Computer Science & Engineering Projects
 

Concrete is one of the most common building materials of the modern age, but its production process releases copious amounts of carbon emissions into the atmosphere. A solution to this problem are SCMs (Supplementary Cementitious Materials): Materials that can substitute cement in concrete. Less cement equals lower emissions. We will discuss some characteristics of common industry SCMs, such as how much cement they can replace and how they may change the properties of the resulting concrete.

Dark matter, which makes up the majority of the universe, can only be observed through its effect on other particles indirectly. To that end, researchers recently found deviations in the expected quantities of electrons and positrons (the antimatter counterpart to electrons) produced when beryllium-8 nuclei decays from a high energy to low energy state. This deviation could be explained if the nucleus emitted a boson with a mass of 17 MeV that split up into an electron and positron later on. This particle change can be detected by atomic clocks; because atomic clocks can sense deviations from the natural oscillation of an atom, they are hyperprecise measurement tools. As such, they are optimal for detecting particle anomalies like the boson described above. Last year's project succeeded in implementing GRASPy, a python module that works to simplify atomic calculations. Furthermore, using electron density outputs, the previous group also calculated the particle shift of Hydrogen 1s to 2s analytically, and graphed these results. Finally, the previous group ran wavefunction calculations for radium with one excitation to ascertain the candidacy of radium for dark matter. The objective of our project is to build upon last year's progress, particularly with regards to including MPI (Message Passing Interface) so we can run calculations with two excitations, and improving the multireferences for radium. We created a loop that optimized the selection of multireferences for radium with one and two excitations. To enable this calculation, we implemented MPI, which allows for parallel processing and will speed up more complex processes. To verify the functionality of the MPI wrapper, we are using Example 4 from the GRASP manual to ensure that there are no discrepancies between calculations in GRASP and GRASPy. Finally, we are creating an integration function to calculate the particle shift of radium.

Colorectal Cancer (CRC) is a malignant tumor type that is found in the colon and rectum. CRC remains to be the second most common cause of cancer-related death and the third most commonly diagnosed cancer among both men and women in the United States (“Colorectal Cancer - Statistics”). To further study CRC, we used single cell RNA sequencing data obtained from a research group at University of Texas, San Antonio(Chun-Lin Lin et al., 2020). We hope to further characterize the role of dasatinib treatment on colorectal cancer cells. Several pipelines were, and will be, conducted. Those pipelines include: clustering DNA expression data and finding which variants are common or different between the treated and untreated cancer cells by using raw sequencing reads from the dataset. So far, our analysis has been conducted using various R packages, especially the DESeq2 package, to normalize the data and calculate important statistical metrics (log fold change, p-value, etc.). This helped us find genes that have the lowest p-adjusted values, meaning that they had the greatest change between the treatment and control (statistically significant). Ultimately, choosing to analyze a cancer cell dataset using bioinformatic methods helps us discover novel genetic relationships in the context of CRC, which can aid in drug discovery.

EEG or electroencephalogram is an activity used to record nerve signals coming from the brain. These signals have the potential to predict behavioral disorders and provide essential information about how the brain functions. However, EEG readings are often filled with unnecessary information such as muscle movement and signals from the environment. In an effort to streamline the process of filtering and extracting information from the readings, we hope to employ the use of machine learning. Our presentation will show the progress we have made in this process as well as discuss our future plan for further research.

Autism Spectrum Disorder is a brain developmental condition that controls lifelong communication, personal behavior, and social interaction. ASD symptoms are diverse with a variety of levels in severity. Disorders, like attention deficiency/hyperactivity, intellectual disability, language delay, and genetic syndromes may co-occur with autism. Electroencephalography is an invasive and non-invasive electrophysiological monitoring technique. It records the electrical activities of the brain over the scalp using electrodes. EEG is a crucial tool for the diagnosis of diseases and disorders affecting the brain. It helps us to realize a much better understanding of the brain’s activities and structures. A recording of EEG signal is weak, nonlinear, non-stationary and contains various noise and artifacts. Our objectives are to study human brain anatomy pertaining to autism spectrum disorder; to analyze the variation of EEG Signals in context with autism; to examine the development of algorithms to detect ASD; comparative analysis of algorithms based on performance parameters; and to validate the results obtained from implemented algorithms and suggest a suitable model for detection of autism as an assistive technology to doctors.

Autism is a developmental and neurological disorder that can be hard to spot or find sometimes. Even though it is typically caught in the early years of one's life, autism is a disorder that can be diagnosed at any point in an individual's life. Since signs can show in a spectrum of different possibilities, having a discrete and unambiguous analysis of symptoms can prove to be extremely beneficial towards accurate diagnoses of autism. By analyzing the different waves that a person produces from their brain, machine learning can help make a model to detect autism in a patient through the use of the 10-20 system to isolate channels to extract data and pandas dataframes to help construct a heatmap to visualize the data. Coded in Python.

