# AI4PAN Seminar Series
Welcome to the AI for Pandemic Seminars organized by the [AI4PAN group](https://ai4pandemics.org/). Our fortnightly AI4PAN Seminars run via [Zoom 2022](https://uqz.zoom.us/j/86567032193) and new link in [2023 Zoom](https://uqz.zoom.us/j/87348920820) at 10am on Wednesday's (AEST = Brisbane time zone). See also our [YouTube Channel](https://www.youtube.com/channel/UChfQzhyKowwzYFAXh6ziRlg).

**UPCOMING SEMINAR TALKS**

**Wednesday, June 7, 2023** [Scott Greenhalgh](https://www.siena.edu/faculty-and-staff/person/scott-greenhalgh/), Siena
**Title:** On the application and theory of ODE compartmental models in epidemiology
**Abstract:** Many methodologies in disease modeling have proven invaluable in the evaluation of health interventions. Of these methodologies, one of the most fundamental is compartmental modeling. Compartmental models come in many different forms, with one of the most popular amongst disease modelers being the use of systems of ordinary differential equations.
<details>
<summary>Click to expand</summary>

So, to begin my talk, I will illustrate how such models can inform on public health and evolutionary biology issues, using HIV and Hepatitis G co-infection as proof of concept. Next, I will show how the theoretical extension of ODE compartmental models beyond their traditional formulations can provide even greater insights into disease transmission. Specifically, I will show how generalizing model rates can inform on measles elimination and re-emergence times. Furthermore, I will explain how generalizing the concept of “disease quantity” yields simple gonorrhea models with potentially periodic behavior, reduces model complexity, and more accurately reflect the epidemiology of diseases. Finally, I provide a brief summary and overview of current work and future directions.
</details>

**PAST SEMINAR TALKS**

**Wednesday, May 3, 2023** [Kernel Enrique Prieto Moreno] Cancelled (https://www.fciencias.unam.mx/directorio/53588), Facultad de Ciencias, UNAM
**Title:** Forecast of COVID-19 dissemination in Mexico: a Bayesian and Machine Learning approaches
**Abstract:** The COVID-19 pandemic has been widely spread, affected and caused the death of millions of people worldwide, especially in patients with associated comorbidities. In this talk, firstly, I present a projection of the spread of coronavirus in Mexico, which is based on a com-
partmental contact tracing model using Bayesian inference. Secondly, I present a projection of the hospital care demand and mortality of COVID-19 patients based on their health profile.
<details>
<summary>Click to expand</summary>

Specifically, I analyze the comorbidity associated with coronavirus using Machine Learning. I have implemented a classifier to predict the type of care procedure (outpatient/hospitalized)that a person diagnosed with coronavirus presenting chronic diseases will need, in this way I estimate the hospital care demand; next, I implement a second classifier to predict the survival/mortality of the patient. I present two techniques to deal with these kinds of imbalanced datasets related to outpatient/hospitalized and survived/deceased cases which occur in general for these types of diseases to obtain a better performance for the classification.
Finally, I present a metapopulation model to forecast the spread of the new coronavirus in Mexico City, using Bayesian inference. The daily mobility of people in Mexico City is mathematically represented by an origin-destination (O-D) matrix using a combination of three sources: the Mexico City government, an INEGI Transportation Mexican Survey, and COVID-19 Google Data Community Reports. Next, this O-D matrix is incorporated in a compartmental model. Given that working with metapopulation models leads to rather high computational time consumption, and parameter estimation of these models may lead to high memory RAM consumption, we do a clustering analysis that is based on mobility trends to work on these clusters of boroughs separately instead of taking all of them together at once.Our analysis of mobility trends can be helpful when making public health decisions.
</details>
**Wednesday, April 5, 2023** [Nicolas Smoll](https://scholar.google.com/citations?user=n4AIAh4AAAAJ&hl=en), Sushine Coast Hospital
**Title:** Data-linkage and Real-time Vaccine Effectiveness in Queensland Australia
**Abstract:** In this talk, I will introduce the machine learning data-linkage of the Queensland Nationally Notifiable Diseases dataset to the state-wide hospitalization dataset (Queensland Hospital Admitted Patient Data Collection) to estimate vaccine effectiveness in close to real-time. I will walk you through a pilot study that has completed the linkage and has estimated vaccine effectiveness against hospitalization for symptomatic COVID-19.

