Pandemonium: Risk Analytics for COVID-19

COVID-19 has decimated economies, destroyed jobs and caused half a million deaths and counting. For individuals, companies and local governments, the lack of data-driven risk assessment tools has continued to hamper efforts to return to normal. 

Quantum Risk Analytics, Inc. is developing a robust risk assessment tool for predicting real-world infection and mortality based on age, health, the location’s risk and individuals’ activity and behavior. By aggregating multi-dimensional data and using highly advanced techniques, we will give individuals and organizations a complete risk assessment tool and empower them to make informed decisions about their activities. 

Our tools will be widely available via our freemium business model; corporate users (and advance users) will pay for advanced features. We rely on partnerships to obtain relevant data and will prioritize our solution on high risk use-cases.  Our Pandemonium model will also prepare humanity for future pandemics. 

https://pandemonium.dev/pitch-MIT_Solve.pdf

COVID-19 has been extremely disruptive for everyone globally. It has resulted in numerous deaths, healthcare infrastructure being over-extended, high joblessness, financial losses and most importantly, uncertainty about restarting the life and economy. While many COVID-19 models exist, these models do not have the granularity to assess risks at the micro-level of individual or localized groups. 

2020 GDP is projected to decline from around 3.0>#/span### to 2.4>#/span###.  40 to 60 million people are expected to enter into extreme poverty (defined as living on less than $1.90 a day) due to the COVID-19 pandemic.

The aviation industry is facing a deep crisis, with 90% of the global fleet grounded. Meanwhile, global commodity prices have seen their largest fall on record, falling 20.4% in March 2020. Global trade for the second quarter of 2020 is now forecast to drop by a precipitous 27>#/span### year-on-year. Tourism is forecast to fall between 58% to 78% this year.

In terms of social costs, the education of 1.6 billion learners has been disrupted; that is 9 out of every 10 students in the world.

Most concerning of all, the impact of COVID-19 is expected to continue for years to come post COVID-19.

We are working on providing everyone a tool to assess their individual and families’ risks, based on information they choose to provide like their recent history & future plans, their location, demographic profile and pre-existing medical conditions. 

Our web-based portal will enable them to upload all these data. This, together with our and other modelling data, will allow us to deliver a personalized risk assessment, and enable users to assess the impact of their choices on infection risk, as well as those of their immediate family and roommates. And they will also receive an assessment of their societal risks. All personal data is optional.

The corporate version will include additional modeling of their operating environment to assess their employees and customers’ risks, including custom modeling of unique situations.  We enable organizers of large-scale events to assess risks from crowd aggregation and dispersal.  At the city/county level, we enable public health officials to assess macro risks related to various measures like limiting gatherings and mandatory masking in public

We intend to first offer our solutions first in the US, but will quickly roll it out internationally.  The international versions will factor in specific cultural and environmental differences in each individual market.

Our solution serves the public to assess their risk. This solution may be used by individuals, companies, schools, event organizers, government etc. The solution will empower the individuals to make informed decisions by using their unique multiple factors and situations into our risk analytics solution.

Quantum Risk Analytics, Inc. is directly addressing several UN Sustainable Development goals (3, 4, 8, 10, 11). Attainment of goals has been severely curtailed and even dialed backwards by COVID-19, hence we are directly targeting the COVID-19 root cause, empowering individuals and organizations to make informed decisions and enabling them to resume working towards those goals. As a stop-gap mitigation measure before a vaccine is found, our solution can significantly reduce the second order impact of COVID-19.

COVID-19 has disproportionately affected the disadvantaged groups.  Our solutions priorities helping these disadvantaged groups, which is the core focus of the UN goals.

Our solution brings diverse components together in an innovative fashion resulting in greater capability of the solution to perform even more complex analyses holistically. Additionally, our solution strives to bring the most advanced tools such as quantum machine learning to the public in an easy to use manner and free of cost which is innovative.

