Coronavirus Trend Analysis – A Realtime COVID Tracker Via Hashtags & Emojis (Abu Shoeb)

Emojis and HashTags have quickly become ubiquitous and they often help researchers to mine targeted information especially from the social media data. In my Ph.D. thesis, I have shown how emojis can help detect emotion intensities for tweets. A similar approach can be applied to Twitter data to grasp the negativity and the positivity of real-time tweets related to COVID. First, our AI system will filter and only collect Coronavirus related posts from Twitter. Then, this will be divided into positive and negative buckets based on their sentiments and emotions. Later, a web application will be developed to demonstrate both positive and negative posts side by side along with their intensities and demographic information. This would help doctor, researchers as well as policymakers to quickly visualize the current trend of the COVID along with the curated data for further research.

Simulation of Human Behaviors in Virtual Scenes for Pandemic Prediction and Strategy Making (Che-Jui Chang)

Predicting the pandemic and making efficient strategies are very difficult mainly because of the large scale of human populations and the high complexity of human behaviors. In this proposal, I want to leverage efficient and accurate human behavior modeling methods to simulate real-world human activities. By adding a transmission model to the simulation, we can understand how the pandemic spreads among people. Besides, trials can be made in the virtual scene to find effective strategies. Applications include (a) prediction of the pandemic, (b) visualization of spreading patterns, (c) evaluation of social distancing/quarantine policies, and (d) crowd flow design to avoid contacts and transmission.

The core techniques of this proposal may include scripting, crowd simulation, and human contact modelings. Optimization of parameters in deep learning models as well as decision-making agents can also be done with the simulation, allowing non-heuristics based strategies to be discovered to combat the pandemic.

Ultra-Wind Band for Contact Tracing and Social Distancing Application (Sanyam Jain)

Social distancing has become the new normal in this COVID pandemic. Health experts believe that even after a vaccine is developed, social distancing will be essential to prevent the spread of COVID. Complete lockdown, though an effective measure to achieve social distancing, can be detrimental to world economics. Thus there is an urgent need for an innovative solution to track and maintain social distancing. This study aims to propose and analyze the Ultra-Wind Band based application to monitor social distancing in public and workplaces, compare UWB based approach with existing approaches based on RFID, Bluetooth, Computer Vision, Thermal, etc. These technologies will be compared based on accuracy, privacy, scalability, and reliability. High localization accuracy, immunity to multipath and interference, low latency, and security of UWB communication makes UWB a strong candidate for contact tracing application. This study’s main challenge is to deploy and test these social distancing applications as a practical solution that must be tested in real-life conditions, which is difficult to achieve with an increasing number of COVID cases.
Keywords: Social distancing, UWB

Written All Over Your Face: Analyzing stochastic signatures of audio and facial expressions (Hannah Varkey)

Perhaps the best piece of advice I have ever gotten for hosting on-air radio was ‘to smile and speak’— it made a difference to my voice when I smiled compared to neutral or other facial expressions, making it more engaging, confident and easier to listen to, despite none of the listeners seeing my face. Taking this idea further, I wonder if there is a way to discern emotions and expression from simple voice audio in a more systematic way, with tools of data science.

At the Rutgers Sensory-Motor Integration Lab, we investigated different kinematic trajectories from pixels on regions of face, particularly those innervated by the trigeminal nerve (V1, V2, V3). Using real-time analysis software, we can measure micro-movements of the face seen in each frame of a video or virtual meeting. Referencing Dr. Paul Ekman’s work on the seven universal facial expressions, we could track the micro-movement velocities on a Gamma parameter plane and compute their probability distribution functions to parse out stochastic signatures of these expressions. With this, we could find the personalized signatures of subtle variations in the seven universal micro-expressions of the human face. Attaining a personalized signature would help us track the facial micro-expressions dynamically, as the person talks, particularly if we knew how they change when the person is staging a narrative versus when the narrative spontaneously flows. This would enable us to integrate these biomarker signals into data that can be compared across different emotions for an individual as well as different populations.

It would be interesting to develop this idea towards a project aimed to find the signatures of the person’s emotions in the person’s voice. Are there also universal micro-expressions conveyed through sound? How do features of the person’s voice map onto that person’s universal facial micro-expressions? I propose to look for correlations between facial expressions, voice and auditory responses of an interlocutor, to figure out if the face-voice person’s signatures uniquely evoke another person’s emotional responses.

I would like to visualize a consented use of this project in fields of telemedicine, patient and mental health screenings. It would be intriguing to delve into dyadic facial expressions where those of a patient may be masked in conditions like depression and PTSD. Beyond this, imagine a Rutgers student calling in to a CAPS or VPVA crisis hotline and speaking with a counsellor. While different scenarios may arise from the student not always being comfortable to share information to not realizing they are experiencing symptoms of mental illness, added tools of real-time data analysis could enable us to receive significant information that may lie beyond the eye or ear.