**Wednesday, March 1, 2023** [Emily Pfaff](https://www.med.unc.edu/medicine/directory/emily-pfaff/), UNC School of Medicine
**Title:** Phenotyping Long COVID Using Machine Learning and Electronic Health Records: Promise and Pitfalls
**Abstract:** Long COVID is a life-altering, potentially debilitating sequela of COVID-19.
<details>
<summary>Click to expand</summary>

Though it is clear that long COVID is a major public health crisis, we still lack a firm definition of the disease over two years after the condition was first recognized. Machine learning models trained on electronic health records may provide one avenue to enable us to find patterns among patients diagnosed with long COVID, which in turn may help us identify other patients who lack a diagnosis, but may have long COVID all the same. This is a first step toward a definition of long COVID that can be leveraged for future long COVID research, but there remains a long road ahead.
</details>

**Wednesday, February 1, 2023** [Xiunan Wang](https://www.utc.edu/directory/ptd497-mathematics-xiunan-wang/ptd497), The University of Tennessee at Chattanooga
**Title:** From Policy to Prediction: Forecasting COVID-19 Dynamics Under Imperfect Vaccination
**Abstract:** In this talk, I will introduce a novel method in forecasting the daily number of confirmed cases of infectious diseases by combining a mechanistic ordinary differential equation (ODE) model for infectious classes and a generalized boosting machine learning model (GBM) for predicting how public health policies and mobility data affect the transmission rate in the ODE model.
<details>
<summary>Click to expand</summary>

To illustrate, I will show how to apply the method to obtain a retrospective forecast of COVID-19 daily confirmed cases in the USA, and identify the relative influence of the policies used as the predictor variables. Our results indicate that the inclusion of data on non-pharmaceutical interventions can significantly improve the accuracy of the predictions. Moreover, the most influential predictor variables are the policies of restrictions on gatherings, testing and school closing. The modeling approach used in this work is helpful in designing improved forecasters as well as informing policymakers.
</details>

**Wednesday December 7, 2022** [Rozita Dara](https://www.uoguelph.ca/computing/people/rozita-dara), University of Guelph
[AI4Pandemics Talk #29:YouTube](https://youtu.be/yyWtwxFU93o)
**Title:** The use of social media and online data for early detection of infectious diseases
**Abstract:** Modeling infectious diseases is usually aimed at understanding disease spread mechanisms, predicting future outbreaks and their impact, and evaluating control measures.
<details>
<summary>Click to expand</summary>
Among infectious diseases, coronavirus disease (COVID-19) is of importance and has been the center of global attention in the past few years. In our study, we evaluated the use of social media outlets as a tool for early warning signals of COVID-19 outbreaks. We also used these online data sources to better understand the publics' opinion about COVID-19 control measures, including vaccines, and their perspective on public health agencies and leaders. We also examined the sources of misinformation and strategies to detect and control them online.
</details>)

**Wednesday, November 2, 2022** [Degui Zhi](https://sbmi.uth.edu/aigi/), University of Texas Health Science Center at Houston
[AI4Pandemics Talk #28: YouTube](https://youtu.be/WMHgxTbHFUw)
**Title:** Predicting COVID outcomes from EHR

**Wednesday, October 5, 2022** [Adam Dunn](https://adamgdunn.net/), The University of Sydney
[AI4Pandemics Talk #27: YouTube](https://youtu.be/6rApbf0RekY)
**Title:** What's wrong with health misinformation research and how multidisciplinary research can fix it
**Abstract:** During the COVID-19 pandemic we saw a substantial increase in data-driven research on misinformation but there is a disconnect between applied machine learning and what is needed to make good local decisions about where to focus efforts or how to design localised communication interventions.
<details>
<summary>Click to expand</summary>
Part of the reason is a disconnect between the goals of AI researchers and the needs of public health, and part is because of the challenge of working in multidisciplinary teams and communicating across disciplines. In this seminar, Adam will explain some of the study design flaws in the applied machine learning literature in the area, propose study designs that use machine learning and could be better suited to addressing the need for evidence in public health, and discuss some of the limitations associated with studies that require recruitment and participant involvement.
</details>

**Wednesday, August 31, 2022:** [Chinasa T. Okolo](http://www.cs.cornell.edu/~chinasa), Cornell University
[AI4Pandemics Talk #26: YouTube](https://youtu.be/9buFvXaDbZ0)
**Title:** Making AI Explainable for Novice Technology Users in Low-Resource Settings
**Abstract:** As researchers and technology companies rush to develop artificial intelligence (AI) applications that aid the health of marginalized communities, it is critical to consider the needs of the community health workers (CHWs) who will be increasingly expected to operate tools that incorporate these technologies.

<details>
<summary>Click to expand</summary>

My previous work has shown that these users have low levels of AI knowledge, form incorrect mental models about how AI works, and at times, may trust algorithmic decisions more than their own. This is concerning, given that AI applications targeting the work of CHWs are already in active development and early deployments in low-resource healthcare settings have already reported failures that created additional workflow inefficiencies and inconvenienced patients.
Explainable AI (XAI) can help avoid such pitfalls, but nearly all prior work has focused on users that live in relatively resource-rich settings (e.g., the US and Europe) and that arguably have substantially more experience with digital technologies such as AI. My research works to develop XAI for people with low levels of formal education and technical literacy, with a focus on healthcare in low-resource domains. This work involves demoing interactive prototypes with CHWs to understand what aspects of model decision-making need to be explained and how they can be explained most effectively, with the goal of improving how current XAI methods serve novice technology users.

