AI And Engineering For Healthcare Crises: Rapid Response Strategies For Covid-19 (IET Webinar).
In terms of computational power, he mentioned that a human takes 0.01 FLOPS (100 seconds) to compute a floating point calculation, and the hardware equivalent would need to be at 1000 TPU's to make the same calculation (I still need a calculator). To this hardware it's data analysing techniques like sigmoid functions that can give precision-based answers to identify false negatives in health care mathematics - again, this is another reason why covid-19 data is vital where without real world data the A.I calculators cannot predict how cures might affect mortality statistics.
Only with the right data, can the current vaccination-peer process be simplified to bring to-market speedier vaccinations, further, the real time feeds will also balance against possible bias that will always be present in A.I models, so again, data feeds are important to also provide focus on causation Vs correlation factors, like, do face masks actually help different geo-locations, true effects of asymptomatic people on infection rates, surface life span of the virus and identifying likely infectious variables. Understanding infection variables is again vital to know what mitigative efforts need to be created and without data, it remains a science at best.
In cyber security, the 7 steps of the 'Cyber Kill Chain' are well known i.e. how does a specific digital-virus infect the digital asset, though the same analysis currently cannot be done with covid-19 as it is not well known how host contamination occurs, replicates, identifies ACE2 receptors to then overwhelm the cell (ddos attack) to eventually fail - at this point the cytokine storm immune system gets confused and self destructs the cell causing eventual hospitalisation, here, A.I can help analyse stages of covid-19 infection.
Therefore once actual data is known from existing patients, it will be possible to identify spreading rates for different areas i.e. some form of targeted predicting to identified exponential growth. So, not to say the end goal is to know that 'washing your hands 3 times a day' will reduce infection, but more rather to identify, to what extent does social distancing reduce probable infections . All this data needs to be modeled, and currently, popular models of choice are, SIR, Agent or Curve. He then reviewed the math's on how all this data effects R factors whose output ultimately assists in quarantine and vaccination decisions.
Specifically, A.I will help reduce development factors that typically would have taken over 10 years (without A.I), however the trade-off between 'speed science Vs treatment accuracy' is based on, again, real world data. Dr Arachna explained that, vaccine factories will be another challenge, where estimates of 16 billion vaccines are going to be required, and its here where learning from responses to older Covid strains like SARS or MERS will help, such as how telemedicine efforts were vital to help local policies to aid patient recovery. She explained how these polices have to cover basic considerations now like, how hierarchy of controls effect travel, facility controls, hand sanitizer and mask usage against medical risks for population zone creation to ensure mitigative controls can be applied effectively.
HIV took 40 years of trials to identify possible treatments with some level of certainty.
Register to view the webinar here .