Although information about Mars is limited, NASA and individual researchers have gathered data through the Ingenuity Helicopter, reflecting a growing interest in aerial exploration of the Martian environment. Unmanned Aerial Vehicles (UAV) are developed to maneuver through the rocky terrain and harsh weather conditions on Mars. This study sought to use previous research to form the parameters for an efficient UAV that would carry out the scientific objectives of investigating surface composition and potential microbial life. Given the different UAV rotor blade and wing designs, such as the Fixed Wing Profile, the Multi-rotor blade, the Single-Rotor blade, and the Hybrid VTOL structures, the capabilities of the designs in a Martian environment were compared based on past UAV data. The system, the quad-copter multi-rotor design, was selected due to its efficiency, accessibility, and confined space operation. Due to the aerodynamics of the low Reynolds number Martian atmosphere, a thin, cambered airfoil with a sharp leading edge presents an effective structure for the rotor blades. We also established electrical parameters using constraints such as payload and efficacy in quantitative analysis of topography. For the powering systems we are using a rechargeable lithium-ion battery which would be charged at 30 minute intervals using a solar powered charging station. The propulsion system will be powered by four individual DC brushless motors which would be working simultaneously at ~5000 rpm to provide the adequate amount of thrust through the harsh atmosphere of Mars.

Automation has revolutionized modern society due to the increased efficiency and productivity it brings to otherwise time-consuming and labor-intensive processes. Used within a plethora of industries, such as transportation, finance, ecommerce, etc., automation ensures that tasks are completed with increased precision, speed, and efficiency. Even with this shift towards automation, manual synthesis is still the most widespread approach to creating compounds even though it is resource exhaustive and time consuming. Attempts to automate chemical synthesis have been made, such as Dr. Leroy Cronin’s “Chemputer”, and taking inspiration from such models, we propose a solution to these issues: RoboChem. In this paper, we present a novel adaptation of a robotic chemical synthesis platform which focuses on cost minimization while maximizing performance and reproducibility. Currently in the development and testing stage, the main components of the RoboChem consist of a robot arm to carry out a TLC for the analysis of the compound; liquid pumps, flowmeters, reservoirs, and tubing to facilitate and monitor transport of chemical content; reservoirs to hold reagents; and an elevation unit to allow for ice baths.

The elucidation of the structural function of protein sequences has been a long-standing challenge in information-based structural biology. We seek to develop a heuristic using machine learning concepts to identify a protein in a completely automated process using a K-nearest neighbor classifier, which works based on the assumption that similar points can be found near one another. We utilize the information of the defining features of a protein to map out the orientations and translations within DNA Ligase that define its structure as a means for identification. Compared to previous methods, we suggest a novel geometric approach toward protein classification and structural prediction. With the rise of computational power and increasing costs in physical chemistry, it provides an area for further investigation into easier means of protein classification.

Natural and synthetic products play a large role in the pharmaceutical industry with significant contributions to pharmacotherapy and chemical synthesis. Identifying and isolating characteristics within synthetic and natural products can allow for the development of distinguishable properties between these two types of products. Current methods such as binary QSAR (Quantitative Structure-Activity Relationship) modeling exist to classify compounds as either synthetic or natural products; however, there have been few advances in this area that use machine learning. Here, we propose a novel approach that uses logistic regression and decision trees with more than 1,800 features computed through the PaDEL-descriptor software, to develop a scale to predict whether a compound is a natural or synthetic product. By automating the process, the cost and difficulty with classifying compounds are greatly reduced, in turn resulting in fewer resources needed, and greater time efficiency.

For decades, researchers have searched the night sky for exoplanets. The main criteria for an exoplanet’s habitability is its ability to host liquid water, although many different characteristics must combine to support life. We analyzed the exoplanets in the NASA Exoplanet Archive and determined their CHZ (circumstellar habitable zone) with the bolometric luminosity. We then checked whether or not each exoplanet's orbit lied within the CHZ, thus making them habitable. We believed that we would find at least one habitable exoplanet, and all of the exoplanets found to be habitable would have planetary eccentricities of 0.4 and stellar surface temperatures under 7500 K. We reasoned that a low eccentricity was necessary to ensure that an exoplanet did not stray from the CHZ of its host star, and the host star should also have a low surface temperature so that its exoplanet is not too hot for liquid water. Based on our algorithm, we deemed 60 exoplanets to be habitable, but some of the habitable exoplanets had eccentricities greater than 0.4 or host stars that had temperatures greater than 6000 K, thus disproving our hypothesis.

For decades, many farmlands have had their productivity hampered by environmental pollutants from surrounding industrial infrastructure. However, the development of several physical sensors has allowed farmers to monitor the soil's health and prevent such loss. Here, we focus on developing a framework for collecting and analyzing soil data. Humidity, temperature, and nitrogen/phosphorus/potassium (NPK) sensors will be integrated onto a small form factor compute platform (e.g.: Arduino or Raspberry Pi). A client-server architecture will be built as the repository for a predictive, mathematical model to enable future data analysis. As environmental chemical changes alter the productivity of the soil, data collected by the proposed device will allow fluctuations [which impact soil health] to be identified and used as predictors for remediatiative treatment.

The Voynich Manuscript is a 15th-century text written in an unknown character alphabet, encrypted from an unknown root language; due to this, it has puzzled scientists for many years. By comparing letter frequency distributions of the Manuscript’s script, also known as Voynichese, and various other languages through a statistical Kolmogorov–Smirnov test, we found that out of the languages we tested, Galician Portuguese had the highest similarity to Voynichese. We also analyzed the first word of each page in the Plant section of the manuscript and found that more than half of these first words only appear once, within their respective folios. This could potentially lead to the conclusion that these first words are proper nouns, specifically the names of the plants featured in the manuscript.