</details>

**Wednesday, August 3, 2022:** [Omar Haggag](https://www.monash.edu/it/humanise-lab/about-us/our-students/current-monash-students/omar-haggag), Monash University
[AI4Pandemics Talk #25: YouTube](https://youtu.be/PQqgT3k3OUc)
**Title:** COVID-19 vs Social Media Apps: Does Privacy Really Matter?
**Abstract:** Many people are worried about using COVID-19 apps. There are several reasons behind this, including privacy issues, lack of trust and ethical concerns.

<details>
<summary>Click to expand</summary>

Also, several media platforms play a huge role in swinging reality to scare and panic the public and steer them away from using COVID-19 apps. On the other hand, Social Media & Productivity mobile apps have lots of well-demonstrated and publicised privacy and ethical issues. However, the public are still using them, often without any apparent concerns, and even at a higher rate during the pandemic. To better understand this behaviour, this project addressed the following two research questions.

How is personal data handled by COVID-19 and Social Media mobile apps?
What are the key issues raised by the users of COVID-19 and Social Media & Productivity apps as evidenced in their users’ reviews?

</details>

**Wednesday, July 6, 2022:** [Jiaying Li](https://qaehs.centre.uq.edu.au/profile/3046/jiaying-li), The University of Queensland
[AI4Pandemics Talk #24: YouTube](https://youtu.be/qSdg7ogb2vY)
**Title:** SARS-CoV-2 in sewer networks: tracking the COVID-19 spread in the community
**Abstract:** Monitoring SARS-CoV-2 RNA in sewer systems is an effective approach for understanding COVID-19 transmission in communities with higher spatial resolutions.

<details>
<summary>Click to expand</summary>

Wastewater surveillance for SARS-CoV-2 has been rapidly developed worldwide for monitoring the COVID-19 prevalence at the population level. Sampling in sewage networks, i.e., the upstream of a wastewater treatment plant, is proposed to understand community transmission. Passive samplers are cost-effective and suitable for sewer networks or catchments where autosamplers cannot be operated. This presentation will tell a full story about how we trace the SARS-CoV-2 in sewer networks: 1) how passive samplers are calibrated to provide time-integrative information, 2) how to determine passive samplers’ sensitivity for low COVID-19 cases in an area, and 3) how to apply passive samplers for upstream monitoring to identify the emergence and dynamics of COVID-19 in communities. These findings demonstrate the ability of upstream wastewater surveillance for identifying SARS-CoV-2 in low-case settings and tracking COVID-19 spread in communities with higher spatial resolutions.

</details>

**Wednesday, June 8, 2022:** [Caroline Colijn](https://www.sfu.ca/math/people/faculty/ccolijn/), Simon Fraser University
[AI4Pandemics Talk #23: YouTube](https://youtu.be/Sq9z-f0ikcI)
**Title:** Genomic Epidemiology in SARS-CoV-2: new tools and challenges
**Abstract:** Scientists around the world have sequenced over 2 million SARS-CoV-2 genomes in an effort to monitor and understand the evolution and transmission of this virus.
<details>
<summary>Click to expand</summary>

Virus sequences can help us understand the emergence of new variants with new phenotypes, track the virus' geographical movements and can help us to understand local transmission. However, there are mathematical and statistical challenges in making the most of this potentially rich source of information about viruses and how they spread. In this talk I will introduce the field of genomic epidemiology in general, and then describe recent research in our group. We have developed a method to use SARS-CoV-2 genomes to estimate serial intervals: the time between symptoms (or in some cases, sample collection) in infector-infectee pairs. Serial intervals are important because they underlie estimates of the reproductive number, Rt, which in turn is used to help understand the strength of transmission and the impact of different levels of vaccine coverage. I will describe the results of this method applied to data from Victoria, Australia. I will conclude by noting some broader challenges and opportunities for the genomic surveillance of SARS-CoV-2.
</details>

**Wednesday, May 25, 2022:** [Andres Colubri]
(https://profiles.umassmed.edu/display/24294491), UMass Chan Medical School
[AI4Pandemics Talk #22: YouTube](https://youtu.be/zilmInRvl34)
>**Title:** Operation Outbreak: an app-based platform for infectious disease education and research
>**Abstact:** Together with Harvard University Professor Pardis Sabeti and Dr. Todd Brown from The Inspire Project, we have been working since 2015 on Operation Outbreak (OO).
<details>
<summary>Click to expand</summary>

