidence found for a dose-dependent effect of Roundup on Planaria. Further research will assess alternative herbicides and pesticides to Roundup to see if they have any adverse effects on Planaria. Another future direction is on the exploration of neoblasts cells which allow the Planaria to regenerate. Finally, the future plan is to find an easy way to test the amount of Roundup in the environment.

Catalytic Ability of Ag-coated Ferromagnetic Microspheres Functionalized by TiO2

Qiyang Zhou (PRISMS, USA)

TiO2 has been used to clean wastewater as a photocatalyst that catalyzes the decomposition of organic pollutants through the production of reactive oxygen species. TiO2 has previously been functionalized on Fe3O4 for improved recyclability of the nanoparticles, and Ag-coating is applied to enhance nanoparticle's catalytic ability by reducing the bandgap of the catalyst in other researches.
In my research, I performed an experiment and will present a way of synthesizing Fe3O4@AgNPs@TiO2 microspheres by functionalizing the ferromagnetic microspheres with silver nanoparticles before TiO2. A photocatalytic test on the decomposition of methyl blue will also be performed to determine the catalytic ability of the obtained microspheres.

Quality Control of Brand Name Aspirin drug and Generic Aspirin drug

Jiale Lu (Princeton International School of Mathematics and Science, USA)

Brand name drug is developed and produced by the brand name drug company through a
complicated process while generic drugs do not require complicated development and
testing process. The huge price difference between the generic drugs and brand name drugs makes it important to analyze the differences in properties between these two kind of drugs. This research examines the physically and functional differences between brand name drug and generic drugs. The research primarily focus on the aspirin tablets sold on the market. The physical properties, such as width and weight, are measured. The uniformity of dosage unit is compared through the second derivative UV measurement to the aspirin inside the tablet. The function of the enteric coating is testified with solutions with different pH value in a dissolution tester. The disintegration rate for the tablets are measured and the total time for dissolving is also recorded. The final goal of this research is to find the difference and suggest one possible method for people to select between generic drugs and brand name drugs as well as one possible improvement for generic drug companies. From the result of the measurement, when comparing the aspirin pills from Bayer AG with generic aspirin, the consistency of the amount of aspirin in the brand name drug is greater and the coating is more effective when exposing to the acidic environment. The same method can be further applied to other drugs. Further studies can be done to improve the accuracy of this testing method.

Abstract of Kelvin Water Dropper

Edward Rossi Banfe (Princeton High School & Engineering Projects In Community Service, USA)

My poster is a demonstration of the Kelvin Water Dropper, invented by William Thomson in 1867. The goal of my project is to educate all those at IEEE about the basics of electrostatic induction through the Water Dropper, like William Thomson did. The poster will include a diagram of how the inductors and receivers work, will explain how a spark is generated from the flow of water, will display a working model of a Kelvin Water Dropper, and a laminated copy of the original paper by William Thomson for a demonstration. The Kelvin Water Dropper works via electrostatic induction, in which opposite charges accumulate in each side of the Dropper until a spark is formed, connecting the two sides for a brief moment and completing a circuit created by the built up positive and negative charges. After the spark occurs, the reaction restarts and the positive and negative charges begin to accumulate again. The charges are collected through the two hanging cans (inductors) and are stored in the cans diagonal to them (receivers). The receivers will pull water with an opposite charge to them and deposit it in the can directly below, causing more charge to be built up as the reaction continues and eventually stops to create a spark. My expertise regarding this project is adequate for what I will be presenting, as it is the project I have been learning and perfecting for last six months of my time at Princeton University's Engineering Projects In Community Service Joseph Henry Division (EPICS), led by Professor Michael Littman. In EPICS we specialize in projects relating to the history of electrical engineering, electromagnetism, and physics. We conduct outreach programs at local schools, libraries, and community events in the Central New Jersey area. In wit ion to my participation in EPICS, I am a student at Princeton High School in Princeton, New Jersey, currently at grade 10. I also volunteer for the Julia Robinson Math Festival which in Central New Jersey.

Magnet Dynamo - Princeton University EPICs

Hugo Kim (Princeton University EPICS, USA)

My poster is about the dynamo, a form of electric generator created by Michael Faraday in 1831. The poster will include diagrams and schematics of how the dynamo generates electricity, an early design of the dynamo, and a laminated copy of the original research published by Faraday for historical context. The poster's purpose is to demonstrate how early dynamos generated power, which aids in understanding their historical importance and applications.