Analysis of Stellar Attributes to Determine the Feasibility of Carbon-Based Life in Exoplanets

Having evaded intense efforts toward its translation for over 500 years, the Voynich Manuscript, a text dating back to the 15th century that is written in an unknown character set, remains a compelling mystery for researchers in the areas of linguistics, history, and cryptography. Within this study, the topic of inquiry relates to the Voynichese characters being a compilation from various cultures and their ancient languages, including an extensive basis in Vulgar Latin with additions from Middle Persian and Sanskrit. It is hypothesized that the Voynichese character set is an encoding based upon a source language of Vulgar Latin origin. When the author was unable to identify characters to match the phonemes found in Vulgar Latin-based languages, they likely borrowed characters that matched those particular sounds from other languages. Research methods performed to validate this hypothesis include character mapping, correlation with images within the text, and historical tracing of the origins of the manuscript. With regard to historical tracing, the Voynich manuscript’s constituents bear many connections back to ancient Persia. Most, if not all of the inks and paints found in the manuscript were widely used in Persian manuscript painting. The ink was iron gall ink with traces of copper and zinc which were unusual for the time. It was inferred that those traces were from a brass inkwell, which had widespread use from the 12th-16th centuries in Persia. The constituents of the colors used in the manuscript were Azurite (for blue), Copper Chlorine resinate (green), Red Ochre (red), and Calcium Carbonate (white). Due to the timing and how all of the constituents of the manuscript originated in Persia, it is safe to infer that the manuscript had a chance of being influenced by Persian culture or written in the geographic region of ancient Persia. With regard to character mapping, the widely-accepted Extensive Voynich Alphabet was used to draw relevant connections between specific words in the manuscript and words utilized in Middle Persian and Sanskrit to describe concepts or images present in the diagrams in close proximity to the words in question. In addition, the history and origins of the Glagolitic script were studied in order to better understand how character sets may be created toward a specific purpose or created due to the rapid and concentrated exchange of cultural information. Posited theories toward explaining the origin of the manuscript include that it was created as a collection of medicinal, herbal, and astrological knowledge collected on the Silk Road, where European, Persian, Hindi, and Buddhist cultures were known to interact. As the scientific community continues on its journey toward revealing the secrets contained within the elusive Voynich Manuscript, further research avenues include investigating the frequencies of the decoded words presented in this study with their relation to depicted diagrams in the manuscript.

The search for habitable exoplanets is not a fruitless effort considering that there are thousands of exoplanets in the universe, and at least one must possess the characteristics needed to support basic life. Water, being the basis for life, leads to the main question regarding the habitability of an exoplanet: can the exoplanet support liquid, or molecular, water? Whether or not this is possible is dependent on a multitude of characteristics, including planetary radius, orbital period, stellar radius, and planetary mass, working together to create an environment capable of life. The purpose of our research was to unravel this “mystery” by looking at data values from the NASA Exoplanet Archive (Planetary Systems Dataset). We started with a deduplication Python program to weed out duplicate entries in the NASA Exoplanet Archives, and then analyzed multiple planetary and stellar data values, through Keplerian Mechanics and on their own to filter out a list for us to take a deeper dive into. Subsequently, we conducted statistical calculations for further analysis. Further efforts can be made by applying Albedo and Circumstellar Habitable Zone (CHZ) data to the analysis of the habitability of an exoplanet. In the end our group formulated a list of ten exoplanets that we believe to be capable of supporting life.

For decades, many farmlands have had their productivity hampered by environmental pollutants from surrounding industrial infrastructure. However, the development of several physical sensors has allowed farmers to monitor the soil's health and prevent such loss. Here, we focus on developing a framework for collecting and analyzing soil data. Moisture and nitrogen/phosphorus/potassium (NPK) sensors will be integrated into an Arduino Uno. A client-server architecture will be built as the repository for a predictive, mathematical model to enable future data analysis. As environmental chemical changes alter the productivity of the soil, data collected by the proposed device will allow fluctuations [which impact soil health] to be identified and used as predictors for remediatiative treatment.

In April of 2021, Jayasinghe et al. reported the discovery of a small, black hole binary companion to the red giant V723 Mon. The discovery of a black hole this small is very important as it closes the mass gap between the smallest known black holes and the smallest theorized. The black hole’s mass, distorting the surrounding spacetime fabric, affects its binary companion by stretching it’s stellar envelope causing the star to take the shape of an ellipsoid. Under normal circumstances we expect stellar radius errors 1 and 2 (measures of the accuracy of the stellar radius [st_rad]), to asymptotically approach zero. However, stars with a mini-black hole companion would have stellar radius error increase as the stretching of the star into an ellipsoidal shape makes it harder to ascertain the stellar radius. Using the data found in the NASA Exoplanet Archive, we used an algorithm to initially look through the M and K type stars for any stars that meet our criteria. While we haven’t found any definitive candidates yet, we assume that with more time series data, and by further refining our algorithm, we will have a solid methodology to identify possible mini-Black Hole binary pair candidates.