This project was originally motivated by the ever-present pandemic threat (at the time, made apparent by the West African Ebola outbreak) and the challenge of educating students about it in more engaging ways. Only five years later, COVID-19 reified epidemiologists’ predictions of a global pandemic caused by an emerging pathogen. As public health measures contributed to curbing the spread of COVID-19, innovative educational programs on infectious diseases could also play a role in controlling this pandemic––and in preparing for or preempting the next one. OO started as a mock outbreak activity for middle schoolers using stickers to mimic pathogen transmission, and eventually evolved to comprise three interconnected components: (1) an academic curriculum and textbook on pathogen biology, epidemiology, public health, political decision-making, and science communication during health emergencies; (2) an outbreak simulation experiential learning activity that synthesizes curricular content with a Bluetooth-enabled smartphone app; and (3) a multi-user dashboard that visualizes data generated during the outbreak simulation for informed reflection and skill development in epidemiology and quantitative data analysis. Facilitated by the smartphone app, the outbreak simulation spreads a virtual pathogen across participants’ phones via Bluetooth. Additionally, the simulation incorporates a series of role-playing activities for the students (e.g., governance, research, healthcare), taking place during the simulated outbreak and mirroring a real-world epidemic. The current OO app supports Bluetooth beacons and QR codes that simulate infectious sources and protective items (e.g., face masks and hazmat suits) and interventions such as testing and vaccinations, and we have been making substantial progress last year in the development of the app technology and conducting several large-scale pilots in middle and high schools, as well as colleges. Currently, my lab in the Program of Bioinformatics and Integrative Biology at the University of Massachusetts Medical School is focused not only on developing the technology behind OO, but also using OO as a “real-life” simulator that could help create and validate epidemiological models to better respond to future outbreaks.
</details>

**Wednesday, May 11, 2022:** [Max Menzies](https://maxmenzies.com/), Tsinghua University
[AI4Pandemics Talk #21: YouTube](https://www.youtube.com/watch?v=n5o00FTrZ4o&t=20s)
>**Title:** Targeted mathematical approaches based on topical questions regarding COVID-19
>**Abstract:** COVID-19 has consistently astonished policymakers and challenged researchers. From early predictions that it would "just go away" to later failures to understand precipitous upticks or decreases in cases and/or deaths, this pandemic has consistently been difficult if not impossible to forecast.
<details>
<summary>Click to expand</summary>

Thus, we took a different approach, focusing on the descriptive analysis of surprising phenomena. Throughout the pandemic, we have keenly paid attention to the news to find key questions of interest that lend themselves to the development of innovative and elegant mathematical approaches. Combining backgrounds of pure maths, applied maths and real-world experience, we will aim to present work that is mathematically interesting in its own right as well as findings and interpretations of relevance to a wide audience.
</details>

**Wednesday, April 27, 2022:** [Lock Yue Chew](https://personal.ntu.edu.sg/lockyue/), Nanyang Technological University
[AI4Pandemics Talk #20: YouTube](https://www.youtube.com/watch?v=A1FNfh8e-1I&t=458s)
>**Title:** Modelling COVID-19 Pandemic with Control Measures using a SEIR Multiplex Network Model
>**Abstract:**
>In this talk, I will present a Susceptible-Exposed-Infectious-Removed (SEIR) model that exploits both multiplex and temporal networks to study the outbreak dynamics of coronavirus infectious diseases (COVID–19) in Singapore.
<details>
<summary>Click to expand</summary>

Specifically, our multiplex network consists of a household network, a dormitory network, a workplace network, a temporal crowd network, and a temporal social gathering network. Our model has enabled us to emulate and tweak the different forms of real-world social interactions to study the spread of COVID–19 in Singapore by examining the dynamics of the reproduction number R0. Our approach demonstrates the potential effectiveness of various control measures against the spread of COVID–19 at different moments and stages of the pandemics.
</details>

**Wednesday, April 13, 2022:** [Paulo José da Silva](https://sinews.siam.org/Details-Page/designing-responses-to-the-covid-19-outbreak-from-simulation-to-optimization), University of Campinas
[AI4Pandemics Talk #19: YouTube](https://www.youtube.com/watch?v=1NDr6xtU7hI&t=1098s)
>**Title:** Robot dance: using optimization in pandemic simulations
>**Abstract:**
In this talk we will present the Robot Dance framework that allow us to use ideas from control and simulation in mathematical epidemiology models.
<details>
<summary>Click to expand</summary>

This allows us not only to simulate the model with fixed parameters but also to capture possible natural variations on parameters that appears in situations like fitting past data when different social distancing measures were taken, simulate the response to a spike in the number of case that would result into adoption of non-pharmacological mitigation, design optimal vaccinations campaigns, etc. [Further information](https://sinews.siam.org/Details-Page/designing-responses-to-the-covid-19-outbreak-from-simulation-to-optimization)
</details>

**Wednesday, March 30, 2022:** [Bridget Haire](https://kirby.unsw.edu.au/people/dr-bridget-haire), Kirby Institute
[AI4Pandemics Talk #18: YouTube](https://www.youtube.com/watch?v=2uQJI1Wz7uU)
>**Title:** More than just the numbers: the value-add of qualitative research in the evaluation of the COVIDSafe app
>**Abstract:**
>Mobile phone-based digital proximity tracing applications (apps) were rolled out in many countries early in the COVID-19 pandemic to enhance contact tracing.
<details>
<summary>Click to expand</summary>