Predicting Alcohol Consumption from Cortical Activity Using Machine Learning and Telehealth (My Nguyen)

Alcohol use disorder is characterized by loss of control over drinking behavior and increase in the amount of alcohol intake over time despite adverse consequences (Logrip et al., 2015). According to recent studies, there has been an increase in alcohol consumption and its associated mental health concerns during the COVID-19 pandemic (Pollard et al., 2020). Due to the improved safety and convenience of telehealth, alcohol use disorder can be detected early with the help of artificial intelligence to predict drinking behaviors. Previous work shows that direct projections from the prefrontal cortex (PFC) to the striatum play a key role in alcohol-seeking behaviors and alcohol use disorder (Cheng et al., 2018; Nam et al., 2013; Wang et al., 2010). However, the precise mechanism through which excessive drinking affects the PFC-striatal pathway to promote alcohol-seeking and consumption remains unknown. In Dr. Huda’s lab, we previously recorded the activity of corticostriatal neurons in the anterior cingulate cortex (ACC), a subdivision of the PFC, using two-photon calcium imaging as head-fixed mice actively drank alcohol over days. In this study, I will perform a more in-depth analysis of how the activity of the ACC-striatal neurons relates to alcohol consumption. In particular, I will use a machine learning classification approach to map neuronal activity onto specific aspects of alcohol drinking behavior. This project will apply advanced analytical techniques to identify the relationship between neuronal activity and alcohol consumption, and has the potential to discover novel neural circuit-based therapeutic targets for addressing alcohol use disorders. These data can then be integrated into telehealth devices to correctly predict excessive alcohol consumption from the patient-generated neural activities.

Using Artificial Intelligence to Improve Health Information Interactions about the Coronavirus (Zaynab Khan and Alex Nguyen)

The United States continues to battle the coronavirus as we approach a second wave. Even though it has been 9 months since the World Health Organization declared the novel coronavirus a pandemic, there is a persisting notion that the coronavirus is not a real threat, in part due to the extent of misinformation on social media. Pew Research reports that 5 months into the pandemic, around 25 percent of Americans believed some conspiracy theory about the coronavirus. The denial of coronavirus related risks has made it difficult for public health and government officials to enforce compliance for mitigation efforts, such as wearing a mask. The COVID-19 Vulnerability tool, an online program created by the non-profit Social Progress Imperative, has analyzed data to indicate vulnerability to the coronavirus based on location. This tool can be used to identify which locations need stronger and more effective public health messaging regarding the coronavirus. Furthermore, to communicate accurate information about the coronavirus, messaging needs to be tailored to the user. Research from Rutgers SC&I professor, Dr. Kaitlin L. Costello and University of Michigan professor, Dr. Tiffany Veinot, found that individuals react to health information differently. This research categorized health information interactions into five different approaches: avoiders, receivers, askers, seekers, verifiers. The research also delves into methods of effective communication for each category. Based on the user’s health information reaction type and location, artificial intelligence can leverage search engines and social media algorithms. The user will be better able to understand and believe personalized communication regarding the pandemic. For example, avoiders are characterized to have low trust in healthcare professionals—so search results can prioritize local government announcements and press releases over CDC guidelines. In accordance with demographics and previous internet behavior, artificial intelligence has the capability to classify the avoiders and later implement this individualized communication. More effective messaging about the virus in places that need it the most will be critical in combating this crisis.
Keywords: coronavirus vulnerability, health information, communication, artificial intelligence

MediMap! (Alexander Suponya)

A potential open-source software that could aid in distribution of medical supplies to locations where they are most urgently needed. Throughout the course of the current COVID-19 pandemic, it has become clear that it is difficult for an overwhelmed bureaucracy to properly coordinate the distribution of PPE, ventilators, and other medical supplies to various hospitals across the country. At the same time, viral symptoms have spread rapidly in regions where little to no precautions have taken place to actually work with COVID patients, and the distribution of vaccines, such as those recently developed by Pfizer, will also be subject to the same biases or miscalculations in supply chain distribution. MediMap is a tool that would use a convolutional neural network to analyse areas which have a higher online search frequency for symptoms related to COVID, such as breathing problems or a loss of smell and taste, and would highlight the given areas as sites of higher urgency when it comes to dealing with the vaccine. This could help medical companies and government officials make the best decisions when it comes to distributing more vaccines or other supplies where they are genuinely needed. Symptoms for other diseases can also be analysed by MediMap if trained on search results in areas where a specific disease is known to have occurred at a high frequency. This makes MediMap a versatile tool for displaying where certain diseases are prevalent more than others, even when the country isn’t in a state of pandemic.

A Neural Network to Predict Healthcare Collapse (Andrew Petryna)

The recent 2020 Coronavirus Pandemic has highlighted key vulnerabilities in existing medical and informational infrastructure. Diverse nations across the globe have seen the cyclic rise and fall of virus “waves”, each unique in demographics, geography, and impact. Unlike previous global pandemics however, now the exponential digitalization of the world has allowed data regarding all forms of epidemiological information to be recorded by various government and private organizations at an unprecedented rate. If this flood of information can be properly analyzed and interpreted, epidemiological patterns of a certain disease can be recognized and used to enact vital changes and preparations to prevent healthcare system collapse and save many lives. The complex and multifaceted nature of a pandemic would be best interpreted by an Artificial Neural Network whose machine learning efforts are directed at predicting the strain placed on healthcare systems and resources by specific patterns of virus spread. The fundamental prediction generated would be whether or not a healthcare system will collapse in a given timeframe. This question hinges on whether any of the three most fundamental resources of a healthcare system; Available Staff, Facility Capacity, and Equipment/Medications, will be exhausted. The initial processed outputs of such a network would have its node and connection schemas built on existing epidemiological prediction models. It would then be trained using the data gathered from past or ongoing outbreaks to learn which factors are most taxing on a healthcare system. After proper training this network would be able to recognize areas that are showing the early signs of a healthcare system at risk of collapse weeks before it happens, and can allow authorities to take steps to prevent catastrophe.