The search for habitable exoplanets has been going on for over a decade and it is unreal to think that not even one of the several thousand exoplanets known can support basic life. Whether or not a planet can support life or liquid water requires that a variety of factors, such as planetary and stellar characteristics, work in harmony with each other. The research we performed focuses on filtering the thousands of known exoplanets and their host stars through various methods to search for traces of life. We start by looking at the Circumstellar Habitable Zone (CHZ) of each host star, which is a region of a stellar system in which water could potentially exist. Our group used data from the NASA Exoplanet Archive (Planetary Systems Dataset) and created Python codes that would calculate the CHZ for each stellar type. In the dataset, there were multiple entries for each exoplanet, indicating different sources of information, thus, we programmed a code that provides us with the most recent and accurate data, while avoiding errors, such as false-positives. Apart from that, two main methods were used in our research: Keplerian Mechanics and Albedo. In the end, our group theorized 10 exoplanets that are capable of supporting life.

Universal Robots is a startup that creates robotic arms with a wide range of applications. Research conducted has focused on combining the automatic capabilities of these robots with artificial intelligence and computer vision to enable these robots to dynamically adapt to differing conditions and load demands. Preliminary research and experimentation with the robotic arm were conducted to gain knowledge of the robot’s capabilities. The robot was used to write characters on a whiteboard, where the goal was to guide the robot to a dry erase marker where it would write text given by the user. Initially, the issues of greatest concern were a lack of knowledge of the robot’s functionality and the programming interface. As these became more familiar, the challenges were related to the robot environment, such as placement of the whiteboard and eraser in order to reduce collisions. The culmination of this preliminary research has successfully resulted in a program that can write any five-letter word on the whiteboard. With preliminary research complete, research will now move towards integrating computer vision and creating a tactile feedback loop to identify objects and handle them according to their structural integrity.

For decades, many farmlands have had their productivity hampered by environmental pollutants from surrounding industrial infrastructure. However, the development of several physical sensors has allowed farmers to monitor the soil's health and prevent such loss. Here, we focus on developing a framework for collecting and analyzing soil data. Humidity, temperature, and nitrogen/phosphorus/potassium (NPK) sensors will be integrated onto a small form factor compute platform (e.g.: Arduino or Raspberry Pi). A client-server architecture will be built as the repository for a predictive, mathematical model to enable future data analysis. As environmental chemical changes alter the productivity of the soil, data collected by the proposed device will allow fluctuations [which impact soil health] to be identified and used as predictors for remediatiative treatment.

With advancements in the field of machine learning powered by on-demand computing and information processing on a large scale, computationally driven proteomics and high throughput virtual screening have become increasingly popular in reducing traditional in vitro screening costs and the timeframe for hit-to-lead identification of drug candidates. The efficiency of high throughput fingerprinting using cheminformatics based approaches coupled with machine learning holds immense potential in screening possible inhibitors. To identify these potential targets, we propose a reductionist approach in identifying key pharmacophoric elements of chemical entities, dramatically reducing the relative compute cost for large scale chemical screening efforts. By minimizing the 3D structure of our molecules to their key points we are able to screen a larger sample of chemical space while effectively filtering for ideal small molecule drugs. Platforms such as PaDEL and Mordred were used in identifying notable descriptors of a class of FDA-approved NNRTIs, and this data was later implemented in a machine learning based model when screening for structural similarity between NNRTIs and other datasets of organic compounds. Herein, we present a novel pharmacophore fingerprinting method based on 3D reductions of molecule libraries, enabling a relatively more efficient and rapid screening of effective inhibitors of the HIV-1 RT enzyme.

There are thousands of exoplanets in the universe, made of many different sizes and characteristics. We already know one planet that supports life (planet Earth), so there must be another one somewhere else that also possesses the characteristics needed to support basic life. One crucial factor towards determining a planet's habitability is water: Since water is the basis of life, will we be able to find an exoplanet that supports liquid or molecular water? We tested data from the NASA Exoplanet Archive for characteristics including planetary radius, orbital period, semimajor axis, luminosity, stellar radius, and planetary mass to find planets that might sustain water. In the spring semester, we updated our deduplication file to organize the planetary observations from the raw dataset. Then, using the characteristics mentioned before, we examined the dataset using mathematical techniques and Keplerian Mechanics to infer which planets may be habitable. In addition, we delved into research involving the albedo and circumstellar habitable zone (CHZ) methods to validate our results in Keplerian Mechanics. With these methods, we added multiple new exoplanets to our list of planets that could support life.

Manual synthesis is the most widespread approach to evaluating drug candidates, but it has proven to be costly and inefficient. Using artificial intelligence and mechanical engineering, we propose an efficient approach for the automated synthesis of novel chemical compounds. This approach utilizes a “monkey-bar” frame to support chemical reservoirs flowing to reaction chambers through tubing and pumps. A robot arm will affect the addition of liquids and disposal of the reactants within the reaction chamber when added in excess. The automated synthesis platform, called “RoboChem”, will add the reactants in small, controlled amounts and then use either a spectrometer or high performance liquid chromatograph to verify the completion of the reaction. The AI-driven platform will connect to a drug discovery algorithm to receive candidates to synthesize and identify novel chemical compounds for further testing.

