MATH AND COMPUTER SCIENCE

High School

A Deep Learning Based Approach for Efficient Diagnoses of COVID-19, Viral Illnesses (Other than Covid-19), and Bacterial Illnesses via Chest X-Rays
Vibha Addala, Jesuit High School

This research focuses on using deep learning to accurately diagnose if a person has a COVID-19, a viral illness other than COVID-19, a Bacterial Illness, or none.

A Systematic Approach to Assessing the Link between Parkinson’s Disease, Neuromelanin, and Aging using Gene Expression Data
Sahand Adibnia, Dublin High School

My research was designed to identify biological processes and individual genes involved in age-dependent neurotoxic neuromelanin accumulation in Parkinson’s disease through gene expression analysis. These genes could serve as therapeutic drug targets and reveal new biological pathways involved in this process. Correlations of gene expression in the substantia nigra with age and differential expression analysis of the melanized substantia nigra with respect to the non-melanized ventral tegmental area were performed.

Teaching a Neural Network to Solve Calculus Problems
Spencer Bauman, Pine Crest School

This research focuses on using a long short-term memory neural network to integrate complex functions.

A Deep Learning Machine Vision System to Prevent Sudden Infant Death Syndrome
Vivek Bharati, Homestead High School

Sudden Infant Death Syndrome (SIDS) causes infants under one year of age to die inexplicably. One of the most important external factors responsible for the syndrome, called an ‘outside stressor’, is the sleeping position of the baby. We propose a Convolutional Neural Network (CNN) based machine vision system that can alert caregivers on their mobile phones within a few seconds of the baby moving to a hazardous face-down sleeping position.

Developing a Voice Recognition Program for Search and Rescue Using Deep Learning Convolutional Neural Networks
Grace Chen, Catlin Gabel School

As sound is pervasive and ever-present in daily life, many studies regarding the development of sound recognition systems have risen in prominence. While multiple studies have trained neural networks to classify different types of human voices or environmental sounds, hardly any studies have investigated the use of neural networks to identify human voices from environmental sounds. Additionally, while some studies have developed search-and-rescue (SAR) technology using image recognition techniques to detect humans in the wilderness, comparatively fewer studies have used sound recognition techniques for the same purpose. As sound can travel through a medium whereas light can only travel through empty space, sound recognition would be ideal for SAR technology when navigating difficult landscapes, such as mountainous regions and dense forests. Since human voices have different frequencies than ambient noise, this research hypothesizes that a deep learning convolutional neural network can be trained to accurately distinguish between human voices and natural sounds.

The Simulation and Analysis of Random Bipartite Graphs Synthesized to Model Statistical Properties of Datasets for Computer Diagnostics
Carla del Rio, American Heritage School Plantation

The purpose of this experiment was to simulate and analyze random graphs synthesized to model the statistical properties of datasets for computer diagnostics.

Rip Current Detection - An Orientation-Aware Machine Learning Approach
Boglarka Ecsedi, Istvan Bocskai Secondary Grammar School

An ever-changing hazardous natural phenomenon – called a rip current – causes numerous fatal accidents all over the world. To address this problem, I developed an orientation-aware image processing algorithm to detect and localize rip currents using the framework of a powerful near real-time deep neural network called Faster R-CNN. The development resulted in detecting rip currents with higher efficiency, allowing the algorithm to adapt to many angles, positions of the object, and different perspectives. The developments are applicable for real-life situations, such as an automated rip current detection system using web cameras or a mobile application. This approach contributes to the deeper understanding of rip currents, to the early identification of the hazard, thus preventing accidents and protecting human lives.

The Non-Homogeneous, Incompressible Navier-Stokes Equations
Benjamin Faktor, Canyon Crest Academy

I developed new results related to the analysis of the Navier-Stokes Equations by considering viscosity to be dynamic over space and time. These new results have been proven to be more accurate than those preceding it.

A Novel Machine Learning Pipeline for Accurate COVID-19 Prediction and Risk Factor Identification using Longitudinal Electronic Health Records
Alice Feng, The Harker School

Current COVID-19 predictive models primarily focus on predicting the risk of mortality, and rely on COVID-19 specific medical data such as chest imaging after COVID-19 diagnosis. In this project, we developed an innovative supervised machine learning pipeline using longitudinal Electronic Health Records (EHR) to accurately predict COVID-19 related health outcomes including mortality, ventilation, days in hospital or ICU. In particular, we developed unique and effective data processing algorithms, including data cleaning, initial feature screening, vector representation. Then we trained models using state-of-the-art machine learning strategies combined with different parameter settings. Based on routinely collected EHR, our machine learning pipeline not only consistently outperformed those developed by other research groups using the same set of data, but also achieved similar accuracy as those trained on medical data that were only available after COVID-19 diagnosis. In addition, top risk factors for COVID-19 were identified, and are consistent with epidemiologic findings.