Little was known about their real-world effectiveness. We undertook a mixed method evaluation of the COVIDSafe app in NSW. This comprised of a prospective quantitative study of that measured the number of infections identified by the app that were not identified by other contact tracing methods, and qualitative interviews with contact tracing staff who were incorporating the app into their work practice and those trained in the use of the app who had not used in in practice.
This presentation will briefly summarise the result of both the quantitative and qualitative aspects of the study. It will focus on the qualitative findings and what they added to developing a thorough understanding of problems identified with the app.
</details>

**Wednesday, March 16, 2022:** [Michaël Chass](https://www.chumontreal.qc.ca/en/crchum/researchers/michael-chasse), Centre Hospitalier de l’Université de Montréal (CHUM)
[AI4Pandemics Talk #17: YouTube](https://www.youtube.com/watch?v=wPBngVLJseA)
>**Title:** CODA-19: Collaborative Data Analysis to Improve Clinical Care in Patients with COVID-19

**Wednesday, March 2, 2022:** [James Hay](https://ccdd.hsph.harvard.edu/people/james-hay/), Harvard University
[AI4Pandemics Talk #16: YouTube](https://www.youtube.com/watch?v=1vRiJy-gkcg)
>**Title:** Using viral loads to improve virologic surveillance
>**Abstract:**
Virologic testing has been central to tracking the COVID-19 pandemic.
<details>
<summary>Click to expand</summary>

Most routine tests provide a quantitative result in the form of a cycle threshold (Ct) value -- a metric proportional to the log viral load of the sample. These data are usually reported as a binary result, thereby removing much of the information inherent in their full quantitative value. We propose that, despite their caveats and variability, the Ct value is a useful measure that can be harnessed to improve public health surveillance. In this talk, I will explain why the distribution viral loads changes predictably during the course of an epidemic, and how this relationship can be used to estimate an epidemic's trajectory from a single cross-sectional sample of RT-qPCR data. I will also discuss the implications of these findings for comparing viral loads between emerging variants, demonstrating surveillance scenarios where variant-specific viral kinetics can and cannot be reliably inferred.
</details>
**Wednesday, February 16, 2022:** [Alina Bialkowski](https://researchers.uq.edu.au/researcher/19747), The University of Queensland
[AI4Pandemics Talk #15: YouTube](https://https://www.youtube.com/watch?v=GVPZ9_OSxUI)
>**Title:** Interpretable machine learning for Chest X-Ray diagnosis and COVID detection
>**Abstract:**
The onset of the COVID-19 pandemic pivoted many research fields towards finding solutions to the disease. Machine learning on images offers great promise for fast and accurate detection of diseases, and in the early stages of the pandemic, some researchers attempted to rapidly deploy these techniques to diagnosing COVID-19.
<details>
<summary>Click to expand</summary>

In this talk, I will discuss some of the flaws encountered in this research which can be identified through interpretability analysis, and discuss the importance of working with medical experts, and understanding the full pipeline in machine learning modelling beyond modelling and accuracy metrics alone.
</details>

**Wednesday, February 2, 2022:** [Fahad Ahmed](https://scholar.google.com/citations?hl=en&user=39-51HMAAAAJ&view_op=list_works&sortby=pubdate), Wayne State University
[AI4Pandemics Talk #14: YouTube](https://www.youtube.com/watch?v=jBqgTPaelCM)
>**Title:** Actionable outcomes prediction: Mortality prediction in COVID-19 pandemic and beyond
>**Abstract:**
Background
The nextwave of COVID-19 pandemic is anticipated to be worse than the initial one and will strain the healthcare systems even more during the winter months. Our aim was to develop a novel machine learning-based model to predict mortality using the deep learning Neo-V framework. We hypothesized this novel machine learning approach could be applied to COVID-19 patients to predict mortality successfully with high accuracy.
<details>
<summary>Click to expand</summary>