Using AI to Determine Outbreak Potential Across Towns and Cities (Devin Grenard)

As the pandemic hit areas in the United States, the response to COVID-19 has been largely reactionary. Lockdowns, mask mandates, and government warnings only seem to occur after a massive rise in cases rather than before, often eliminating the idea of any preventative measures beyond mask wearing, social distancing, and waiting for a vaccine. The use of AI could change that and provide local governments and scientists another preventative measure to stop the spread of COVID-19 and possible outbreaks in the future. The amount of data that has been collected over the past 9 months could be instrumental in developing a predictive AI that would be able to assess the risk posed by COVID-19 on specific towns and cities. By isolating a specific town or city and feeding an artificial intelligence factors for each day since the outbreak such as population density, number of sick individuals within a town, average rate of transmission over the past 7 days, and others, the AI would be able to weight each factor’s impact on the spread of COVID-19 and, from those weights, predict when a local outbreak may occur for the days to come. This prediction could then be used as a warning and ultimately save lives if used in a preventative manner.

Social Distancing in the Subway: Multi-Agent Simulation of Public Transportation Regulations (Girish Ganesan)

In 2019, the NYC subway system had an average daily ridership of 5.5 million New Yorkers; due to a combination of public awareness and disease control measures enacted by the MTA, today’s ridership hovers around 1.7 million. Prompted by a catastrophic fall in revenue, MTA Chairman Pat Foye emphasized in a video-conference with the MTA board that “we need $4 billion, but we’re only at the starting point of the crisis,” implying that more federal bailout money will be needed to cover subway’s running costs the longer the pandemic is uncontrolled and the curve untamed. This begs the question: In anticipation of a gradual reopening of the economy, how can public transportation agencies begin to resume their functions in larger capacity without jeopardizing the safety of their passengers and employees? In particular, what strategies and guidelines should they adopt in order to minimize the spread of infection while minimizing boarding/disembarking delay?
I propose to build a robust, flexible software in Unity and C# that models passenger behavior and infection spread. Users will have the capacity to affect both of these observed phenomenon. To influence passenger behavior, users can

  1. encode a passenger Psyche. This is a basic ruleset each individual passenger follows in order to minimize both their risk of infection and their boarding delay. For example, the WHO’s injunction to keep “six feet apart” can be implemented here through an API.
  2. enforce centralized Guidelines. For example, New York City considered making it a misdemeanor offense to sit right next to another passenger on benches when an open, isolated seat was available. The effectiveness of that policy could be tested out here.
  3. establish Beacons and triggers. When some time-based condition is met, agents must head to their associated Beacons in a timely manner. For example, one Beacon could be the arrival of the Line A metro, on which some passengers will intend to board but some will have no interest in (with their own intended destinations).

To view the effect of disease characteristics on the spread of infection, users can

  1. change the “interval of risk”, a parameter which reflects the method of transmission of the disease. So far, it is known that COVID-19 spreads in droplets expelled by coughing and sneezing, so this parameter would have some value between 0 and 1 reflecting the frequency of cough by a COVID-19 patient. Conversely, a disease that spreads from skin-to-skin like the leprosy bacterium would have an “interval of risk” of 0 since there is a constant risk of exposure.
  2. change the “radius of transmission”. Here, COVID-19’s value for this parameter might be around 6 feet (or 2 meters); the transmission of lice, on the other hand, requires much closer contact so it would have a much smaller “radius of transmission”.
  3. alter the “infection visibility”. The interaction of healthy people with asymptomatic carriers can be modeled through this feature. When infection is less visible, healthy people will be more comfortable in the presence of patients and may get closer than they would with perfect information.

After all parameters are properly set, the software will return a stochastically evolving animation of colored dots (i.e. the passengers), reflecting how disease may spread from a finite number of initial patients to other healthy passengers. In the long term, I believe this project has a great capacity for added functionality. Two moonshot ideas I had were to extend the study of the spread of disease to common surfaces (i.e. benches) and to allow the user to create their own two-dimensional Unity models of different public transport stations, like the NJ Transit gates at Penn Station or the various platforms of the Paris Metro, to study the impact of urban design and architecture on infection control. To this end, Prof. Kapadia’s work with the Intelligent Visual Interfaces Lab on modeling human crowd behavior towards enabling better urban design would be of great help in this project.

Assessing COVID-19 Outcomes and Mortality Using Deep Learning Methods (David Natanov)

One of the greatest challenges of the COVID-19 pandemic has been the difficulty in proper resource allocation when critical units are stretched to maximum capacity. Identifying which patients are most at risk for death over a short period of time is critical to reducing COVID-19 mortality and ensuring that scarce supplies such as ventilators are properly allocated. As case numbers and the number of promising yet scarce therapies such as monoclonal antibodies continue to rise, the problem of pandemic-related resource allocation become even more poignant. Using open source deep learning tools designed for medical imaging, genomics, and biomarker identification, I propose creating a tool that will enable physicians to predict clinical outcomes for patients based upon different scans they may have received, their genetic predisposition to the illness, and the levels of various biomarkers. Already, several papers have found various factors to be predictive of COVID-19 outcomes, including levels of “ground glass opacities,” blood type, several SNPs, and various preexisting health conditions. Combining and tracking these various clinically predictive risk factors may prove to be a useful risk assessment for COVID-19, enabling physicians to make better decisions for their patient’s health and isolate more serious cases before the patient is at serious risk of death. A deep learning model may prove to be far more accurate and quick at assessing patient risk for COVID-19 than any individual physician’s judgement by drawing predictive inferences from a variety of complex data types. Lastly, understanding the complex risk factors that govern COVID-19 may also provide druggable targets and provide avenues for mechanistic research into the virus.