Machine Learning has plenty of real-world applications ranging from modeling the universe to computational chemistry. Because probability is the bedrock of machine learning, it is essential to optimize both hardware and software to obtain the best results. Classical computers are generally used for machine learning programs; however, learning from high-dimensional data often requires excessive compute time and may not achieve the highest accuracy. The Quantum Computing environment can be used to create a more accurate model than one created by Classical Computing. To test this quantum advantage, we implemented a Quantum Convolutional Neural Network (QCNN), which parallels the classical Convolution Neural Network (CNN) structure in the quantum domain. Due to the lack of quantum computers with many qubits, engineers have introduced the Noisy Intermediate Scale Quantum (NISQ) concept, which constitutes a hybrid interface between classical and quantum computers. In the context of QCNN, the data processing and the cost function optimization would be performed on the classical computer, while the probabilities generated by the parametric quantum circuits are evaluated in the quantum computer. Thus, the QCNN consists of a classical-to-quantum data encoder, a cluster state quantum circuit to entangle qubit states, a series of Parameterized Quantum Circuits using Quantum Convolutional and Pooling Layers for efficient feature extraction, a quantum-to-classical data decoder, and a fully connected layer leading to the output. Both the CNN and QCNN extract features from data like 2D images, and the networks' performances can be compared using metrics such as accuracy, loss, and compute time. This investigation has shown that the QCNN can perform better than the CNN for specific applications.

Predicting fires is a crucial part of forest fire management. With these predictions, firefighters can more efficiently use their resources to protect the general population and the environment. In this research paper, we will use machine learning to analyze historical fire behaviors and accurately predict the spread of wildfires in California. This data includes fire location (latitude and longitude), the number of acres burned, and meteorological factors (wind speeds, humidity, temperature, and precipitation). Observing methods of prevention and their efficiency is incorporated into this aspect of the analysis. After first classifying wildfire size as A-F based on the number of acres burned, this research paper will use classification algorithms like logistic regression, support vector machine, and random forest classification to predict the size of wildfires using meteorological data and determine the most effective methods of preventing and fighting wildfires based on classification categories.

Machine learning has plenty of real-world applications ranging from modeling the universe to computational chemistry. Because probability is the bedrock of machine learning, it is essential to optimize both hardware and software to obtain the best results. Classical computers are generally used for machine learning programs; however, learning from high-dimensional data and many features often requires excessive computer resources and may not achieve the highest accuracy. The Quantum Computing environment can be used to create a more accurate model than that of Classical Computing. To test this quantum advantage, we implemented a Quantum Convolutional Neural Network (QCNN), which parallels the classical Convolution Neural Network structure in the quantum domain. Due to the lack of quantum computers with many qubits, engineers have introduced the Noisy Intermediate Scale Quantum (NISQ) concept, which constitutes a hybrid interface between classical and quantum computers. In the QCNN context, the data processing and the cost function optimization are performed on the classical computer, while the probabilities generated by the parametric quantum circuits are evaluated in the quantum computer. Overall, the QCNN consists of a classical-to-quantum data encoder, a cluster state quantum circuit, a series of Parameterized Quantum Circuits using Quantum Convolutional and Pooling Layers, a quantum-to-classical data decoder, and a fully connected layer leading to the output. Both the Convolutional Neural Network and the QCNN extract features from data like 2D images. The networks' performances can be compared using metrics such as accuracy, loss, and time. This project has shown that the QCNN can perform better than the CNN for specific applications.

A common view of global trade is that rich countries generously aid and develop poor countries through trade deals and loans. However, research suggests that the opposite is true--rich countries siphon resources and labor from poor countries through unequal exchange and economic restructuring in order to develop and aid themselves. Previous research on global trade and 21st century imperialism has been relatively inconclusive and incomprehensive, particularly on the causes of such exploitation. Through analysis of the monetary outcomes of global trade, our study aims to clarify and add to the discussion of the globalization of production and examine the role of rich countries in global trade. By using multiple regression techniques and analyzing previous literature, we found that GDP per capita has a positive correlation net transfer, suggesting that richer countries exploit poorer countries. We aim to further examine the role of economic planning in reducing the likelihood of exploitation and the implications of such exploitation in modern society.

There are many different large tech companies that are developing new technologies to try to advance society for the better. A recent example of this is with the advancement of facial recognition. Google Vision was the software tested in order to prove and possible correlations. Using their api, it has been determined that there is possible bias in the software. A dataset of mug shots with various different kinds of people was passed into the api and configured many separate times to return back data that would demonstrate if the software was indeed biased. The tests were able to determine a few different things. When calculating the percent of people not even classified as human, Asians were the highest by a wide margin. On top of this the api deemed that the confidence of their software that asians were being recognized was of the highest. This project on Google Vision proved a possible bias by using an outside database and passing that through the api.

