A new COVID-19 model at Caltech, using AI to predict the disease’s impact, is dramatically outperforming other models. Its performance is so impressive that it has attracted public health officials across the country.
While various models to predict a disease’s spread already exist, few, if any, uses artificial intelligence as efficiently as this one. Incorporating AI allows this model to make predictions based on the observation in real-time. It relies on the accurate data of what’s happening instead of traditional models that behave based on what the model’s designers think will happen. By using AI, it is possible to find patterns hidden in data sets that the human mind alone cannot recognize.
According to Yaser Abu-Mostafa, professor of electrical engineering and computer science at Caltech, the man behind the AI-Driven COVID-19 model, AI is a powerful tool. He said that it only makes sense to apply AI to one of the most urgent problems that the world is facing. He leads the development of the new CS156 model, named for Caltech’s computer science class where it originated.
The researchers are evaluating the CS156 model’s accuracy by comparing it to the predictions made by the ensemble model that the Centers for Disease Control and Prevention has built using 45 major models from various universities and institutes across the nation. Using 1,500 predictions as comparison points, the researchers found that as of November 25th, the CS156 model is 58% of the time more accurate than CDC’s ensemble model. This AI-driven model also regularly outperforms the benchmark projections of IHME (Institute for Health Metrics and Evaluation).
Abu-Mostafa said that he is currently expanding the CS156 model based on public health officials’ feedback so the model can become a lifesaving tool for guiding policy decisions. California Department of Public Health (CDPH) and representative of the New York City Commissioner of Health have contacted Abu-Mostafa regarding the CS156 model.
Based on the feeding, the researchers have tweaked the model to allow public health officials to predict how interventions like maks mandates, stay-at-home orders, etc., can control the disease’s spread. With these new predictions at their disposal, public health officials can now better evaluate the situation, and their intervention will be more likely to help. Abu-Mostafa has learned that the CDC is already using predictions from CS156 in its decision-making process.
Abu-Mostafa said that they are working on CS156 feverishly, and his group has gathered data on every COVID-19 policy in California’s every county since the onset of the pandemic. “It’s a tricky problem,” Abu-Mostafa said. He also stated that demographics play an essential role in prediction making. Younger people usually don’t abide by public health guidance as much as older people, and policies targeted towards individuals are less effective than those targeted towards businesses. Despite all these complexities, Abu-Mostafa and his team are digging into the data to make more accurate predictions using the CS156 model.
The AI-driven COVID-19 model, now known as the CS156 model, initially started in Abu-Mostafa’s computer science class, CS/CNS/EE 156, Learning Systems. It began during Caltech’s spring 2020 term. Usually, each time they typically apply AI to a much generic topic like movie recommendations. However, Abu-Mostafa and his students recognize the opportunity to make a real difference with their work. At the start of the project, eighty students were in the class. An additional seventy signed up when Abu-Mostafa announced the new challenge. By the term’s end, they had created 40 viable models, out of which ten were already competitive with existing epidemiological models.
Throughout the summer, Abu-Mostafa and a core student group continued working and refining their data collection and modeling. The group of researchers officially launched the CS156 model on August 24th. In the fall of 2020, when students returned to their studies, Abu-Mostafa continued working on the model, managing the models’ data aggregation. However, he attributes the continued success of the CS156 model to his students’ hard work.
Abu-Mostafa describes his work as, “At this point, I am like a chef cooking a meal.” He said that you need the right ingredients because you cannot have the proper meal without them. He hopes that the news of the CS156 model’s efficacy will bring it to the other public health policymakers’ attention who can utilize its predictions in their decision making and save lives.
Abu-Mostafa said that everyone working on COVID-19 models is working towards the same goal of winning the war against this raging pandemic. “We are here to do our part,” he concluded.