In short, the magnetic dynamo works by using rotating magnets to cut through the lines of flux created by magnetic wire (a "stator"). In other words, the dynamo generates electricity by rotating one magnet under the influence of a separate magnetic field. The moving magnetic field pushes electrons through the wire, which generates an electromotive force (a phenomenon described by Faraday's Law of Induction). As a result, electrons move through the wire, generating a current that can be used to power devices.

Dynamos were the first electric generators powerful enough for industrial use; although the first dynamos used permanent magnets as the stator, the first industrial dynamo used electromagnetic coils as their stator. Passing a current through a conductive coil creates a much stronger electromagnetic field than permanent magnets. This happens because the magnetic flux lines produced all pass through the coil's center and overlap to create a very strong field. This principle allowed the dynamos using electromagnetic coils to produce enough power to be industrially viable.

Today, most power stations have phased out larger dynamos for alternators. Alternators are also a type of electric generator, but unlike dynamos, alternators produce alternating current, which periodically switches direction. In contrast, dynamos only produce direct current (current that flows only in one direction). AC is now the way electricity is delivered to houses and businesses; this is because AC voltage is easier to control with a transformer, which makes the energy transfer more efficient. Nevertheless, the low-voltage DC current provided by dynamos is still often used in modern electronic devices, whose circuit boards only function with a unidirectional, constant current.

I've been working on this project for over two months at a program offered at Princeton University by Professor Littman, known as Engineering Projects In Community Service (EPICS). The purpose of EPICs is to merge community service and engineering; we do this by creating projects relating to electromagnetism and presenting them at local schools and libraries during community events. Outside of EPICS, I'm a junior at Montgomery High School in Montgomery, New Jersey.

Demonstrating Lorentz Transformation Using Computer Simulation

Saniya Nagali (Allentown High School, USA); Anisha Iyer (Princeton High School, USA); Vanisha Nagali (Allentown High School)

Lorentz transformations are at the heart of Special Relativity as they are the correct description of how motion looks from moving perspectives in our universe. Lorentz transformations were developed to align with experimental observations which proved that speed of light is a constant in all frames of reference including moving ones.
Spacetime diagrams - with distance as the horizontal axis and time as the vertical axis - are typically used to visualize how objects in relative motion perceive each other. To understand the perspective of the moving objective, we need to transform the spacetime diagram such that the relative velocity, represented as the angle between the curves of two objects in the spacetime diagram, stays the same. The easiest way to visualize are shear transformations where the "time" of the moving object is kept the same and the "distance" coordinate is moved to the right or left on the spacetime diagram. However, such shear transformations do not maintain the constant speed of light. Lorentz transformations were then derived to obtain spacetime transformations that maintained the constant speed of light.
For high school students studying physics Lorentz transformations can be non-intuitive and difficult to understand as they require the spacetime coordinate plane to slide, rotate and stretch in the correct proportions to maintain the constant speed of light. A simple visualization of different spacetime transformation approaches can be a helpful aid.
We have developed a computer simulation that explains different transformation approaches (shear, Galilean, and Lorentz). We first modeled the coordinate plane using the AutoDesk Inventor software to develop a physical apparatus that mimics Lorentz transformations could be built. We then used a Java programming language to simulate the mathematical and movement concepts.

Best Predictors for Major Food Allergy Reactions Abstract

Will R Morrison (Princeton High School, USA)

Best Predictors for Major Allergic Reactions Abstract

I have recently received a dataset with information about 680 Children's Hospital of Philadelphia food challenges and whether they had a severe reaction or a mild reaction. Food challenges are appointments where a child is given a food that he has been tested to have a small or nonexistent allergy for to see if they will react. If they don't, they are cleared of the allergy and can eat it outside of the controlled environment. In each of the 680 tests in this dataset the child reacted and data was recorded about the type of food, how much they ate, how they reacted to it, and how severe the reaction was based on a standardized scale.

The goal with this data is to find out which of the 20+ columns is the best predictor for whether someone will have a severe reaction. For example, does a history of asthma make someone more likely to have a severe reaction? Or does sneezing during the test mean that they will have a severe reaction.