Our biggest differentiators are granularity and personalization.  We intend to take into consideration individual risk factors, including (but not limited to) medical history, detailed travel history/plans, and potentially even genetic data.  As far as we know, none of the existing models have all of these.

We use a combination of technologies as the core of our solution – Bayesian inference, probabilistic programming language (Pyro), quantum and other machine learning, and computational fluid dynamics.

More about the technology and how we will use it is discussed in https://pandemonium.dev/proposal-draft.pdf

The evidence of our solution working will be assessed from the accuracy it achieves. Accuracy depends on the risk model and data fed into the model. We will pilot our solution with partner organizations to test, refine and confirm the risk model. In parallel, we will continue to gather data to feed into the model to continue improving the accuracy of the model’s prediction. The technologies that we are using have been utilized successfully in various applications. Some have already been used successfully in COVID19 modeling, except in less comprehensive ways. Here are two examples of machine learning applied to COVID19:

Youyang Gu (an MIT alum) has developed a relatively simple model (https://covid19-projections.co...) that employs machine learning (ML) techniques and has done better than many more complex models (see the analysis he provides on site above). It demonstrates the power of ML applied properly. He is relying on death data which the most reliable. We will be using death & intubation data to ground our modeling due to data reliability, (see discussion in https://pandemonium.dev/propos...), but will also utilize many other data in order to further improve the modeling.

The MIT DELPHI project has an infection calculator & mortality calculator:

https://www.covidanalytics.io/...

https://www.covidanalytics.io/...

These are trained on data using machine learning. These are independent of their macro model and are specifically for the clinical setting, whereas we are addressing the much broader domain of outside the clinical setting and integrating with the macro model to do that

Our solution is built on the concept that individuals can make better decisions for themselves and others if they are provided with all the necessary information. Therefore, our solution strives to empower the individuals by allowing them to input and analyze their unique situations into our solution anonymously to assess the health risk associated with them and make the best decision for themselves and others around them.

We currently have no users as the solution is in development phase.

Conservatively, in one year, we intend to serve 3% of the US population and 100 companies of average 1000 employees each and have a significant user-base outside the US.  In 5 years, we hope to serve 10% of the US population and 1000 companies and have very substantial global usage.  By that time the pandemic is expected to be over, but our tool can still be used for other infectious diseases like the seasonal influenza.

Our immediate goal is to develop our model (and incorporate relevant data) to the extent that we can help anyone who wants such a tool (particularly among urban, low income residents).  Within the next five years, we intend to work withepidemiologists, infectious disease experts and other scientists to continue incorporating the latest and advanced knowledge, technologies and models at that time (e.g. for specific use cases in specialized environments), and leverage the collective expertise of the medical and scientific community to make this the most useful model for risk analytics for future pandemics.

We anticipate our biggest barrier to be availability and access to accurate data of root causes and their impact on infection spread and mortality.  The fragmented way data is collected in different global markets makes it difficult for us to normalize the data for global use while privacy concerns would hamper data to be shared.  More importantly, organizations who collect and have access to relevant data are unlikely to share initially until we build a network and reach critical mass of data coverage.

Secondly, once we have the required data and our refined model, organizations still need to be persuaded to use data-driven insights to make their risk-mitigation decisions.  There could be other (e.g. political & business) considerations that may prevent decisions around minimizing infection spread and mortality reduction based purely on scientific evidence and irrefutable data. 

In next five years, the major barrier we anticipate is the scale-up of the solution to global locations, incorporating the diversity of different populations and the large user base.

Our ability to demonstrate effectiveness in the first few use cases will be fundamental to solving both the above-mentioned barriers.  For example, if we can successfully help a factory or a small city minimize infection using our tool, other organizations/cities will be persuaded to use our tool and provide us with more data to refine our tool.

Additionally, MIT network and MIT Solve community will be leveraged to help us obtain high quality data, refine our model and demonstrate effectiveness in initial use cases.