Our project serves to develop a machine learning model that analyzes the features of audio clips from political speeches and addresses given by Trump and Biden during the 2020 presidential campaign season. Upon analysis, the model measures relation to a specific emotion from the following categories: neutral, calm, happy, sad, angry, fearful, disgust, and surprise. First, we use web-scraping to collect audio data of the 2020 presidential candidates’ speeches. These are then rescaled into sizes comparable to the training data used to develop our machine learning model to determine emotions identified in each. The result provides more insight into both Donald Trump’s and Joe Biden’s stances in their presidential campaign. Our project aims to provide a more calculated way to determine the stance of a given speech.

Earthquakes pose numerous threats to this world, significantly due to their unpredictability. These natural disasters are a very common and dangerous phenomenon, and being able to predict their location and magnitude can help save lives. Seismologists are cognizant of the various signals that precede an earthquake, including P-Waves, and EEW systems. However, these scientists have yet to find much concrete results in predicting earthquakes. As a result, our team aims to ultimately discover new earthquake patterns, and forecast the magnitude of these natural disasters in the near future. The following presentation will discuss a time series analysis model to predict seismological patterns, and various forms of regression to predict the magnitude of earthquakes.

Earthquakes pose numerous threats to this world, significantly due to their unpredictability. Seismologists are aware of the various signals that precede an earthquake, including P-Waves, seismic waves that are undetectable by humans, and EEW systems, which consist of sensors that could possibly inform local residents of an earthquake through the recognition of seismic shaking; however, these scientists are yet to have found much concrete results in predicting earthquakes that may occur a month or two in the future. As a result, our team aims to ultimately discover new earthquake patterns, and forecast the magnitude of these natural disasters in the near future. The following presentation will discuss a time series analysis model to predict seismological patterns, and various forms of regression to predict the magnitude of earthquakes.

Sudden Oak Death is a disease found in different species of trees that causes the leaves of infected trees to decay and eventually kills the trees. The decay is characterized by parts of the affected leaves turning darker in color, typically brown or black. The disease is caused by a water mold pathogen. California, having high tree population density compared to the rest of the United States, is especially subject to the consequences of Sudden Oak Death. In order to make handling the task of researching this disease easier, our group's aim is to create a machine learning model capable of differentiating between leaves with and without Sudden Oak Death. This presentation will go over the steps and process behind creating the model, including getting our dataset, preprocessing it, choosing a model, and training it.

Many approaches have been taken to predict stock market variability. Prior research has suggested a correlation between public sentiment and the Dow Jones Industrial Average Index (DJIA). This paper explores the relationship between Twitter sentiment and stock price fluctuations using machine learning and time series analysis. We analyze the sentiments of tweets concerning top NASDAQ companies from 2015 to 2020 and plot their correlations with S&P 500 stock prices over the same time period. We then use Granger’s Causality Analysis to verify whether twitter sentiments can be used to predict stock prices for various companies. Finally, we attempt to forecast specific companies’ stock prices from twitter sentiments using vector autoregression and evaluate the forecasts by comparing their Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE), finding certain stocks to be more predictable than others.

The ongoing COVID-19 pandemic, also known as the coronavirus pandemic, is a global pandemic caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Multiple vaccines have been developed, underwent clinical trials, and distributed along with many boosters and doses. While these vaccines are currently being distributed the virus still spreads as we have not reached herd immunity. As the virus spreads new variants like to enter. This study aims to map the vaccination candidate connectome using network analysis with the Barabási-Albert model. This network science will provide scale-free networks that accurately model the powerful distribution of SARS-CoV-2 vaccines on a global scale. From there, the study will use R & Java Script as well as Gephi to visually model these networks. Along with modeling, we will use machine learning to predict the next pandemic among vaccinated candidates on a global scale.

Neurodegenerative diseases often have underlying genetic causes that are difficult to pinpoint and analyze. Recently, however, topological algorithms have become popular for extracting information from large genomic datasets. Topological Data Analysis (TDA) is a field that utilizes persistent homology and computational geometry to provide insight into the underlying shape and structure of data. Applications to cancer research and disease analysis have demonstrated the value of using TDA to identify significant patterns in genetic markers. We aim to use TDA to analyze genomic data in an effort to better understand how neurodegenerative diseases manifest. Using TDA algorithms such as Kepler Mapper, we will analyze the structure of the specific areas of the genome that are associated with such diseases. Identifying indicators of neurodegenerative disease can revolutionize diagnosis and respective treatments.

While epitomizing the conjunction between information sciences, linguistics, and psychology, natural language processing (NLP) helps to collect data that can determine how computers can understand human language and engage in two-sided communication. From applications in healthcare, the military, and robotics, NLP has been a primary translation method for several centuries. In addition, NLP is often associated with sentiment analysis, which essentially allows for the close examination of feelings or moods based on textual evidence. Although NLP is used heavily in information and task management, we decided to combine its technical facets with those of neuro-linguistic programming (also NLP), the enhancement of physiological skills, and leverage our knowledge on the extraction of keywords to reconstruct shortened versions of conversations. Our vision for this project is to merge the gap between neuroscience and technology through creating and using a Raspberry Pi-based cognitive linguistic assistant device that records voice and sends shortened communication through email or a potential complementary mobile app. This device would benefit those with speech or memory impediments, including neuro-divergent students in the classroom, and those struggling with mental health, autism, ADHD, and Alzheimer's.