So far, I have done a logistic regression with the data. Some findings have surfaced, but for the final project I would need to find what variables to drop and focus on analyzing the results. Below are some screenshots of the data and work that I have done with it:

The first few entries in the dataset

A heatmap of the variables to determine which need to be dropped

Understanding Platonic Solids: Turning a Polygon into a 3 Dimensional Object

Sowmya Natarajan (Whittle School and Studios, USA)

My teachers had a difficult time teaching me how to find the surface area of a 3-D object, especially when I was looking at a 2-dimensional diagram. My goal is to teach people the concept of area and volume of a platonic solid through the use of 3-D pull up nets. A platonic solid is a regular, convex polyhedron. It is constructed by congruent, regular, polygonal faces with the same number of faces meeting at each vertex. Five solids meet these criteria: a tetrahedron, cube, octahedron, dodecahedron, or icosahedron.

In 1994, mathematics educator Bob Vertes introduced E.B. Meenan to the idea of Pull-up polyhedron nets. These nets could be created using only a card and string and easily folded up into a beautiful, three-dimensional shape.

Learning about volume and area through the use of platonic solids facilitates understanding and therefore easier for a person to apply these concepts in life. Using Pull-up nets is helpful to students who are visual or hands-on learners. Platonic solids are the basis for engineering, architecture, and geometry. Pull-nets can be used in many areas of life. Pull-up nets can form the basic design element of multiple objects from tents and bowls to prosthetic limbs. I want to advance the use of pull-up nets for tent-design, and as the basis for prosthetic limb design.

One other interesting questions I will explore include:
1. Is there only one pull-up net for each Platonic solid. A good starting point to explore this question is to consider the eleven distinct nets of a cube. I will explore if each of these formations form a string based Pull-up net.

2. What about other nets for other shapes like a tetrahedron (triangular pyramid)?
3. What about other polyhedra, do they have pull-up nets?

My research based on the work of Bob Vertes, EB Meenan and BG Thomas makes understanding volume and surface area of a 3 dimensional object fun and easy.

[1] E.B. Meenan. "Be a Paper Mathemagician", from Motivate: Videoconferences for Schools [online]. [Accessed 15/01/2008.] Available from World Wide Web:
[2] B.G. Thomas. Form, Shape and Space: An Exhibition of Tilings and Polyhedra. The University of Leeds International Textiles Archive, UK. 10 October 2007 - 16 May 2008.
[3] P. D. Turney. "Unfolding the Tesseract", Journal of Recreational Mathematics 17, no.1, pp.1-16, 1984-85.
[4] B.G. Thomas and M.A. Hann. "Patterned Polyhedra: Tiling the Platonic Solids" in R. Sarhangi and J. Barrallo (eds.) Bridges Donostia: Mathematical Connections in Art, Music, and Science, pp.195-202, 2007.
[5] B.G. Thomas and M.A. Hann. Patterns in the Plane and Beyond: Symmetry in Two and Three Dimensions. Monograph no. 37 in the Ars Textrina series, The University of Leeds International Textiles Archive (ULITA). 2007.
[6] Pull-up Patterned Polyhedra: Platonic Solids for the Classroom E.B. Meenan* and B.G. Thomas School of Education* and School of Design University of Leeds Leeds, LS2 9JT

What Effects Do Ultra Violet Rays Have on Yeast Colony Growth

Roshan S Natarajan (Whittle School and Studios, USA)

UV light triggers thymine to form thymine dimers inducing cell death.

Though the sun provides heat and light, which are essential for life on Earth, ultraviolet (UV) rays in sunlight can cause damage to DNA.In this science fair project, I will experiment with a strain of yeast that is super-sensitive to UV light.The goal for this project is to find out what percent of yeast colony growth has been killed.
Bakers yeast, or saccharomyces cerevisiae, is a eukaryotic unicellular organism. Cerevisiae is used in many laboratories as a model organism because it has internal organs such as a nucleus and a mitochondria. Since cerevisiae's genes have been well-studied, researchers are able to separate genes of interest from others, called knockout genes. In this project, a knockout strain of yeast will be used. This modified yeast is designed to be DNA-repair deficient which means that this strain of yeast does not have the enzymes needed to repair damaged cells while regular yeast and human cells do. When UV light destroys DNA the light initiates a reaction with thymine creating a thymine dimer. If the thymine dimer does not repair properly there are two paths it can follow, become a cancer cell if the thymine dimers are not widespread, or die, if they are widespread. In this project, there are many thymine dimers that will be formed when the modified yeast is exposed to UV light causing the yeast to die.