An Optimal Pandemic Treatment and Vaccination Intervention Model for Controlling COVID-19 (Eunseok Kim)

Limiting the adverse outcomes of COVID-19 has become a global priority. While biological-science-driven research efforts are advancing towards discovering prevention and treatment options, there remain significant challenges in their efficacious deployment. This research aims to find the most efficacious portfolio of treatment and vaccination strategies that can reduce the societal harm caused by the pandemic. In this research, we develop the Stochastic Dynamic Pandemic Intervention Optimization Model (SDPIOM), which minimizes the COVID-19 societal/intervention costs by finding the optimal deployment of vaccines and treatment medication. SDPIOM can advise state and federal health agencies on the most impactful way to allocate these scarce treatment/prevention resources. While SDPIOM is being developed for the current COVID-19 pandemic, its applicability can be to a many other pre-/post- infection intervention problems relating to other diseases.

With more than 150,000 known cases of COVID-19 infection in the US, there is a growing realization among policymakers that these numbers will continue to increase for the rest of this year. Controlling the pandemic will depend, critically, on the timely and effective deployment of vaccines/treatments as pre/post-infection intervention. Under the current uncertain efficacy and timeline of promising vaccines/treatments, planning the right mix of vaccines/treatments is a crucial problem that needs urgent attention.

COVID-19 pandemic has several unique characteristics that make the problem especially difficult. Most of the challenges discussed in the literature have focused on biological research challenges. However, the production aspects of the vaccine supply chain under such circumstances is often overlooked. Also, epidemiological characteristics of COVID-19 – asymptomatic infections and re-infections – suggest that a customized epidemic model should be considered to estimate the impact of intervention strategies accurately. Existing intervention studies have adapted classical epidemic models such as the SIR model (and its variants) without considering the idiosyncratic attributes of COVID-19.

In this study, the proposed SDPIOM leverages a tailored compartmental epidemic model. The epidemic model distinguishes susceptible and infectious into subpopulations according to the known characteristics of COVID19’s spread. By interlinking the vaccines/treatment medications directly into the epidemic model, SDPIOM quantifies the impact of these interventions. Incorporating delayed and capacitated intervention availability, which reflects the current pandemic circumstance, will help accurately model the reality, thereby leading to the most appropriate actions for decision-makers. Thus, this model will provide actionable insights to the public healthcare decision-maker by investigating policy-related questions on achieving herd immunity and the minimum efficacy required for the vaccine. For multiple disease progression scenarios provided by the U.S. Centers for Disease Control and Prevention(CDC), SDPIOM will determine the optimal intervention strategies incorporating uncertainties stemming from the pandemic and the availability of vaccines/treatment. Following the validation of the SDPIOM and its parameters, the model will be shared with the federal and state health agencies.

Targeted Interventions Reduce the Spreading of COVID-19 based on Human Mobility (Haotian Wang)

Major interventions have been introduced worldwide to slow down the spread of the SARS-CoV-2 virus. Large scale lockdown of human movements are effective in reducing the spread, but they come at a cost of significantly limited societal functions. We show that natural human movements are statistically diverse, and the spread of the disease is significantly influenced by a small group of active individuals and gathering venues. We find that interventions focused on these most mobile individuals and popular venues reduce both the peak infection rate and the total infected population while retaining high levels of social activity. These trends are seen consistently in simulations with real human mobility data of different scales, resolutions, and modalities from multiple cities across the world. The observation implies that compared to broad sweeping interventions, more targeted strategies based on the network effects in human mobility provide a better balance between pandemic control and normal social activities.

Implementation of nation-wide social support platforms to improve COVID-19 patient outcomes (Kaleigh Hinton)

Research Question: Would creation of a centralized platform to create social support networks for people going through traumatic experiences related to the COVID-19 pandemic improve overall patient outcomes?

There is no doubt that the COVID-19 pandemic has caused increased levels of loneliness and feelings of isolation in communities throughout the US, and things are only getting worse (Twenge and Joiner, 2018). Americans need an improved infrastructure of social support if they are to make it through the COVID-19 pandemic, and inevitable future pandemics with similar devastating outcomes. Development of a centralized social support platform, which shares experiences of survivors, as well as families impacted by COVID-19 is critical if we are to make it through present and future pandemics. Such a platform would allow a large network of people impacted by the virus with differing severity to share their experiences with treatment and healthcare outcomes. It would enable healthcare workers nation-wide to view a platform that provides honest patient insights on different types of treatment. It would be a voluntary open-collaborative platform/forum (edited and maintained by healthcare media specialists to verify identities and manage content) for people to share their experiences with the virus and resulting treatments that they may have to undergo throughout the entirety of their lives. Initially, would be catered to the COVID-19 pandemic, with the goal of evolving the platform to focus on providing social support for as many infectious/chronic illnesses in the US as possible.

