SARS-CoV-2 transmission dynamics in South Africa and epidemiological characteristics of the Omicron variant

Within days of first detection, Omicron SARS-CoV-2 variant case numbers grew exponentially and spread globally. To better understand variant epidemiological characteristics, we utilize a model-inference system to reconstruct SARS-CoV-2 transmission dynamics in South Africa and decompose novel variant transmissibility and immune erosion. Accounting for under-detection of infection, infection seasonality, nonpharmaceutical interventions, and vaccination, we estimate that the majority of South Africans had been infected by SARS-CoV-2 before the Omicron wave. Based on findings for Gauteng province, Omicron is estimated 100.3% (95% CI: 74.8 - 140.4%) more transmissible than the ancestral SARS-CoV-2 and 36.5% (95% CI: 20.9 - 60.1%) more transmissible than Delta; in addition, Omicron erodes 63.7% (95% CI: 52.9 - 73.9%) of the population immunity, accumulated from prior infections and vaccination, in Gauteng.

Framework of Electrical Fire Probability Estimation Based on Bayesian Network Model Inference

Electrical fire had become one of the main parts in total fire accidents. Most of researches rely on the complex combustion models, which consume a huge number of computational resources. However, few studies focus on evaluating fire disaster by probability theory, and estimate the likelihood of fire occurring by the calculation result of probability based on the current data from the sensor. Bayesian Network is introduced due to the advantage of calculation complexity, ability of expressing uncertain factors and the accuracy of model with incomplete data. Some problems should be solved before using Bayesian Network to inference events based on given evidences. In this paper, the structure and the parameter of the Bayesian Network is created by the discussing result of the experts and scholars in electrical fire research field. A frequently-used fuzzy function called Sigmoid function to process data from raw data to the probability. Inference result by Bayesian Network is calculated by the Variable Elimination algorithm. A case study about the simulation of analyzing the probability of electrical fire happened when the load of circuit is under the high status. Research result shows that Bayesian Network model is suitable for estimating and analyzing in the scenario of electrical fire. Model has a good robust to express probability of electrical fire probability, which is of vital importance for estimating whether the fire occurs or not, thus providing significant information and instruction for preventing electrical fire and the sustainability of the environment. Based on the simulation result, it can conclude that the Bayesian network model inference is suitable for the electrical fire estimation scenario, and the introducing of this scheme is possible for predict electrical fire.

Tree-based machine learning performed in-memory with memristive analog CAM

Tree-based machine learning techniques, such as Decision Trees and Random Forests, are top performers in several domains as they do well with limited training datasets and offer improved interpretability compared to Deep Neural Networks (DNN). However, these models are difficult to optimize for fast inference at scale without accuracy loss in von Neumann architectures due to non-uniform memory access patterns. Recently, we proposed a novel analog content addressable memory (CAM) based on emerging memristor devices for fast look-up table operations. Here, we propose for the first time to use the analog CAM as an in-memory computational primitive to accelerate tree-based model inference. We demonstrate an efficient mapping algorithm leveraging the new analog CAM capabilities such that each root to leaf path of a Decision Tree is programmed into a row. This new in-memory compute concept for enables few-cycle model inference, dramatically increasing 103 × the throughput over conventional approaches.

Development of a model-inference system for estimating epidemiological characteristics of SARS-CoV-2 variants of concern

To support COVID-19 pandemic planning, we develop a model-inference system to estimate epidemiological properties of new SARS-CoV-2 variants of concern using case and mortality data while accounting for under-ascertainment, disease seasonality, non-pharmaceutical interventions, and mass-vaccination. Applying this system to study three variants of concern, we estimate that B.1.1.7 has a 46.6% (95% CI: 32.3–54.6%) transmissibility increase but nominal immune escape from protection induced by prior wild-type infection; B.1.351 has a 32.4% (95% CI: 14.6–48.0%) transmissibility increase and 61.3% (95% CI: 42.6–85.8%) immune escape; and P.1 has a 43.3% (95% CI: 30.3–65.3%) transmissibility increase and 52.5% (95% CI: 0–75.8%) immune escape. Model simulations indicate that B.1.351 and P.1 could outcompete B.1.1.7 and lead to increased infections. Our findings highlight the importance of preventing the spread of variants of concern, via continued preventive measures, prompt mass-vaccination, continued vaccine efficacy monitoring, and possible updating of vaccine formulations to ensure high efficacy.