Reflection of Political Bias within YouTube Search and Recommendation Algorithms
Sanjana Gadaginmath, Lynbrook High School and Michael Lutz, Bellarmine College Preparatory

Given the reach of YouTube as a proliferator of contemporary news and ideas, it is important to understand how YouTube reflects (either magnifies, minimizes, or preserves) pre-existing political bias. Principally, our research aims to provide a novel understanding of YouTube’s reflection of user base political bias within its search and video recommendation algorithms.

A Novel Approach to Citrus Disease Management: Leveraging Computer Vision, Machine Learning and Convolutional Neural Networks
Arko Ghosh, C. Leon King High School

Huanglongbing (HLB) is an incurable disease that affects citrus trees. This citrus disease is also often referred to as "citrus greening". This disease was first spotted in China in the early 1900s and was kept under check for a century. The disease started infecting plants in Florida in 2005 – the largest orange-producing state in the United States. By 2012 HLB had spread much of citrus farms in the entire state of Florida and started to wreak havoc in the billion-dollar citrus industry in Florida. The disease also threatens the citrus industry in California and other states. There are microscopic, molecular, and spectroscopic techniques to detect HLB on citrus plants today but require a laboratory setting. These techniques are practically hard to implement on a large citrus grove. The only way growers could manage this disease today is by visually inspecting the plants and then removing the infected plants as the best way to control the problem. In this research project, an AI-powered method and process is defined, and an economically sustainable system is developed to identify HLB disease on the citrus plants and provide early detection signals on the citrus growers. The goal of this project is to research and develop a predictive model using Computer Vision, Artificial Intelligence, and Deep Learning using Convolutional Neural Networks to identify citrus leaves infected with the HLB disease and to implement a remotely piloted aerial system to detect the citrus leaves infected with the disease in citrus groves using the predictive AI model.

Determining the Optimal MRI Sequence for the Automatic Segmentation of Multiple Sclerosis Using Convolutional Encoder Networks
Shaurnav Ghosh, Pine Crest School

Multiple Sclerosis (MS) is a neurodegenerative disease that is caused by white matter lesions. These lesions are simply brain scars that often affect older patients and can cause severe damage to the motor and sensory functions. The diagnostic procedure calls for a process known as segmentation which is the detection of lesions within a patient's brain. This is an arduous task that is left to be done manually by expert radiologists. However, even with expert radiologists segmentation is very time-intensive taking upwards of 60 minutes per patient. Additionally, radiologists have to wait to obtain three different types of MRIs before the process can start at all. Thus, this study sought out to use machine learning to determine an optimal MRI and automatically segment a patient's lesions.

Increasing the Accuracy of Deepfake Detection Using an Ensemble of Convolutional Neural Networks
Chris Gu, The Pingry School

First, I briefly discuss what deepfakes are. They are defined as fake images of people created through the utilization of deep learning techniques. More importantly, they are dangerous as they have manipulated elections and have facilitated the spread of misinformation. However, differentiating between real and deepfaked images remains imprecise and inaccurate. Convolutional Neural Networks (CNNs) are most prevalently researched in deepfake detection. In my research, I successfully attempted to increase the accuracy of deepfake detection through an ensemble method. I first gathered multiple CNN base models designed by other researchers that determine if an image of a person is real or deepfaked. I then input the predictions from each of these models into a Deep Neural Network that I designed, which outputs a final prediction. The method of making predictions from multiple base models and then analyzing them using another deep learning technique in order to produce a final prediction is called ensembling. The result of my experiment was that the accuracy of my ensemble model was higher than that of the best base models. Thus, my research project indicated that ensembling the prevailing deepfake detection techniques has the potential to increase overall detention accuracy.

Bounding Polygons: Adding Precision and Resemblance to Region Proposals Through Iterative Slicing
Nicholas Jiang, University of California Santa Cruz

This project examines the effectiveness of an iterative slicing framework to finetune traditional object detection models like Faster R-CNN and YOLOv3.