Possible Barriers:
Potential barriers to address this research question could include: privacy issues, lack of participation of those impacted by COVID-19, lack of media coverage that would stifle growth of the platform I plan to address this issue by proactively recruiting participants using resources and networks that the Rutgers School of Public Health has to offer. In addition, I would leverage social media platforms of research associates as well as the network of public health professionals that the School of Public Health has at its disposal. As a current student, I would ask my classmates to promote the platform and advocate for the platform’s usage across Rutgers’ entire student body.

Specifics related to the platform:
• Including a connect page where current patients can chat with/contact past patients to create a network of social support for those dealing with the same illness
• A news page where healthcare workers, patients, and family members of those with a specific illness can share their stories and others can read about them
• A stats page where all different statistics are available, including mapping the impact of COVID by preexisting condition and age (those hospitalized at each age with different pre-existing conditions), and mortality mapped by preexisting condition.
o In the case of type 2 diabetes, obesity, heart disease, etc. (diseases that people can make lifestyle choices to prevent), include links to helpful information/counseling to reduce the preexisting condition.
o Include interactive graphs and maps that can show people their risk for contracting severe cases of the disease, and areas in which cases are rising in number
• A page that includes all of the latest legislation around the pandemic/virus, as well as short summaries of each piece of legislature that are quick and easy to understand by the general public.
• Will provide social support even for those that are not able to leave their homes/nursing homes

Twenge, Jean M., and Thomas E. Joiner. “U.S. Census Bureau‐Assessed Prevalence of Anxiety and Depressive Symptoms in 2019 and during the 2020 COVID‐19 Pandemic.” Wiley Online Library, John Wiley & Sons, Ltd, 15 July 2020.

Bolstering Supply Chain Networks via AI (Jonathan Chan, Aditya Garg, Myera Mian, and Julia Nowak)

The ongoing COVID-19 pandemic has exposed many problems in the supply chains across the world, perhaps none more so than that of the medical world. From when the pandemic first broke out in China, the news was full of information of supply chain shortages around the world for supplies such as protective personal equipment (PPE). In South Korea we saw shortages of masks for both citizens as well as for medical professionals. Now with COVID spread around the world, we have seen these same issues arise in many nations, including shortages of PPE, but also for things like ventilators and other medical equipment. Current defaults in supply chains can be attributed to lack of inventory visibility and a sharable data structure. As countries like that of Canada have shown, not having transparency and shareability of data can have devastating impacts on the healthcare industry. Canada has been constantly dealing with shortages of PPE, due to aspects such as PPE expiration, lack of tracking, loss of suppliers, and even competition between its own provinces for the few sources of PPE there are. Though shortages and loss of PPE may be due to a simple lack of structure and data availability, the effects are much more dire leading to rising deaths and sickness. The US has had a similar story, except that it is in terms of states rather than provinces.

A comprehensive data infrastructure as well as a federal reserve can help mitigate many of the aforementioned issues. By using both blockchain and cloud networks through government contracted services like Amazon Web Services, Azure, or Google Cloud Platform, supply chains can securely update their inventories while notifying government agencies of available supplies. Blockchain technology allows supply chains to be made more transparent, accessible, and secure, while also providing more information and depth than other technologies offer. Blockchain allows for ledgers to be updated in real time, and with all these ledgers being interconnected, supply chains suddenly offer much more information than before. If a government orders PPE, they can track each individual order much more effectively through blockchain technology. They will know which orders are about to expire, where they are, and how much they need to reorder based on how much has been used. This can prevent waste of unused supplies, shortages, and loss. While competition between states or provinces can be mitigated with a data infrastructure, use of a federal reserve can bolster this solution further. A federal reserve allows PPE to be distributed to provinces or states that are in need of them immediately while maintaining the natural supply chain. This takes away state and provincial governments’ fear of not having enough supplies resulting in over-ordering and shortages. Additionally, a reserve benefits healthcare organizations too as supplies soon-to-expire can be bought at cost. The uses of both a data infrastructure and federal reserve can prevent many of the issues that have weakened supply chains during COVID-19. However, humans are not free from error. This is where emerging innovations like Artificial Intelligence (AI) can improve processes even further.

The use of AI ensures that the blockchain system is as efficient as possible. Through keeping track of usage numbers, expiration dates, demand, as well as other statistics, AI can accurately predict how much PPE needs to be reordered. This ensures that even the smallest changes are kept in mind, as well as accounting for things such as losses from accidents, expiration, or misplaced inventory. This is crucial to ensure that people on the front lines like nurses, doctors, EMTs, etc all have access to PPE to keep them safe during a time like this. They could be interacting with hundreds of people who have tested positive for COVID, and having PPE is a way to help protect themselves from the risks they put themselves in for us. AI can also quickly pick up on new trends and growing models of future demand. AI could have analyzed how countries around the world were suffering from COVID, and bolstered order numbers ahead of time. AI can also be used in terms of distribution of equipment in the supply chain to areas which need it most based on these same statistics. If it knows that a state, province, etc historically has a higher need of equipment during a certain time, or that it is using equipment at a faster rate, it can place more orders and help distribute existing resources to the areas which need it most. With the implementation of a data structure and federal reserve both optimized by AI, many countries can see a drastic improvement in the management of their supply chains. Humanity needs every advantage it can get and with a solution like this, we will be better equipped to fight this pandemic and be better prepared for possible ones in the future.