Methods
We collected clinical and laboratory data prospectively on all adult patients (≥18 years of age) that were admitted in the inpatient setting at Aga Khan University Hospital between February 2020 and September 2020 with a clinical diagnosis of COVID-19 infection. Only patients with a RT-PCR (reverse polymerase chain reaction) proven COVID-19 infection and complete medical records were included in this study. A Novel 3-phase machine learning framework was developed to predict mortality in the inpatients setting. Phase 1 included variable selection that was done using univariate and multivariate Cox-regression analysis; all variables that failed the regression analysis were excluded from the machine learning phase of the study. Phase 2 involved new-variables creation and selection. Phase 3 and final phase applied deep neural networks and other traditional machine learning models like Decision Tree Model, k-nearest neighbor models, etc. The accuracy of these models was evaluated using test-set accuracy, sensitivity, specificity, positive predictive values, negative predictive values and area under the receiver-operating curves. A new external validation dataset was collected from Detroit Medical center in the United States. Validation of was done on this dataset (test-set accuracy, sensitivity, specificity, positive predictive values, negative predictive values and area under the receiver-operating curves).
Results
After application of inclusion and exclusion criteria (n=)1214 patients were selected from a total of 1228 admitted patients. We observed that several clinical and laboratory-based variables were statistically significant for both univariate and multivariate analyses while others were not. With most significant being septic shock (hazard ratio [HR], 4.30; 95% confidence interval [CI], 2.91–6.37), supportive treatment (HR, 3.51; 95% CI, 2.01–6.14), abnormal international normalized ratio (INR) (HR, 3.24; 95% CI, 2.28–4.63), admission to the intensive care unit (ICU) (HR, 3.24; 95% CI, 2.22–4.74), treatment with invasive ventilation (HR, 3.21; 95% CI, 2.15–4.79) and laboratory lymphocytic derangement (HR, 2.79; 95% CI, 1.6–4.86). Machine learning results showed our deep neural network (DNN) (Neo-V) model outperformed all conventional machine learning models with test set accuracy of 99.53%, sensitivity of 89.87%, and specificity of 95.63%; positive predictive value, 50.00%; negative predictive value, 91.05%; and area under the receiver-operator curve of 88.5. The external validation shows test set accuracy of 64.5%, sensitivity of 84.3%, specificity of 34.6%, positive predicative value of 66.0%, negative predicative value of 59.4%, true positive = 70, true negative = 36, false positive = 19 and false negative = 13.
Conclusion
Our novel Deep-Neo-V model outperformed all other machine learning models during development and showed of good accuracy in one of the external validation. The model is easy to implement, user friendly and with high accuracy.
Keywords: COVI-19, Pandemic, Machine Learning, Deep Neural Network, Mortality, SARS-COV-2
</details>

**Wednesday, January 19, 2022:** [Jude Kong](https://judekong.mathstats.yorku.ca/), York University
[AI4 Pandemics Talk #13: YouTube](https://youtu.be/aDE8WgOj3k8)
>**Title:** The impact of social, economic, environmental factors on the dynamics of COVID-19
>**Abstract:**
The COVID-19 pandemic has reached a stage where there is now sufficient data to infer whether the basic reproduction number (R0) varies across countries, and what demographic, social, and environmental factors, other than interventions, characterize vulnerability to the virus.
<details>
<summary>Click to expand</summary>

In this talk, I will present the first global estimate of R0 across all continents, and the results of a comprehensive investigation on what social, economic, and environmental factors characterize vulnerability to the virus. Understanding how space and time dependent factors predispose a community to a different COVID-19 rate of increase is essential to assessing the efficacy of interventions.
</details>

**Wednesday, December 15:** [Pranesh Padmanabhan](https://researchers.uq.edu.au/researcher/11648), The University of Queensland
[AI4Pandemics Talk #12: Toutube](https://www.youtube.com/watch?v=uxyE3RsQov4&t=1979s)
>**Title:** Predicting the effectiveness of COVID-19 vaccines and treatments
>**Abstract:**
The raging global COVID-19 pandemic has triggered enormous global efforts to develop and improve COVID-19 vaccines and antiviral treatments.
<details>
<summary>Click to expand</summary>

Predicting the efficacy of COVID-19 vaccines and treatments would aid in the identification of optimal vaccine/treatment development and usage strategies. In this talk, I will first describe our multiscale mathematical model that presents mechanistic links between COVID-19 vaccine efficacies and the vaccine-induced neutralisation antibody responses. I will then discuss a mathematical model of SARS-CoV-2 entry into target cells that unravels an unexpected synergy between antivirals targeting the different virus entry pathways.
</details>

**Wednesday, December 1:** [Amalie Dyda](https://public-health.uq.edu.au/profile/5957/amalie-dyda), The University of Queensland
[AI4Pandemics Talk #11: YouTube](https://www.youtube.com/watch?v=jDHIxLHDkwY&t=39s)
>**Title:** Associations between vaccine coverage and vaccine information exposure on Twitter
>**Abstract:**
Vaccination is a key public health measure to control the spread of vaccine preventable diseases, highlighted by the current COVID-19 pandemic.
<details>
<summary>Click to expand</summary>

However, uptake of adult vaccination is influenced by a number of factors including awareness, perceived risk and safety, which may be affected by exposure to misinformation. With current high levels of social media use, many individuals may now be influenced by health information in this medium which can include exposure to vaccine misinformation. This seminar will describe a case study, using Human papillomavirus (HPV) vaccination as an example, of an investigation into associations between information people are exposed to on social media and levels of vaccination coverage and discuss future directions for research.