Prediction of Drug-target Interactions For Selective Androgen Receptor Modulators (SARMs) Using Machine Learning Methods
Soh Yun Kang, Sungjoon Kang, and Nicholas Tran; Admission AG

Early clinical studies have demonstrated potential uses for the selective androgen receptor modulators in the treatment of cancer-related cachexia, benign prostatic hyperplasia, hypogonadism, and breast cancer, with positive results. Accurate prediction of drug–target interactions (DTI) is crucial for drug discovery. We present an implementation of a machine learning model that is able accurately to characterize the binding of compounds to the drug targets. In this project, we tested the performance of the machine learning model on the set of selective androgen receptor modulators.

Concatenated CNN and Transformer-Encoder Neural Network to Diagnose Cancerous Thyroid Nodules in Ultrasound Cine Images
Tara Kapoor, Palo Alto High School

Thyroid cancer prevalence has increased notably in recent years. Biopsy of nodules is invasive and expensive, and risk stratification of ultrasounds is done manually by radiologists through TI-RADS system, motivating the need for effective computerized diagnosis solutions. I developed a novel deep-learning algorithm for diagnosis: MobileNet-v2 CNN for feature extraction concatenated with a Transformer-Encoder network (2-head encoders, self-attention/feedforward sub-layers). I addressed two main technical challenges: extreme class-imbalance (benign/malignant ratio=175/17) and processing patient-wide cine-clip sequentiality. For the former, I incorporated weighted oversampling and focal loss (alpha=0.9, gamma=2.4). For the latter, I stacked CNN inputs (adjacent or equally-spaced patient frames), then channeled extracted feature vectors into the Transformer-encoder for patient-wide attention-mechanism processing. My deep-learning techniques—adjacent-frame (AUROC=0.867) and equal-spaced-frame (AUROC=0.858) model —performed superior to manual radiologist TI-RADS scoring (AUROC=0.798), meaning my deep-learning models can advance diagnosis efficiency and patient outcomes.

CropMates: A Low-cost And Efficient IoT And Al- Based Integrated System For The Detection And Treatment Of Crop Diseases And Deficiencies That Maximize The Quality And Quantity Of Crops To Potentially Eradicate World Hunger
Shraman Kar, duPont Manual High School

CropMates is an integrated system that detects disease and nutrient deficiencies in the plants and recommends appropriate pesticides and fertilizers. CropMates uses the state of the art machine learning and IoT technologies.

Automated Recyclable Detection in Industrial Processing Facilities Using Deep Learning
William Kirschner, Pine Crest School

Given the difficulty of sorting recyclables, we propose a deep-learning-based solution that is capable of identifying contamination in waste streams within industrial processing facilities.

An AI-Based Global News Service using Cross-lingual Text Summarization and Translation
Riley Kong, Archbishop Mitty High School

Current methods for understanding news in foreign countries are lacking and result in little awareness of events around the world. To solve this problem, we present a global news service that dynamically interprets news by performing cross-lingual summarization (CLS). A website is built that queries the user for an article or news site and returns corresponding CLS summaries. The model used in our service improves on existing CLS models by combining features from a transformer-based and extractive-based summarization model, which dramatically improves the readability and faithfulness of summaries. Using this service, users are able to easily understand global news at a glance and stay up to date with current events.

Using Different Machine Learning Classifiers to Examine the Relationship Between Prodromal Huntington’s Disease and miRNA Expression in Blood
Deeksha Kumaresh, American Heritage School Boca/Delray

This presentation outlines the research conducted to determine whether there is a relationship between microRNAs and Huntington's disease in order to assess the feasibility of using miRNAs as a treatment for Huntington's.

SeizureSeeker: A Novel Approach to Epileptic Seizure Detection Using Machine Learning
Hamza Lateef and Gabriel Ralston, Charles J Colgan High School

The main objective of this research was to develop an efficient method for analyzing EEG data using machine learning algorithms that allows for complex data processing. This way, healthcare professionals can automatically distinguish between normal EEG signals and epileptic seizures.

Reducing False Positives in Pulmonary Nodule Detection Using Convolutional Neural Networks
Chaeyoung Lee, Tenafly High School

Developed a deep learning model that reduces false positives in lung nodule detection using CNNs. The model increases the overall reliability and accuracy of lung nodule detection. Could potentially be applied to other detection systems.