Resolving the Issue of Cold Storage Chains for Vaccine Delivery (Kritika Singh)

The COVID19 pandemic has been the biggest crisis in recent times. In the US alone, we are close to a staggering 300,000 deaths. What is most frightening is that we do not have any real solution to contain the virus, leaving high hopes built around vaccinations. Billions of dollars have been invested through Operation WarSpeed to pharma companies that have the capabilities of testing and producing a vaccine in record time. With all this support, Pfizer, Moderna, and a few companies have come up or are about to come with a vaccine. Both Pfizer and Moderna vaccines have been shown to be 95 percent effective, which is truly remarkable. We are now out of the woods and there is light at the end of the tunnel.

While the development of these novel vaccines is promising, many new challenges are to arise. For example, Pfizer’s vaccine must be stored at -80 degrees Celsius and Moderna vaccine needs -20 degree Celsius. This will require a cold chain storage to be maintained after the vaccine is produced and finally delivered to the public. This will not only be almost impossible in many developing countries, it will also be difficult in many of the remote areas and interiors of even the most developed nations.

My solution would be to build drones with cold chain facilities that can be used to deliver the vaccines. There are several advantages in taking this approach. Drones are small and compact and can travel at almost 70 mph at low altitudes. They do not need an infrastructure of roads and rail lines. They can run on self-powered engines that do not need a lot of power. If we engineer a drone to have in-built cold storage facilities, it may help resolve the issue of getting the vaccines to areas that do not have the infrastructure to maintain cold storage chains.

Utilizing Demand Sensing to Aide Food Pantry Coalitions in Efficient Inventory Management (Lily Chang)

The economic fallout of the COVID-19 pandemic caused a rapid increase in Americans suffering from food insecurity, with widespread media coverage showcasing hundreds of cars lined up at food banks and pantries across the country. The pandemic highlighted food insecurity as a rising issue in the U.S., and the long lines served to reveal the severe deficiencies in the Emergency Food System, especially in terms of supply chain, food aggregation, and food distribution. A problem that is often overlooked is the lack of interconnectivity between food pantries operating as a coalition, leading to avoidable deficiencies in inventories. Thus, I propose a solution to this by implementing demand sensing into a centralized database to forecast future pantry and pantry coalition needs.

For background, a food pantry coalition is a group of food pantries located in close proximity that organize together to share resources and aid each other to operate more effectively and generate a wider impact. However, many food pantries in a coalition often do not share inventory or clientele information; they often have separate data management systems isolated from one another. Since they cannot access each other’s data, this impedes their abilities to maintain stable and reliable inventories, which may lead to underutilization of not only the pantries themselves but the Emergency Food System as a whole and exacerbate food insecurity. Thus, these coalitions struggle to collaborate effectively: if one pantry has an excess of a particular food product while another has a shortage, there is no dependable resource with which to utilize to notify, communicate, ship, and receive the food product among those two pantries. Typically, food pantries rely on word of mouth, calling/texting, and/or social media to ask other pantries in the coalition for aid. While there do exist software applications that are specifically geared towards pantries with common features such as client tracking and inventory management, none of them are geared towards coalitions specifically. Coalitions require a different and more complex set of features and capabilities as it needs to cater to the needs of multiple pantries in addition to ensuring a functional interchange of information (otherwise, there is no point of a centralized database). Moreover, no current software integrates demand sensing, which has unprecedented potential to improve the operation of not only pantry coalitions but the Emergency Food System as a whole during unforeseen circumstances like a pandemic. Demand sensing can be particularly useful for food pantries as it aims to make the most accurate predictions of demand for the near future based on short-term demand data: whether it be hours, days, or months. This differs from traditional models that focus on yearly, or macro, predictions that are not as useful for pantries as they are constantly experiencing shifts in food supply and will need to rapidly alter operations in a pandemic. Taking this one step further, demand sensing is more about making the best short-term decision based on recent events, vital during a pandemic.

To startup the software system, all pantries in the coalition will have a responsibility to input the number of households served per month with an approximation of what quantity and types of food are needed each month to serve those households. This can always be revised according to new information, increase or decrease in clientele, and changes in food needed. After receiving food, the pantries will input the type (perishable/non-perishable, vegetable, meat, etc.), quantity, and expiration dates (if applicable). The system will automatically adjust the inventory based on requirements and demand of other pantries, subsequently calculating excess inventory, deficiencies, and need for the pantry in question and pantries in the coalition. Then, using location and demand, will alert both pantries (one with excess and one with need). It is then the pantries’ responsibility to collaborate and arrange for shipment and transportation. Since the system will make use of demand sensing, it will continuously learn what the coalition’s needs are, leading to increasingly more accurate projections of future demand. Essentially, the software will generate calculations based on a time-scale of growth.

Resources required include a team of data scientists, programmers, and data engineers. Data consolidation among food pantries is key, a cloud-based system may be ideal. An interesting option would be to employ Rutgers computer science and IT students partnered with faculty to work with these coalitions to create something similar to my proposal as a research project or internship.