Being a charitable non-profit corporation enable us access proprietary data which are available to non-profits / research institutions.  We also plan to crowdsource highly localized data which is not available from centralized source.  We will use advanced techniques like machine learning both to extract meaningful data from social media and digital photos & videos, as well as to ascertain elusive parameters that are hard to measure by optimizing the fit of the model to the most reliable data. With our corporation, we plan to engage the local and state officials to obtain data and adopt our solution. 

Our team is geographically diverse that works remotely to source the appropriate resources with the required skills.

6 people total working on the solution, all volunteers, including 2 in Europe, 3 in US, 1 in Asia.

1 working full-time now on it, another nearly half-time.

Various advisory contributors.

As a global solution team, we bring all relevant skills ranging from management to information technology to healthcare. Specifically, we bring skills and experience such as in life science industry, strategy and innovation, both business management and not-for-profit corporate management, GIS, data science, software development, artificial intelligence, including quantum machine learning, combinatorial optimization, as well as numerical modeling and fluid mechanics. ~50% of our team are MIT alumni and the team is led by a core team of MIT alumni; they bring strong technical experience and background to the team. Our team has also adopted a strategic approach to build upon the existing solutions to lead to complex analyses to revolutionize the management of COVID19.

Our team is exploring partnership with Research Science Institute for scientific networking with the leading scientists in fields such as big data analytics and infectious disease, and we are participating in their mentorship program, offering student internships through them this summer. A faculty at University of Illinois (also an MIT alumnus) is acting as an advisor of our team.

In the short-term, we plan to focus on the free solution that is available through a web-portal to the general public and small-businesses. We are in the process of seeking funding through grants, donations, crowdsourcing and MIT Solve to fund our effort to develop and rollout the free solution. In the long-term, we will offer a customized, paid solution to the large, for-profit corporations. This customized, paid solution will allow us to sustain the refinement, availability and rollout of the free solution to the general public.

Initially, we will rely on volunteers, grants and crowdfunding sources to get us to a workable prototype.  Once we are able to offer the solution via our freemium model to the general public, we expect at least a small percentage to sign up as our premium paid users.  In parallel, we will approach organizations whose core businesses are most severely impacted and demonstrate that our tool can enable them to resume operations.  Once we achieve our critical mass of corporate users, that will be our primary revenue source.

On the cost side, we intend to be lean operationally, and scale our costs proportionately to revenue.  This way, we will always be able to maintain a financially viable business model.

Seed funding, brand recognition, feedback on our solution, mentorship by and connection with MIT faculty to further develop and refine our solution, more effective outreach to our customers and organizations that own data through MIT Solve.

First priority for partnership is to obtain data to feed our model.  Data can be very costly if bought on a commercial basis, and the huge data requirement of our model necessitates that we obtain them via partnerships, e.g. in exchange for data, we can offer the organization a discounted or free subscription to our customized, premium solution for the first year.  We will offer our partners more value in return for their data contributions.

Our second partnership goal is to identify and pilot our tool in some of the most impacted use cases, e.g. airlines, cruise ships, schools/universities, factories.  This will help us to rapidly refine our model and make it ready for general adoption.

Third, we wish to partner with established research and scientific organizations in order to tap into their scientific expertise and accelerate the development of our model.

First of all, as MIT is grappling with the decision of how to open the Fall 2020 semester in a way that minimizes the risk of infection, we would like to partner with the Institute to apply our tool to help manage the micro-environment within the campus - the re-design of how lectures are conducted, how student living, athletic center, hallways, lab arrangements and how dining facilities need to be reconfigured, which students to bring back on campus for the Fall 2020 semester. This will help student and faculty members return to campus in a safe manner.

We certainly would like to get help from MIT Venture Mentoring to scale our operations.

LATER

We are developing a solution that employs various AI & ML technologies for the benefit of humanity. Our primary goal is to leverage these and other advanced technologies that would normally be out of reach for most people to utilize directly for assessing their risks for COVID-19 with the necessary specificity to allow them to optimize their choices to minimize COVID-19 risks weighed against other priorities. Simply staying home all the time indefinitely is not something that works for most people. So how do people understand what their true risks are in the various real-life scenarios that people face? We have formed a not-for-profit corporation to both develop the technology and make it available for free to help people get those specific answers.