Functional Magnetic Resonance Imaging, known as fMRI, is a non-invasive neuroimaging technique. It utilizes BOLD signals to construct high resolution images of brain activity from subjects instructed to perform tasks or respond to stimuli within an MRI machine. Extracting meaning from fMRI data requires choosing an appropriate raw dataset and performing numerous preprocessing steps to reduce noise and artifacts. This summer, the dataset accompanying the paper “Dynamic intersubject neural synchronization reflects affective responses to sad music” (Sachs et al., 2020), publicly available at openneuro.org, was chosen for analysis. The dataset contains scans of brain activity of 40 participants listening to music asserted to evoke either sadness or happiness. Analyzing this dataset will provide insight into how stimulus-driven changes in activity and connectivity in the brain correlate to emotional enjoyment and intensity. This term, the dataset was preprocessed through an fMRIprep pipeline, as well as using ICA-Aroma. In the fall, analysis will begin with the aim to examine the regions of the brain most connected to musical and emotional experiences.

The COVID-19 pandemic has drastically altered typical living habits, including vital circadian cues, and is linked to new pressures, new responsibilities, and new concerns about one's health and financial security, all of which are likely to disrupt sleep. Our research group aims to study the effects of the COVID-19 vaccine on sleep quality during the pandemic by assessing vaccinated and unvaccinated participants. Our team designed an eligibility questionnaire based on our analysis of various populations (healthcare workers, children, night shift workers) to reduce the chances of outliers in our study. Here, we will further discuss the reasoning behind our eligibility questionnaires and how we’re currently creating a questionnaire to assess the sleep quality of our participants in which each question is backed by previous research on sleep and COVID-19.

The ongoing COVID-19 pandemic, also known as the coronavirus pandemic, is a global pandemic caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Multiple vaccines have been developed & underwent clinical trials to prove efficacy, such as the one produced by Pfizer & BioNTech. While these vaccines are currently being deployed, no vaccine is fully effective and thus, breakthrough infections, in which a fully vaccinated individual may contract the virus, occur, albeit infrequently. With the appearance of the Delta Variant, a more infectious & now most predominant strain of the SARS-CoV-2 variant that was first identified in India in December 2020, it is now being questioned whether vaccinated individuals who are subject to a breakthrough infection may contract the Delta Variant as well. This study aims to map the vaccination candidate connectome using network analysis with the Barabási-Albert model. This network science will provide scale-free networks that accurately model the powerful distribution of SARS-CoV-2 vaccines on a global scale. From there, the study will use R & Python, as well as Gephi to visually model these networks and predict the next pandemic among vaccinated candidates on a global scale.

With an estimated 50 million dementia cases worldwide, this neurodegenerative disease is becoming more prolific than ever. Methods of CBCT/CBCR (Computer-based Cognitive Training/Rehabilitation or Brain-Care) and music therapy have shown effectiveness as a means of positive intervention for geriatric groups of Alzheimer’s dementia patients, leading to mental state and quality of life improvements and decreases in patients’ Clinical Dementia Rating. Though these interventions have led to positive outcomes, their development and testing occurred irrespective of one another. However, a combination of these techniques would potentially produce an unprecedented level of amelioration. In this presentation, an overview of Jahanikia Neuro Lab’s efforts to achieve this result will be delineated, contextualized by current works in the field that prove the isolated efficacy of these methods.

This project aims to analyze the post-election sentiment of Joe Biden in contrast to his sentiment prior to the election. The opinions of Biden are presented through his words, and sentiment analysis is used determine his emotions and thoughts regarding key topics in political discussions. For example, when positive words such as “unison” or “we the people” are used in general, it leaves a lasting effect on the audience because the speaker is creating the impression that everyone is one, together, leading to a positive sentiment. In this project, our group will be analyzing speech transcripts of Joe Biden, to determine his polarity and subjectivity within the topics of the economy, covid/vaccine, social justice, and security, along with a neutral category for speeches that do not belong to any of the topics listed. We can then contrast Joe Biden’s post-election polarity/subjectivity with his pre-election polarity/subjectivity, which were recorded from the project prior to the current one, using plots and figures. The polarity score is defined as the positivity/negativity of a speech, on a scale of -1 (most negative) to 1 (most positive). The subjectivity score represents how subjective a speech is, on a scale of 0 (objective) to 1 (fully subjective). By using transcripts of Biden’s pre- and post- election speeches, we identify his sentiment based on polarity and subjectivity. His speech is then categorized into lists to determine whether the he is more positive/negative and objective/subjective in regards to the above topics.

Through this research, we aim to analyze political speech audio from Trump and Biden using machine learning. The overall goal of this research is to analyze Donald Trump’s and Joe Biden’s political speeches and provide their sentiment scores on controversial topics through computational methods. We will compile several of the presidential candidates’ speeches and addresses, then feed it into our machine learning model to give each one an emotional classification as surprised, fearful, angry, sad, happy, calm, or neutral with a provided measure of accuracy. The result will be a more precise idea of both Donald Trump’s and Joe Biden’s stances in different conversational methods.