In summary, while this is only a very brief overview of the proposed system, it provides a meaningful substructure with which to illustrate a potential solution to the lack of interconnectivity and communication between pantries in a coalition. Ideally, to prepare for a pandemic or any other circumstance, this will be implemented and in use for as long as possible so that the system can use extensive historical data to produce reliable projections. With a pandemic, there is a need for a faster response rate and increased agility and collaboration in a coalition for supply and demand. Using a system such as this will promote maximum usage of food, delivery to the people most in need, and a decrease in food waste.

Applying Machine Learning to Demographic and Resource Data to Determine Optimal Vaccine Distribution Routes (Sanjiv Prasad, Mervin James)

Vaccine distribution is a challenge many scientists and companies face each year. The current pandemic has established the importance of an efficient rollout of vaccines, for a successful one can prevent many deaths. Factors that affect vaccine distribution include the availability of cold storage, location of transportation routes, available supply, costs, and need. I propose the use of data science to determine optimal routes of vaccine distribution that take into account known prevalences of risk factors in certain populations, such as diabetes, lung disease, and cancer, the location of cold storage facilities, and established flight patterns. Machine learning and predictive analytics can be used to account for projected supply and demand. Data science is already widely used in supply chain management. The purpose of this study would be to model vaccine distribution in the United States as such a supply chain. If successful, various organizations can use such predictive models to establish efficient and cost effective routes to bring vaccines to those who need it the most.

Scarlet Tutors: Using AI to Empower Students of the Next Generation During COVID-19 and Future Pandemics (Meera Krishnan)

The high unemployment rates associated with the COVID-19 pandemic have resulted in further stress on low-income communities. Online schooling has had a huge impact on these communities, as it decreases the opportunity for interactive learning, foreclosing on a routine environment where students focus on their work and listen attentively in class without worries of service or homelife. With most of life occurring now at home, there is more responsibility on the student to learn the material and more pressure for parents to help in their education than would be necessary with in-person instruction. These families often cannot afford to pay for the high cost of tutoring, and many are tasked with trying to find a job during this pandemic. As a result, few have the time to tutor their child like the school would otherwise have done. Currently, the students who would benefit the most from interactive live tutoring often cannot afford it, and this socioeconomic problem is further exacerbated during a pandemic such as COVID-19.

Plausible Solution
Many students are interested in tutoring as well as gaining volunteer hours at Rutgers University. Rutgers University has the resources to launch a virtual tutoring program to bring much needed academic aid to students not just in New Brunswick, but across the state. If successful, the next step is to design an application that would allow university students nationwide to register themselves as tutors with an account, and students who need academic assistance can search for a subject on the site, find a tutor who is also online, and set up a virtual tutoring session.

One possible barrier is allowing tutors to virtually meet with the students in a way where they can effectively teach virtually. Since Rutgers has an agreement with Zoom, we can start by limiting the tutors to Rutgers students who can register with netIDs (thereby being able to create a zoom meeting room). We can also connect this to the tutors’ hourly log, once they upload their recorded tutoring sessions their volunteer hours will be counted for. This is just one solution to this barrier, and I am open to other ideas.

While I understand the importance of health-related solutions, this pandemic has greatly jeopardized the education of the next generation by forcing schools and students to adapt to online infrastructure and stressing the preexisting inequities that have chronically plagued disadvantaged populations. I want to work with Rutgers to provide a solution that will make sure the education of our future workforce is not compromised regardless of any measures we need for safety during this pandemic.

Resources and Partnerships
I would like to collaborate with the AI and Pandemic Initiative to set up this application, which would allow for Rutgers students to register with their netID and become tutors. We can test the program at the New Brunswick, Newark, and Camden campuses. Tutors would also have the means to track their hours on the application. The application would be able to match a student looking to be taught in a certain subject with a tutor who is registered to help in that same topic.

The Guru: A Personalized Education Platform (Namarata Battula)

The COVID-19 Pandemic has highlighted serious concerns regarding the education system. Students are overwhelmed by the number of assignments and assessments they are expected to complete, while simultaneously dealing with the uncertainty brought on by the pandemic. Their anxieties lead to burnout, a lack of interest in schoolwork and learning, and high levels of cheating and plagiarism.

I propose to use Artificial Intelligence to create a personalized education platform. Through this platform, students can progress through a curriculum (designed by professors) at their own speed, while also receiving feedback on their progress. AI can be used to create this flexible and engaging learning system that will adjust to each student’s unique strengths, weaknesses, and interests. This program will recreate the beautiful relationship between a guru and a student, a valuable one-on-one teacher-student relationship, that is often difficult to create and maintain in a course with 100+ students over Zoom.

Janice Gobert is a professor in the Graduate School of Education (GSE) at Rutgers University. She has done “cutting edge work [focused] on personalized learning of science and assessment to replace summative tests.” The first step to approaching a feat as large as reforming education might be to replace traditional testing methods. Gobert is a valuable resource here at Rutgers, whose work in creating performance assessment technology will be useful to kickstart this AI proposal. I hope to reach out to her and learn more about the specifics of her research. This proposal is important to me as a student because I have seen how the education system has been affecting the mental health of students during this difficult time and I hope that this technology can reinstall the curiosity and excitement that we all started school with back when we were children.