Using Instance Segmentation to Determine the Movement of a Point on its Differentiated Graph
Jake Malis, Pine Crest School

Despite the rise in popularity of educational apps designed to help students better understand math with useful, artificial intelligence-backed tools, many features are missing in these apps because of technological limitations. Photomath - the second-ranked educational app on the Apple App Store - is a valuable mathematics app that allows the user to take pictures or type in math problems and see a step by step guide on how to solve it; it is unique for its ability to input math problems taken from pictures but falters in its inability to use pictures of graphs as an input. This research aims to use more advanced neural networks to create additional educational tools to be added to apps such as Photomath, this being a functioning network that can take in a picture of a graph and determine its movement: whether the represented point is accelerating, decelerating, or stationary; whether its position is increasing, decreasing, or stationary; and determine when any of the values of the graph are negative. The results of this research will significantly benefit students studying calculus and eventually other mathematics courses.

Augmented Reality Chess Analyzer (ARChessAnalyzer): In-Device Inference of Physical Chess Game Positions through Board Segmentation and Piece Recognition using Convolutional Neural Networks
Anav Mehta, Cupertino High School

A novel method to provide an instantaneous feedback learning tool for chess players using a combination of computer vision and machine learning models. This method can be used to analyze a multitude of other physical games. This app and technologies underneath will help chess learners improve their game and hopefully will be widely used in chess clubs.

Improving Diabetic Diagnosis and Prevention with Machine Learning on Retinal Imaging
Yushan Min, American Heritage School Boca/Delray

Using four different machine learning models and test which one gave the best results for detecting signs of diabetes from retinal images.

Using Automated Infant Posture Recognition to Reduce SIDS Risk
Isha Narang, Ardrey Kell High School

According to the CDC, approximately 3500 infants die annually in the United States from sleep-related infant deaths, including Sudden Infant Death Syndrome (SIDS). These deaths occur among infants less than 1 year old. The American Academy of Pediatrics (AAP) recommends supine positioning for infants: sleeping on the back. My project’s purpose is to find an efficient way to alert caregivers when an infant’s posture is unsafe (high-risk as per the AAP).

The Effectiveness of Deep Learning Models on the Detection of Pneumonia
Maya Neeranjun, American Heritage School Plantation

The purpose of this experiment was to find out if deep learning models accurately diagnose patients with pneumonia. During a time of crisis, there are a limited amount of doctors to aid patients. As a result, these different models could potentially be the key doctors rely on.

Mathematical Modeling, Analysis, and Simulation of the COVID-19 Pandemic with Behavioral Patterns and Group Mixing
Comfort Ohajunwa, The Governor's School at Innovation Park

This presentation will focus on a novel mathematical modeling framework for studying the impact of behavioral patterns and group mixing on the spread of COVID-19. In our work, we model the interactions and behaviors of two groups of people: one group with normal behavior who do not reduce their social contacts and one group with altered behavior who reduce their contacts through interventions, such as social distancing and confinement. Furthermore, we show the basic reproduction number for our model, provide the results of our numerical simulations, and display a dashboard to share our work.

Improved Classification and Prediction of Head and Neck Squamous Cell Carcinomas Using A Novel Generative Adversarial Network Model
Ashwin Parthasarathy, American Heritage School Boca/Delray

The purpose of this project was to develop a model that can take an input of radiographic images of head and neck squamous cell carcinoma (HNSCC) and provide a diagnosis of the stage of cancer and the predicted stage progression of the disease with state-of-the-art accuracy. To do this, I created a novel generative adversarial network (GAN) model, a type of machine learning architecture that uses multiple sub-models, and both trained and validated this model with publicly available CT scans. Using this research, doctors will be able to diagnose and predict the future development of HNSCC with high accuracy in a completely automated fashion.

Using Natural Language Processing to Detect Drug Discussion on Reddit
Adam Peles, Pine Crest School

Addictive and detrimental substances/drugs can cause much harm for the people that use them and their loved ones. Identifying drug trends early on can help alleviate these destructive effects. My research explores the possibility of using artificial intelligence to detect discussion regarding novel substances on the social media platform, Reddit.

Designing a Recurrent Neural Network (RNN) using Deep Learning and SVM modelling to Analyze Covid Pattern
Risha Raman, Saratoga High School

During the pandemic, 40% of adults reported having anxiety or depression regarding coronavirus as well as fear of the unknown. The project, "Designing a Recurrent Neural Network (RNN) using Deep Learning and SVM modelling to Analyze Covid Patterns" educates people on the recent coronavirus cases, hospitalizations and deaths in Santa Clara County. Furthermore, the SVM modelling makes potential predictions for the upcoming trends ahead.