This visual emotion classifier utilizes machine learning to output a measurement of the emotions being exhibited on camera. The program aims to access a live camera feed and trained off of a data set consisting of thousands of categorized facial images, it will read the emotional state of the user. This will prove useful in situations in which being aware of social cues is a challenge for the user. Additionally, by being able to visually classify the emotion of whoever is visible on the camera feed, the program has the potential to signal blind people on the emotional state of those around them, helping them connect with others easier. Finally, after the emotional analysis is complete, a playlist that corresponds with the user’s mood is suggested to the user to be played at their convenience.

This project aims to analyze the sentiments of Donald Trump and Joe Biden on various topics, including the pandemic, SCOTUS, economy, racial justice, climate change, national security, health care, and immigration, leading up to the 2020 United States presidential elections. Sentiment analysis is the use of natural language processing, text analysis, and computational linguistics to analyze statements and determine whether they are positive, negative, or neutral about a certain topic or product. Since there are varying levels of polarity, algorithms can be used to assign these levels to statements. A statement's subjectivity can also be determined, allowing the text to be substantiated as fact or opinion. Sentiment analysis has various implications in society.

The associations between roots and fungi are called mycorrhizae. There is a mutualistic relationship between them fosters better water and nutrient absorption by plants, and can enable plants to share resources. The main nutrients required by plants for optimal growth are nitrogen, phosphorus and potassium (NPK). We hypothesized that when there is deficiency of any of these nutrients, arbuscular mycorrhizae may compensate by sharing the limited amount of the nutrient between various plants. We are studying the growth pattern of Lactuca Sativa (Buttercrunch Lettuce) in low and optimal phosphorus soil environments. We are testing the effect of low, natural and elevated levels of mycorrhizae in each of these soil environments. The growth of plants will be monitored based on leaf size, plant height, root depth and more. If the plants growing in phosphorus deficient soil compare or grow better than plants in optimal phosphorus soil, this will support our claim that mycorrhizae can replace soil nutrients.

Machine learning has plenty of real-world applications ranging from modeling the universe to computational chemistry. Because probability is the bedrock of machine learning, it is essential to optimize both hardware and software to obtain the best results. Classical computers are generally used for machine learning programs; however, learning from high-dimensional data and many features often requires excessive computer resources and may not achieve the highest accuracy. The Quantum Computing environment can be used to create a more accurate model than that of Classical Computing. To test this quantum advantage, we implemented a Quantum Convolutional Neural Network (QCNN), which parallels the classical Convolution Neural Network structure in the quantum domain. Due to the lack of quantum computers with many qubits, engineers have introduced the Noisy Intermediate Scale Quantum (NISQ) concept, which constitutes a hybrid interface between classical and quantum computers. In the QCNN context, the data processing and the cost function optimization are performed on the classical computer, while the probabilities generated by the parametric quantum circuits are evaluated in the quantum computer. Overall, the QCNN consists of a classical-to-quantum data encoder, a cluster state quantum circuit, a series of Parameterized Quantum Circuits using Quantum Convolutional and Pooling Layers, a quantum-to-classical data decoder, and a fully connected layer leading to the output. Both the Convolutional Neural Network and the QCNN extract features from data like 2D images. The networks' performances can be compared using metrics such as accuracy, loss, and time. This project has shown that the QCNN can perform better than the CNN for specific applications.

Since the outbreak of COVID-19, the economy and stock market have both been negatively affected as a whole; both the numbers and numerous studies have confirmed that to be a fact. However, there are certain stocks which are less of a focus than others. This study aims to look at the impact of COVID-19 related news articles on the energy sector, a sector which is seldom covered by the mainstream news media in regards to the pandemic. We will be using an outsourced sentiment analysis algorithm to find the sentiment of our articles, which we then will map with the price of energy stocks, renewable and non renewable, on that given time after the article was published. Our ultimate goal is to find some correlation between the sentiment of certain articles and the stock prices immediately after and draw conclusions about COVID-19 news impact on the energy sector in general.

During soil rehabilitation research, the researchers at ASDRP collect a large amount of data such as the ppm of various heavy metals, pH measurement, and soil bacterial average. Such data needs to be stored in an efficient way to ensure scalability. As a solution, we built a mySQL database to store this data. Through foreign key architecture, stored functions, and prewritten SQL code, this database can be used for future statistical research and communication from machines. This will take in data from remote instruments, physical location measurements as geographical coordinates, and lab results from instruments and be able to relate to one another through foreign key architecture. Additionally, to create more comprehensive and rapid analyses, we define functions to return date wise or id wise selection for average, count, minimum, and maximum.

Animal migration and climate change are increasingly stepping up to become the most important problems of our decade. The complexity with which natural ecosystems evolve in the face of global temperatures presents many problems for our modern era, especially for the saving of these species. On the other hand, recent developments in machine learning technology have made it possible for us to find and chart changes in these species migration patterns, particularly in Machine Learning, where we can use a modified least cost algorithm as weights for a Dijkstra's algorithm-based solution for a variety of creatures. Our project then, focuses on predator-prey dynamics in the face of these patterns, using a weighted Dijkstra’s algorithm to plot the changes to migration in a supervised learning model, making it possible for us to predict probabilistic changes to animal migration patterns in the face or climate change, and changes in an animal's habits as it tried to navigate the path to its prey, and away from its predators.