There will be two dishes next to each other with grown modified yeast. One dish will have aluminum foil on the top and the other one will not have aluminum foil. Then both of them will be exposed to UV light. This is the equation that is used to find out what percent of the yeast colony has died:

100 × ( 1 - colonies on exposed plate/colonies on control plate) = % killed

Two more tests will be done on the effects of pure UV light and the effects of regular light with no UV rays on yeast cells. This will show that the light is not effecting the yeast but the UV rays are.

This project will demonstrate how DNA in yeast cells are damaged by UV light, causing yeast cells to die. Similarly, UV rays cause human cells to mutate by destroying DNA, which leads to skin cancer. Although modified yeast does not have the enzymes that unmodified baker's yeast and human cells have, it will still show how UV rays affect eukaryotic cells' DNA. A future application for this project would be using skin cells to see how they interact with UV rays and by doing this more research can be done on skin cancer. When I find out what percent of yeast died when exposed to UV lights I will compare it to the effects of skin cancer and see how the enzymes react differently to UV light and look at the difference between the modified yeast and the skin cell.

Homopolar Motor

Tvisha Bhanushali (Hillsborough Middle School, USA)

Good afternoon!

My name is Tvisha Faria and I am currently an 8th grader studying at Hillsborough Middle School. I am part of Professor Littman's Engineering Projects in Community Service program at Princeton University. Using the vast concepts of science, we duplicate original experiments and share them via our events which occur in local schools, libraries as well as community programs.

I have recreated a project known as the homopolar motor. My essential goal for this presentation is to teach individuals about the homopolar motor. I doubt many people know about the homopolar motor and I would like to inform people about their capabilities. or Faraday's motor uses electromagnetism to create a rotational movement around a battery. Michael Faraday, the creator of the motor, was an English scientist of the 19th century, known for studying electromagnetism and electrochemistry. The homopolar motor essentially uses Lorentz force, a combination of electric and magnetic force, to spin a copper coiled wire around a battery and three magnets. This invention led to the making of motors and the discovery of electromagnetism in 1821.

My poster board will include information about the history and the science behind homopolar motors. Students will also be able to learn about homopolar motors through innovation and interactive experiments. My group and I have done this project in the past and have presented it in front of other individuals.

My presentation will include mini-experiments using the homopolar motor concept. Some of these experiments may include the battery and copper wire rotation or the circular battery motion on aluminum foil. My poster will include research about homopolar motors: their origin, evolution, fun facts, and more. My future goals are to continue to teach all about homopolar motors to people. My team and I often go to different locations to expose people to the realms of engineering and science. Homopolar motors fall under this category, and I am exotic to teach individuals about how they work. Also, I have a younger sister at home who loves learning about circuits. In the future, I would like to show her this experiment, which could be incorporated in her life. Her elementary school hosts an annual science fair every year, which is when I can teach her this project in such a way that she can share it with her classmates. Essentially, my goal is to show people how homopolar motors work. I would like to show people how interesting science and engineering is.

Thank you,
Tvisha Faria
[email protected]

How STEM can help save lives in Tsunami prone areas

Joshua Tewolde (Grand Bland West Middle School, USA); Girma Tewolde (Kettering University, USA)

This poster is about how tsunami warning systems can be used in Tsunami prone regions. Tsunamis are devastating natural forces that are dominant in Southeast Asia, where there are a lot of developing countries that cannot recover quickly when a disaster strikes. With tsunamis come a lot of loss of life because tsunamis can be deadly. One of the best ways to prepare for a tsunami is to know in advance that a tsunami is coming. If you know that a tsunami is coming, you have time to leave the coastal area. Approximately 65% of Indonesians (about 171 million people) live within 50 miles of the coast. Moreover, coastal resources have been used for further economic growth in countries within that region. For example, these economic sectors account for 25% of the GDP and 20% of the workforce in Indonesia. Further income inequality in this region will drive more people away from their homes and towards the coastal areas where there is an influx of new jobs in manufacturing, fishing, and agriculture. This will mean that there will be more people in this area with high risk of tsunamis, intensifying the need for a reliable tsunami warning system that will keep the hundreds of millions of people in this situation safe. The tsunami warning system should be able to meet multiple criteria, including, but not limited to, keeping costs low, maximizing warning time, and minimizing false alarms to avoid loss of public trust in the system. This system will have multiple parts, including the warning system and all the components that make it work. This poster makes an excellent example that ties into STEM in many ways. For the science portion, it relies on concepts of geology, such as how tsunamis are created, ways to detect tsunamis, and their impacts. We need to tie in technology in multiple parts of the system. First of all, we need to determine how to warn the public in the event of a tsunami, whether it be through mobile alerts/social media or possibly an alarm system throughout the region. We will also need to work out the technology needed to detect tsunamis, from the sensors to the control rooms. This includes learning the types and specialties of each sensor that could be used. Engineering will be a big part of this system, especially factoring into its design and deployment. We will need to use our knowledge of sensor types and tectonic plate boundaries in the area to make a prototype warning system. Mathematical equations and software simulation tools will be used to calculate the probability of a tsunami within a certain period of time and the effectiveness of the warning system. In conclusion, this poster demonstrates that knowledge in STEM is critical to help solve real world problems.