BLEPNet: In-Silico B-Cell Epitope Prediction Using Deep Learning
Mihir Rao, Chatham High School

B-cell epitope prediction plays a critical role as a computational alternative to existing time- and resource-consuming epitope mapping approaches. Epitope prediction is important in early vaccine development stages, as it can identify highly immunostimulant regions of antigens that can be implemented in the vaccine. My research introduces a novel B-cell epitope prediction system that has significantly higher predictive performance compared to existing tools.

Implementation of Time Frequency Analysis for Seizure Localization: Phase II
Asha Reddy, Lake Highland Preparatory School

Epilepsy is generally diagnosed using electroencephalograms, or EEGs, which detect the electrical activity in a person's brain. Signals recorded from neurological tissue are extremely noisy, causing uncertainty in the analysis of EEG scans. The purpose of my research is to find an alternative method of modeling to improve epilepsy analysis.

Comparing the Effectiveness of an Artificial Neural Network to Logistic Regression in Predicting the Diagnosis of Pancreatic Cancer
Ruhi Reddy, American Heritage School Boca/Delray

Pancreatic cancer is the fourth leading cause of death from cancer around the world. This research compares the effectiveness of using an artificial neural network in comparison to using logistic regression in order to predict the diagnosis of pancreatic cancer given certain factors about individuals.

Comparing the Efficiencies of Various Convolutional Neural Networks on the Recognition of Melanoma
Anshuman Seetharaman, American Heritage School Boca/Delray

Using images of skin lesions as well as 3 convolutional neural networks that I programmed using python, I aimed to compare the efficiencies of the network. I tested each network against the image set to see which network would perform the best when given the task of recognizing malignant lesions.

Are You Social Distancing? A Computational Approach
Aparnaa Senthilnathan, The Bronx High School of Science

In the wake of the coronavirus pandemic and a lack of an economical system, a mask recognition program using convolutional neural networks, fundamental tools of artificial intelligence, was created with goals of big functionality with a wide range of possibilities, ranging from an analytical tool to calculate the percentage of masks to a function on security cameras inside public buildings.

Using a Convolutional Neural Network (CNN) to Distinguish between Benign and Malignant Acute Lymphoblastic Leukemia (ALL) cells
Shreya Shenoy, American Heritage School Boca/Delray

This project utilizes a convolutional neural network based on the SqueezeNet model to distinguish between images of benign and malignant Acute lymphoblastic leukemia cells.

Descriptive Epidemiology on Vestibular Schwannoma
Kailey Takaoka and Kaitlyn Greppin, Hathaway Brown School

Statistical Analysis on Vestibular Schwannomas.

The Effects of Quantized Parameters on the Accuracy and Efficiency of Neural Networks
Ritvik Teegavarapu, American Heritage School Boca/Delray

Artificial neural networks are computation systems that belong to a class of machine learning approaches under the general subset of artificial intelligence, which is used to recognize patterns and create functions based on data provided as input. The primary input of most artificial neural networks are numbers, usually real numbers with a large number of trailing digits. Quantization, however, is the process of reducing the number of bits used to represent the number. While performing this significantly improves the bandwidth and the storage needed, it also reduces the accuracy of the neural network. Thus, this experiment will attempt to determine how to optimize the amount of quantization such that accuracy and efficiency are relatively preserved. This experiment was able to definitively determine that intermittent quantization added was more effective than back-end quantization, or optimization introduced at the end of processing a data set.

Defining Curve Behaviors on the Surface of a Möbius Strip with One Boundary using the Double Cover
Lio Thomas, The Bronx High School of Science

The Mobius strip is a simple one-sided non-orientable 3D object that can be constructed by giving a half twist to a rectangular surface and attaching the ends. In topology, properties of an object are preserved, even if the object is moved, bent, stretched, or twisted; however, altering a surface, such as cutting or puncturing, can drastically change its properties. In my paper, I explore the properties of a surface when a hole or boundary is opened. One way to study the properties is to simulate walking on the surface in such a way that we can trace our way back to the starting position. By doing this, we can understand the topology of the surface better, such as how the surface will behave when it is subject to topological deformation. In this paper, I find the shortest paths that maintain the properties of this new derived surface. This paper proves that there are only 3 basic directions and 4 basic paths that can be taken on a Mobius strip with a hole.