</details>

**Wednesday, November 17:** [Lewis Mitchell](https://lewismath.github.io/), The University of Adelaide
[AI4Pandemics Talk #10: YouTube](https://www.youtube.com/watch?v=7gfYOjX6DYc&t=20s)
>**Title:** Risk mapping for COVID-19 using social media data
>**Abstract:**
Much has been made of the possible application of novel datasets such as social media and search data for combatting infectious diseases, particularly during the COVID-19 pandemic.
<details>
<summary>Click to expand</summary>

In this talk we evaluate one such dataset – mobility data provided by Facebook’s Data for Good program – for quantifying risk associated with human mobility in Australian cities during 2020. We describe the dataset, how it is created, its limitations, and present a methodology for using this dataset with a simple mathematical model to predict spatial risk. Results show that the dataset and method were effective in estimating outbreak risk for 2 key outbreaks in 2020.

</details>

**Wednesday, November 3:** [Romesh Abeysuriya](https://www.burnet.edu.au/people/578_romesh_abeysuriya), Burnet Institute
[AI4Pandemics Talk #9: YouTube](https://www.youtube.com/watch?v=KbAJWx2rEhE)
>**Title:** Long-term COVID-19 strategies with intermittent control measures
>**Abstract:**
The COVID-19 vaccines used in Australia have all demonstrated high efficacy against severe disease and death. However, multiple models have shown that the combination of imperfect protection against infection and more infectious variants means that Australia, and other countries globally, are unlikely to achieve herd immunity.
<details>
<summary>Click to expand</summary>

A key question is therefore what control strategies are proportionate and sustainable, in a world with high vaccine coverage, no herd immunity, and ongoing importations of cases into the community from relaxed quarantine and increased travel. In this seminar, I will present a modelling study carried out by the Burnet Institute where we used the Covasim model to explore options for using intermittent measures to maintain long-term epidemic control.
</details>

**Wednesday, October 20:** [Rebecca Chisholm](https://scholars.latrobe.edu.au/rchisholm/publications), La Trobe University
[AI4Pandemics Talk #8: YouTube](https://www.youtube.com/watch?v=0mOuBGtfX3M&t=3s))
>**Title:** Contribution of mathematical modelling to COVID-19 response strategies in regional and remote Australian Aboriginal and Torres Strait Islander communities

>**Abstract:**
The health and science communities recognised early on in the SARS-CoV-2 pandemic that Aboriginal and Torres Strait Islander Australians were likely to be at high risk of COVID-19 infection and severe outcomes, due to high rates of comorbidities associated with severe outcomes, and multiple factors predisposing to increased SARS-CoV-2 transmission.
<details>
<summary>Click to expand</summary>

In March 2020, the Australian Government convened the Aboriginal and Torres Strait Islander Advisory Group on COVID-19 (IAG), co-chaired by the Department of Health and the National Aboriginal Community Controlled Health Organisation. The role of the IAG was to develop and deliver a National Management Plan to protect Aboriginal and Torres Strait Islander communities. Our research groups—located at the Doherty Institute, the Kirby Institute and La Trobe University—were commissioned to carry out modelling, under the guidance of the IAG, to help inform aspects of this plan related to regional and remote communities. In this presentation I will describe how we used modelling to address a number of questions of interest to the IAG regarding the importance of a timely response to the first identified case of COVID-19, who should be quarantined and/or tested in communities, and whether there is a role for community-wide lockdown in initial containment.
</details>

**Wednesday, October 13:** [Colleen Lau](https://researchers.uq.edu.au/researcher/2260),The University of Queensland
[AI4Pandemics Talk #7: YouTube](https://www.youtube.com/watch?v=uctiEO8hNDc&t=12s)
>**Title:** CRISPER: COVID-19 Real-time Information System for Preparedness and Epidemic Response
>**Abstract:**
A major challenge during the COVID-19 pandemic has been the need to share data and information public while protecting data privacy.
<details>
<summary>Click to expand</summary>

Effective communication of real-time data is critical for informing risk assessment and decision making, and to support a unified national response. Real-time dashboards have become important platforms for information sharing during the COVID-19 pandemic. This presentation will describe the development of CRISPER, a COVID-19 Real-time Information System for Preparedness and Epidemic Response, including the challenges experience. CRISPER includes a suite of interactive visualisation and mapping tools and automatic alerts for COVID-19 cases, deaths, testing, and contact tracing exposure sites across Australia. The system also allows the use of differential privacy algorithms to protecting data privacy.

</details>

**Thursday, October 7:** [Clair Sullivan](https://researchers.uq.edu.au/researcher/13187), The University of Queensland.

>**Title:** The Algorithm will see you now

**Wednesday, September 8:** [Peter Frazier](https://people.orie.cornell.edu/pfrazier/), Cornell University
[AI4Pandemics Talk #5: YouTube](https://youtu.be/QHVruw121RA) [Presentation Slides](https://docs.google.com/presentation/d/1q4G4sURbQL7xrivO9SvAal9kNKL5OYsL8HCkiNCnIPs/edit?usp=sharing)
>**Title:** Fighting COVID-19 at Cornell University