The AI related technologies that will employ include Quantum Machine Learning, Deep Learning and other Machine Learning, probabilistic programming languages (such as Pyro), Bayesian inference, Markov Chain Monte Carlo (MCMC), Motion Tracking, Image Analysis and Classification, etc. Here is a further description on some ways machine learning will be utilized in our solutions:

Machine Learning

Machine learning (ML) techniques are planned to be used in multiple ways. ML offers many potential advantages, including the potential for:

• optimizing the model parameter values based on the observed data

• utilizing large, disperate, additional and updated data sets with relative ease

• finding hidden patterns that may be missed in traditional statistical analyses

• and doing so relatively quickly.

Assessing Mask Usage in Public

A parameter that is likely to have a lot of spacial and temporal variability even when there is consistency in public policy (i.e., a state-wide requirement or CDC recommendation). Deep learning (DL) would be very good at recognizing whether people are wearing masks, and potentially even do well at categorizing the masks between types, such as N-95, surgical, cloth, bandana, other homemade, etc. as well as whether it is being worn properly (e.g. covering the nose and well fitted). Then a large set of publically-posted geotagged and timestamped images could be analyzed using the trained DL algorithm to identify how the usage and types vary by location and time. The geotag would determine whether the photo was taken in a public location and the kind of location, which could then be combined with similar locations in the area to estimate the usage rate. Certain locations do not tend to have many people taking photos normally despite the high number of people, however. (Grocery stores are not the most photogenic spots!) Volunteers could be welcomed to take and upload photos at such locations directly to be analyzed for this purpose. There are other ways such random photos could be analyzed to assess the number of people, their density and how much social distancing is occurring. (Trying to use any publically-posted photo would not be good for this, because there is high chance that such a photo is of a group of people who know each other and came together just for the shot, but aren’t usually that close.)

Estimating Viral Exposure Probability
Distributions

One of the key challenges to doing a detailed risk assessment is not having any direct data of how many active virons (viral particles that are potentially infectious) everyone is exposed to every day. The micro mechanistic sub-model helps, but it is not enough, because we generally don’t know who is infected, what their viral load is or how much virus they are shedding, nor their exact movements. But that does not mean that all hope is lost. We can deal with this problem probabilistically.

The motion for people at a particular place can be analyzed from a video recording, if available, ideally from security cameras since these will avoid sampling bias (at least if they are fixed-frame and distributed to provide full coverage). Motion tracking software is well-established AI that can extract the centroidal location of the moving object (in this case humans) as a spatial point as a function of time. Once we have that we can use machine learning to analyze the human movements relative to each other as well as the venue and as the human density increases and decreases. And we can take a set ML-trained virtual automaton bots that will randomly behave characteristically to the real motions on which they were trained and with the same distribution of behaviors, which would be encoding this very complicated joint probability distribution, and run many Monte Carlo (stochastic) simulations with them, randomly selecting which bots were infectious and by what degree, etc. And then along with the mechanistic micromodel we can analyze the level of exposure to a particular bot on each run. This will result in a probability distribution itself due to the uncertainties involved in that process. Then the combined distribution results from all the runs taken together. That is our best estimate of the viral exposure, based on the lack of knowledge of actual movements as would be the case for any future and most past estimates. But if we happen to know one person’s actual track from the past, we can run the simulations with that locked-in, to improve the estimate.

(Drawn from https://pandemonium.dev/proposal-draft.pdf ---this will periodically updated with more details about how we will utilize these and other technologies.)