Defending Convolutional Neural Network-Based Object Detectors Against Adversarial Attacks

Victor Hu (Watchung Hills Regional High School, USA); Jeffrey Cheng (Bridgewater Raritan Regional High School, USA)

At the heart of many state-of-the-art image classification systems, including facial recognition systems and object detectors, is a convolutional neural network (CNN). A CNN uses thousands of pre-classified images to train a collection of numerical weights, which the network applies to each pixel of the input image in a series of layers to produce classification percentages. However, CNNs are by nature susceptible to adversarial examples. An adversarial example is an input image specifically generated to trick a CNN's collection of weights to incorrectly classify an image, although the image would look no different to a human observer. In safety-critical systems, such as autonomous vehicles, it is paramount that object detection is resistant to adversarial attacks. Autonomous vehicles rely on object detectors to identify things such as road signs and humans in their surroundings. We generated physically robust adversarial examples that successfully caused real-time state-of-the-art object detectors to misclassify road signs as other objects, a scenario where misclassification could result in damage and loss of life. In addition, we proposed defenses to mitigate misclassification. First, to prove that CNN-based object detectors are capable of reliably classifying stop signs under standard conditions, we tested the YOLOv3 object detector with normal stop signs as well as stop signs with sticker graffiti. A Raspberry PI car with a front-facing camera was used to simulate a passing car, reproducing dynamic perspective and lighting conditions. The car successfully detected a normal stop sign in 100% of the video frames and a stop sign with graffiti in 89.02% of the video frames across three trials. We then tested YOLOv3 with our adversarial attack designed to increase nonexistent "person" detection rates, which lowered "stop sign" detection rates to 58.74% and increased faulty "person" misdetection rates to 66.90%. To counteract the effects of the adversarial attack, we implemented defenses such as color thresholding and classification based on Haar features. The color thresholding snapped certain pixels to their closest associated color, thus perturbing the adversarial attack and allowing the CNN to successfully ignore the adversarial attack. Classification based on Haar features was a different method of classification that searched for certain features of a target classification. It acted as a safety net to our CNN, as the adversarial attack targets the CNN, but does not sufficiently perturb the Haar features. Implementing defenses such as color thresholding and classification based on Haar features returned "stop sign" detection rates back up to over 99%. Our work shows that adversarial attacks are substantial threats to the safety of autonomous vehicles by tricking their object detection pipelines, but their effects can be mitigated by using a variety of defense methods.

Object Recognition Using TensorFlow

Nahuel E Albayrak (Chesapeake Science Point High School, USA)

Computers can apply vision technologies using cameras and artificial intelligence software to achieve image recognition and identify objects, places, and people. Deep learning algorithms set up basic parameters about the data and train the system to learn on its own by recognizing patterns using many layers of processing. Thanks to recent advances in small computers such as Raspberry Pi and Deep Learning algorithms, object detection applications have become much easier to develop.