Optimization of the ADME (Absorption, Distribution, Metabolism, and Excretion) properties of the drug molecules using machine learning methods.
Thao Tran and Jiwoo Yoo, Admission AG

ADME influences the performance and pharmacological activity of the compound as a drug.Poor compound solubility, chemical instability and inability to permeate the intestinal wall can all reduce bioavailability of a drug. Drugs that absorb poorly when taken orally must be administered in some less desirable way, like intravenously or by inhalation.Machine learning methods are useful tool in prediction of ADME properties.In this project we study performance of machine learning model Message Passing Neural Network (MPNN) as ADME predictor.

Achieving Quantum Parallelism by Analyzing the Interaction between Particles through a Novel Design of Generalized Grover's Algorithm
Christopher Um, Torrey Pines High School

The presentation is a walkthrough of my research in Grover's algorithm, and it has been formatted as I have submitted to other science fairs and following Sigma Xi's guidelines.

Ranking Countries’ Responses to the COVID-19 Pandemic Using Learning to Rank
Alexander Wilentz, Pine Crest School

This study ranked countries based on their response to the COVID-19 pandemic using learning to rank (LTR). From these rankings, insights into the best or worst policies and practices when responding to a pandemic such as COVID-19 can be made.

Utilizing Mathematical Modeling to Simulate Combination Drug Therapies & Elucidate the Differential Effects Stromal Fibroblasts Impart on Breast Cancer Sensitivity to Lapatinib
Jonathan Williams, Pine Crest School

My work utilizes mathematical modeling to simulate how combination drug therapies affect the growth of HER2+ Breast Cancer Cells.

Optimizing COVID-19 Interventions in Florida
Megan Yang, American Heritage School Plantation

I designed a neural network and used reinforcement learning to first predict the rise and decline of COVID-19 cases and then use that to design a plan to hinder the spread of COVID-19 while taking into account not only the number of cases but things like economy, education, and precaution against harmful short-sighted decisions.

Undergraduate

An AI application to Diagnose Melanoma by Analyzing Image Files
Sneha Iyer, University of Virginia

Automated diagnosis can help detect, treat, and prevent threatening diseases. This project utilizes a convolutional neural network (CNN) built using the FastAI Library, OpenCV Library, and Google Colaboratory that classify Melanoma images as cancerous vs. noncancerous. Melanoma can become deadly in as little as 6 weeks. Thus, it is important for a solution to be developed that aids in the early-stage detection of cancer. The app allows the user to upload a picture. Once they do that, the image is fed to the Flask server and the CNN classifier returns a result (positive or negative). The app also features a comprehensive platform so users can access resources such as MayoClinic, cancer support centers, and other information. Due to its malignant nature and tendency to affect relatively young individuals, the timely diagnosis of Melanoma is very important.

Visualizing Dichotomous Data Correlations Using Two-Sample Corrgrams
Rithika Tummala and Rohan Tummala, Vanderbilt University

We introduce the two-sample corrgram, a novel graphical representation we created to display correlation matrices for two samples simultaneously. We also introduce “corrarray,” an R package we developed to streamline the generation of 2-sample correlation matrices and hence corrgrams to more efficiently visualize multivariate correlations of dichotomous data.

Augmented Reality Version of The LEGO Game
Jiahao Yang, Nankai University

There are five main parts in the presentation. All of these parts are around one topic, that is the AR version of the LEGO game. Within the URL, you can see how the work is done and how much I want to make this programme into reality.

Automatic Keyword Detection from Social Media
Yifei Yue, University of Nottingham Ningbo China

With the rapid development of social media such as Twitter and Weibo, detecting keywords from a huge volume of text data streams in real-time has become a critical problem. In this project, I propose a novel method combining the TF-IDF and LDA models to better cope with the distinct attributes of social media data to address the keyword detection problem.

Graduate

Deep Reinforcement Learning Based Self Driving Car
Chandrasekaran Sakthivel, Colorado State University

My research is focussed on
(i). Motion-based on object recognition. The proposed method identifies and differentiates objects according to their speed.
(ii). The artificial intelligence aspect of autonomous self-driving is under consideration. It is essential to establish the training of an autonomous driving AI agent through reinforcement learning is introduced.
(iii). In this thesis, a direct perception approach is proposed to drive a car on a road or the highway environment.

MATH AND COMPUTER SCIENCE

High School

Undergraduate

Graduate

About

Programs

Chapters

Meetings & Events

Members

News

Publications