Bringing these advanced AI and other technologies together, we will provide the following solutions to humanity to help it deal with this crisis:

1. We are working on providing everyone a means to assess their own risks and their families’ risks, based on as much information they choose to provide about their recent history and future plans, as well as their location and demographic information and medical information (concerning their predisposing prior conditions if any, and whatever COVID-19 testing that they have had). They will be able to specify in our web-based risk analytics portal where they have recently gone (and for convenience we’d plan to enable them use a track recorded on their smart phone if they have one) and where they plan to go and/or are thinking about going in the future. This information with our modelling and other model data, will allow us to deliver a personalized risk assessment, and enable users to explore the impacts of their choices on risk. Information about their living situation and family and/or roommates can also be provided to further estimate and show the propagation of risks (to and from them). And they will also receive an assessment of their societal risks. All personal data is optional; it is up to each to balance how personalized to them they want the assessments to be with their willingness to provide data.

2. We will be able to analyze the risks associated with holding certain events, large and small, and the circumstances under which those events are held and mitigation measures taken (if any), and the projected impact on the spread of COVID-19 not just locally, but to great distances upon dispersal. We plan to have a no-code option for specifying generic small events to model. (Large events are much more complex to model than small ones, and will require a dedicated effort from a member of our team.)

3. We are also developing a no-code platform designed for many small businesses that can be used by businesses and organizations to customize the modeling of their establishment so that they can better understand their operational risks both to their employees and the public and evaluate (potential) mitigations.

4. Larger companies with greater complexity would be able to work with us directly to model their particular situations and needs.

5. Governments would be able to utilize our work to help evaluate policy options that they are considering.

The basic solutions provided to individuals, businesses and organizations will be made available for free. We further expect to make our code open-source with the first release intended for regular use, so that others with the means can further build on it to have the greatest impact for humanity.

We would use the AI for Humanity Prize money, if we are so fortunate to receive the high honor of winning any, primarily to hire more highly-skilled technical staff to

develop our solution more quickly and secondly to scale deployment, so it can help more people sooner.

Quantum Risk Analytics, Inc.'s solution directly addresses at least 5 of UN Sustainable Development Goals (3,4,8,10, 11). Although this is largely a global health crisis, it has first order impact on the economy & jobs, numerous industries are on the brink of collapse, education has been halted, cities have been locked down and across the board, the social-economical divide has been widened, with the poor, minorities being disproportionately affected by the pandemic.

COVID-19 has been extremely disruptive for everyone globally. It has resulted in high number of deaths, healthcare infrastructure failure and unemployment. On a global level, it is estimated that gross domestic product (GDP) over 2020, will drop from around 3.0 pQuantum Risk Analytics, Inc.ercent to 2.4 percent and 40 to 60 million people are expected to be pushed into extreme poverty.

We believe that one the fundamental ways to holistically address this myriad of issues is to empower individuals, companies and cities to get back to their feet, by providing them with the tools that accurately assess their infection and mortality risks in order to make decisions on when and how to return  closer to normal. We truly believe short of finding a vaccine, Quantum Risk Analytics, Inc.’s tool is the best stop-gap measure that will fundamentally enable a reboot of our society and economy.

Our fundamental purpose is to try to improve the chances of protection for the people worldwide by giving them the possibility to calculate the risk of contracting COVID-19 and developing aggressive forms themselves as well as their risk to others, based on their travel plans and various choices as well as pre-existing conditions and other factors. This could leExplain how you are qualified for this prize. How will your team use The People's Prize to advance your solution?ad to an increase in individual safety as well as a decrease the spread of the disease in the community and therefore provide better chance for people to live better by not taking unnecessary risks.

We are proposing an application from the people for the people which will be free of charge and accessible to anyone anywhere.  The communities and demographic groups that are shown to be most impacted by the pandemic stand to also benefit most from Quantum Risk Analytics, Inc.’s tool.  Hence we truly feel that we deserve consideration for The People's Prize.

If we receive the great honor of winning, we intend to use The People’s Prize money in order to pay developers to help us perfect the application more quickly. Since more data equates to more accuracy of the model we would like to invest in capturing best data and in creating the best model.