The objective of this project is to leverage these new technologies to detect an automobile as it drives by and estimate its location, travel direction, and speed. The procedure involved building an image capturing and classification system utilizing camera lenses and Raspberry Pi's small computers and using OpenCV as the image processing tool. In order to identify car features and predict a particular car's model and color, we used a machine learning framework called TensorFlow, and Google's pre-trained image model Res-Net-152 that was built from ImageNet dataset and Stanford car models data [1], [2], [3]. Finally, code was developed in Python to create a universal clock for each camera that recorded the detection time. As an automobile drove by, the camera system captured its image, recognized its model and color, and recorded this information together with the time and location on a log file. The information from the log was used to successfully identify a specific car's location and approximate speed. We performed trials using 4 different car models and obtained high confidence levels above 80% for most models. The lowest confidence levels belong mostly to sedan categories where models tend to look similar. Although our system was successful, it had a few limitations. Budget restrictions limited the number of cameras built to two. In addition, the information captured by the cameras was not transmitted in real time because WiFi or LTE capability were not available at the time. With a larger budget this system can be extended to include multiple cameras and real time data transmission. There are many applications for this system, offering significant benefits; from assisting law enforcement with vehicle identification in an emergency such as an Amber alert or detecting traffic infractions, to automating parking and drive-through systems.


[1] Imagenet data set

[2] Stanfords Cars Dataset, (contains 16,185 images of 196 classes of cars)

[3] Foam Liu, "Car Recognition with Deep Learning" Open Source;

[4] OpenCV; Open computer vision;

Using properties of electromagnetism to construct speakers from paper cups

Anisha Iyer (Princeton High School, USA)

An electromagnet is a soft metal core that demonstrates magnetic properties after passing a current through a coil that surrounds. Modern speakers, which function as a result of interactions between electromagnets and permanent magnets, can be modeled using a paper cup, copper wire, neodymium magnets, a razor blade or sandpaper, an MP3 player, and some other related materials.

First, the copper wire must be made into a coil of uniform radius. This is easiest when using a tool like a test tube. After coiling the wire, the coil must be unwinded such that a few inches of free wire are available on either end of the coil. Next, the ends of the wire must be stripped, with a razor blade or sandpaper, until the copper-colored wire turns silver. Using the corresponding materials, the wire must be stripped to reveal silver-colored wire which can pass current from the MP3 player.

To replicate the interaction between the electromagnet and permanent magnet, an electromagnet must be induced and paired with a permanent magnet such as the neodymium magnet. Audio connector cables should run from the headphone jack of the MP3 player to the terminal conductor. Two alligator clips, each from separate cables, should then be attached to the conductor, preferably by soldering. The other ends of each cable should be clipped onto the stripped ends of the coiled copper wire.

These connections allow the electric current coming from the MP3 Player to pass back and forth through the coil. Passing a current through the copper coil creates an electromagnet by inducing a magnetic field, the direction of which depends on the direction of current flow. The copper coil should be secured to the bottom of the paper cup, preferably using a material such as double-sided tape or glue that does not interfere with the integrity of the coil. If this material dislodges some turns of copper wire from the copper coil, this will cause a muffled sound.

For simplicity, the neodymium magnets can be placed on a cake pan. With the copper coil secured to the bottom of the paper cup, the paper cup can be placed onto the stacked neodymium magnets such that the coil hangs over the column of magnets. The magnetic field from the electromagnet will stretch over the neodymium magnets in a helical shape. Such arrangements allow the magnetic properties of a copper coil to interact with the north and south poles of permanent magnets, namely the ones attached to the cake pan.
As current passes back and forth through the coil, it is alternately attracted and repelled from the magnets on the cake pan. The resulting force pushes the bottom paper cup back and forth, which in turn pushes the air back and forth to produce sound waves. The cone shape of the paper cup concentrates and amplifies these sound waves.

Levitating a graphite rod using the camelback effect

Anisha Iyer (Princeton High School, USA)

This project aims to investigate the relatively recently discovered "camelback effect" using my knowledge magnetic properties. In a system of two lines of transverse dipoles, the "camelback" field confinement effect can be recreated in a parallel dipole line system (PDL).

The "camelback effect" occurs when two rows of magnetic dipoles are aligned to measure the strength of the field along the center axis. The magnetic field is stronger at the center and diminishes away from it. However, if the length of the dipole line exceeds critical length the field get stronger towards the edges of the dipoles and produces a confinement profile on the center axis that looks like a camel's back.
This camelback effect can be produced using special cylindrical magnets with poles on the curved side. The effect can also effectively trap an object at the center of the axis along the positive y-axis. A graphite rod can work as the trapped object and will levitate perpetually without any input power as a result of the camelback effect.

The graphite rod can also be levitated using a checkerboard of magnets, alternating according to their North and South poles.

Session Chair

To Be Determined

Made with in Toronto · Privacy Policy · © 2022 Duetone Corp.