Covid Warrior: Where Technology Meets Education (Sanjana Pendharkar and Natalie Chen)

When COVID-19 hit our towns and cities, news channels focused on the damages to our economy, while academic publications were geared towards understanding the biology of the virus. However, there was little to no exploration of the pandemic’s impacts on some of the most promising members of our society: elementary school children. Taken away from their familiar class time ritual of show-and-tell, and having to cancel on informative field trips to local museums and galleries, children were left feeling lost, ignored and marginalized.

During the first few months of the pandemic, I noticed a change in my little cousin’s behavior and general excitement for school. Whereas previously, he was excited to map out recess with his friends, in our recent conversations, my little cousin expressed a general sense of confusion, as he didn’t exactly understand the withstanding implications of the pandemic. He didn’t want to be a part of Zoom classes, and would often complain about the lack of excitement in lessons and interaction with his classmates. When young children lose interest in their education, they risk not developing refined understanding of the world around them, and pursuing ideas that change our societies. As informative as teachers would like their Zoom classes to be, online instruction is often a form of passive learning, as students are not actively applying their skills or furthering their progress through collaborative problem solving. With the 2019-2020 school year being cut by at least one third relative to its normal length, there was a loss of at least 0.1 standard deviations in retained learning across the board, deemed to be even larger in elementary school children, with about 60-80% of elementary school students (even larger with respect to STEM) experiencing a performance gap. Adding onto that, the budget cuts taken by several schools due to a 15% reduction in state education funding lead to a loss of more than 300,000 teaching positions, implying less one on one interaction with teachers and less educational resources available for students. Clearly, the impairments in education faced by children are a serious problem that needs to be addressed, as they have the power to be leaders in our society.

In order to combat the problem of a loss in educational retainment, our team is proposing Covid Warrior, an adventure game in which students can explore different worlds (Fantastic Forest, Dungeon of Doom and Castle Capitol) while answering curriculum-based questions in math and science to defeat biological monsters (Covid Dragons and Virus Zombies) and advance within levels. What makes our game different from all of the other educational games out there is the prominence of public health education within our game (the collection of masks and hand sanitizers to help the classroom community in bonus levels), along with detailed, step-by-step instructions on how to solve a problem if it was missed the first time around by a student. It’s important to recognize that going to school is not just about learning the basics of various educational subjects, rather it is about personal development. After school activities, such as playing our game, support the mental and emotional well-being of children, and instill a sense of self-confidence within them. By introducing our game as an addition to instructional time, we hope to facilitate student interest in course material, as extending instruction by an hour increases reading scores by 0.05 standard deviations in elementary schools, and teach them the skills of patience and collaboration. This game involves an application of cognitive science tools and
technologies, including the study of epistemics, which explores the transmission of knowledge from a sociological and psychological point of view. We are using common aspects of cognitive science to initiate UI/UX research and testing within the student population, utilizing terms such as the scientific method, simulation, data visualization and modeling. Currently, we are working with Durham School District in North Carolina, to pilot our game through an inter-institutional independent study. At Rutgers, we are working with Dr. Aanjaneya and Dr. Kapadia to understand game design and game development. Through this experiment, we will be drawing conclusions from the data collected using data analysis and statistical techniques to see how we can enrich student experience and deduce if our product was successful. In future models, AI will be implemented to help the student practice their weaker areas, based on the questions they have answered correctly. We plan to have a friendly AI chatbot that students can interact with through the duration of their studying within the game, which would function as an accountability buddy our users can be motivated by.

In introducing this solution, we will reach some barriers in development and implementation. One barrier is sparking continued interest, especially in older students. To combat this challenge, we plan to code different difficulty levels and incorporate engaging game play, structure, as well as vivid and playful character graphics and aesthetics. We also don’t have prior experience in game development, game design, and education content creation. We plan to self learn Unity / C#, refer to existing curriculum worksheets, and speak to professionals for guidance and advice. Finally, we don’t have a funding platform for gaining traction in different schools and for game development. We hope to move forward in this competition to provide the financial means to cover better softwares and other classroom equipment, so we can advance learning during this crisis. The financial support and guidance can help us, help others.

Using Artificial Intelligence to Improve Online Engagement for Young Students (Steven Liang and Justin Lee)

The COVID-19 pandemic has presented significant challenges for education. Young students are likely to be the most affected because primary education has a strong influence on their development and social behaviors throughout their lives. Nevertheless, the proliferation of technology like artificial intelligence could alleviate the impacts of the pandemic by providing a personalized educational experience for struggling students. Therefore, we propose the development of a rational, utility-based agent that could utilize classification and regression models to best model each student’s performance and understanding. This AI system would utilize an expansive knowledge base of educational-related materials on various topics to help address shortfalls in each student’s learning. For example, our model would gather data from raw user inputs such as mouse and eye tracking, facial expressions, student free response answers, and gather insights that will allow teachers to pinpoint areas that students are struggling with and make the appropriate adjustments for tailored assistance. Furthermore, research from the Indonesia University of Education identified that students who learn through a problem based learning model—utilization of educational material tailored to real life problems and questions—had improved literacy when compared to a traditional direct instruction model, characterized by teacher explanations of the curriculum followed by practice guided questions. With the combination of AI classifying student struggles, our solution will additionally involve AI regression that would integrate problem based learning more by generating tailored practice problems that will help students gain better understanding based on their identified skill level. Overall, the use of AI in online education will serve to better assist students in understanding concepts by identifying their needs and offering specific assistance and guidance.

Keywords: primary education, coronavirus, artificial intelligence, remote learning