>**Abstract**
Universities around the world faced a challenging decision during the summer of 2020: whether to reopen for in-person instruction despite the pandemic and how to protect campus populations if they did.
<details>
<summary>Click to expand</summary>

Operations research and data science were a fundamental part of these decisions at Cornell University in the USA. First, models developed by Cornell's COVID-19 Mathematical Modeling Team were used to design the testing interventions that are a cornerstone of Cornell’s COVID-19 control strategy: targeted asymptomatic screening that tests all undergraduates twice per week and an adaptive testing program that goes beyond traditional contact tracing to test the full social circle of positive cases. Second, these same models were the basis for Cornell's decision to reopen for a fall semester with in-person instruction. They showed that reopening with aggressive mandatory testing was surprisingly less risky than virtual instruction. Data suggested that thousands of students would return to the area whether in-person instruction was offered or not, and a weaker ability to enforce mandatory testing for these students risked being unable to control clusters in that population. Reopening with asymptomatic screening was successful, with only 0.5% of students, staff and faculty infected over the semester. This talk will share insights from this experience and explain practical tools that supported this work.
</details>

**Wednesday, August 25:** [Joel Miller](https://scholars.latrobe.edu.au/jcmiller), La Trobe University
[AI4Pandemics Talk #4: YouTube](https://www.youtube.com/channel/UChfQzhyKowwzYFAXh6ziRlg)

>**Title:** COVID and the misunderstood denominator…

>**Abstract:**
Like past epidemics, the efforts to stop the transmission of SARS-CoV-2 have been hindered by the parallel transmission of misinformation (inaccurate information) as well as disinformation (intentionally deceptive inaccurate information).
<details>
<summary>Click to expand</summary>

Unlike historical epidemics, the social media landscape has accelerated the spread of misinformation. I will discuss the role misinformation has played in the pandemic.
</details>

**Wednesday, August 18:** [Jeremy Howard](https://en.wikipedia.org/wiki/Jeremy_Howard_(entrepreneur)), fast.ai & University of San Francisco.
[AI4Pandemics Talk #3: YouTube](https://www.youtube.com/watch?v=9jQE27bZOcU&t=195s)

>**Title:** How a little-known data scientist convinced the West to wear face masks

>**Abstract:**
The title of this talk is actually the title of an article in [The Telegraph](https://archive.is/Jkxdj) about my journey in co-founding the [Masks4All](https://masks4all.co/) movement. I share how I found myself becoming the face of Masks4All globally, and what I learned about making an impact as a data scientist.

**Wednesday, July 28:** [Kirsty Short](https://scmb.uq.edu.au/profile/4618/kirsty-short), UQ.
[AI4Pandemics Talk #2: YouTube](https://www.youtube.com/watch?v=KYufp__7NQg&t=246s)
>**Title:** The role of children in the spread of SARS-CoV-2

>**Abstract:**
The role of children in the spread of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) remains highly controversial.
<details>
<summary>Click to expand</summary>

To address this issue, we performed a meta-analysis of the published literature on household SARS-CoV-2 transmission clusters (n = 213 from 12 countries). Only 8 (3.8%) transmission clusters were identified as having a pediatric index case. Asymptomatic index cases were associated with a lower secondary attack in contacts than symptomatic index cases (estimate risk ratio [RR], 0.17; 95% confidence interval [CI], 0.09-0.29). To determine the susceptibility of children to household infections the secondary attack rate in pediatric household contacts was assessed. The secondary attack rate in pediatric household contacts was lower than in adult household contacts (RR, 0.62; 95% CI, 0.42-0.91). These data have important implications for the ongoing management of the COVID-19 pandemic, including potential vaccine prioritization strategies.
</details>

**Wednesday, July 14:** [Christopher Rackauckas](https://chrisrackauckas.com/), MIT.
[AI4Pandemics Talk #1: YouTube](https://www.youtube.com/watch?v=7yPU_04Unb8&t=359s)

> **Title:** Learning Epidemic Models That Extrapolate.

>**Abstract:**
Modern techniques of machine learning are uncanny in their ability to automatically learn predictive models directly from data.
<details>
<summary>Click to expand</summary>

However, they do not tend to work beyond their original training dataset. Mechanistic models utilize characteristics of the problem to ensure accurate qualitative extrapolation but can lack in predictive power. How can we build techniques which integrate the best of both approaches? In this talk we will discuss the body of work around universal differential equations, a technique which mixes traditional differential equation modeling with machine learning for accurate extrapolation from small data. We will showcase how incorporating different variations of the technique, such as Bayesian symbolic regression and optimizing the choice of architectures, can lead to the recovery of predictive epidemic models in a robust way. The numerical difficulties of learning potentially stiff and chaotic models will highlight how most of the adjoint techniques used throughout machine learning are inappropriate for learning scientific models, and techniques which mitigate these numerical ills will be demonstrated. We end by showing how these improved stability techniques have been automated and optimized by the software of the SciML organization, allowing practitioners to quickly scale these techniques to real-world applications.
</details>