Intelligent electromagnetic metasurface camera: system design and experimental results

Electromagnetic (EM) sensing is uniquely positioned among nondestructive examination options, which enables us to see clearly targets, even when they visually invisible, and thus has found many valuable applications in science, engineering and military. However, it is suffering from increasingly critical challenges from energy consumption, cost, efficiency, portability, etc., with the rapidly growing demands for the high-quality sensing with three-dimensional high-frame-rate schemes. To address these difficulties, we propose the concept of intelligent EM metasurface camera by the synergetic exploitation of inexpensive programmable metasurfaces with modern machine learning techniques, and establish a Bayesian inference framework for it. Such EM camera introduces the intelligence over the entire sensing chain of data acquisition and processing, and exhibits good performance in terms of the image quality and efficiency, even when it is deployed in highly noisy environment. Selected experimental results in real-world settings are provided to demonstrate that the developed EM metasurface camera enables us to see clearly human behaviors behind a 60 cm-thickness reinforced concrete wall with the frame rate in order of tens of Hz. We expect that the presented strategy could have considerable impacts on sensing and beyond, and open up a promising route toward smart community and beyond.

The basic reproduction number of COVID-19 across Africa

The pandemic of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) took the world by surprise. Following the first outbreak of COVID-19 in December 2019, several models have been developed to study and understand its transmission dynamics. Although the spread of COVID-19 is being slowed down by vaccination and other interventions, there is still a need to have a clear understanding of the evolution of the pandemic across countries, states and communities. To this end, there is a need to have a clearer picture of the initial spread of the disease in different regions. In this project, we used a simple SEIR model and a Bayesian inference framework to estimate the basic reproduction number of COVID-19 across Africa. Our estimates vary between 1.98 (Sudan) and 9.66 (Mauritius), with a median of 3.67 (90% CrI: 3.31 - 4.12). The estimates provided in this paper will help to inform COVID-19 modeling in the respective countries/regions.

Contraction bias in temporal estimation

When asked to compare the perceptual features of two serially presented objects, participants are often biased to over- or under-estimate the difference in magnitude between the stimuli. Overestimation occurs consistently when a) the two stimuli are relatively small in magnitude and the first stimulus is larger in magnitude than the second; or b) the two stimuli are relatively large in magnitude and the first stimulus is smaller in magnitude than the second; underestimation consistently occurs in the complementary cases. This systematic perceptual bias, known as the contraction bias, was demonstrated for a multitude of perceptual features and in various modalities, but it is yet unknown whether it also exists in the temporal domain. Here, we tested whether estimation of time-duration is affected by the contraction bias. In each trial of three experiments (n=20 each), participants compared the duration of two visually presented stimuli. Findings revealed over- and under-estimation effects as predicted by the contraction bias. In addition, we found that the bias was asymmetrical, indicating that, in some cases, the subjective center of the distribution was shifted to the left. Here, we discuss this asymmetry and describe how these findings can be explained via a Bayesian inference framework.

Multifaceted empathy differences in children and adults with autism

Although empathy impairments have been reported in autistic individuals, there is no clear consensus on how emotional valence influences this multidimensional process. In this study, we use the Multifaceted Empathy Test for juveniles (MET-J) to interrogate emotional and cognitive empathy in 184 participants (ages 8–59 years, 83 autistic) under the robust Bayesian inference framework. Group comparisons demonstrate previously unreported interaction effects between: (1) valence and autism diagnosis in predictions of emotional resonance, and (2) valence and age group in predictions of arousal to images portraying positive and negative facial expressions. These results extend previous studies using the MET by examining differential effects of emotional valence in a large sample of autistic children and adults with average or above-average intelligence. We report impaired cognitive empathy in autism, and subtle differences in emotional empathy characterized by less distinction between emotional resonance to positive vs. negative facial expressions in autism compared to neurotypicals. Reduced emotional differentiation between positive and negative affect in others could be a mechanism for diminished social reciprocity that poses a universal challenge for people with autism. These component- and valence- specific findings are of clinical relevance for the development and implementation of target-specific social interventions in autism.

PopART-IBM, a highly efficient stochastic individual-based simulation model of generalised HIV epidemics developed in the context of the HPTN 071 (PopART) trial

Mathematical models are powerful tools in HIV epidemiology, producing quantitative projections of key indicators such as HIV incidence and prevalence. In order to improve the accuracy of predictions, such models need to incorporate a number of behavioural and biological heterogeneities, especially those related to the sexual network within which HIV transmission occurs. An individual-based model, which explicitly models sexual partnerships, is thus often the most natural type of model to choose. In this paper we present PopART-IBM, a computationally efficient individual-based model capable of simulating 50 years of an HIV epidemic in a large, high-prevalence community in under a minute. We show how the model calibrates within a Bayesian inference framework to detailed age- and sex-stratified data from multiple sources on HIV prevalence, awareness of HIV status, ART status, and viral suppression for an HPTN 071 (PopART) study community in Zambia, and present future projections of HIV prevalence and incidence for this community in the absence of trial intervention.

Wilson statistics: derivation, generalization and applications to electron cryomicroscopy

The power spectrum of proteins at high frequencies is remarkably well described by the flat Wilson statistics. Wilson statistics therefore plays a significant role in X-ray crystallography and more recently in electron cryomicroscopy (cryo-EM). Specifically, modern computational methods for three-dimensional map sharpening and atomic modelling of macromolecules by single-particle cryo-EM are based on Wilson statistics. Here the first rigorous mathematical derivation of Wilson statistics is provided. The derivation pinpoints the regime of validity of Wilson statistics in terms of the size of the macromolecule. Moreover, the analysis naturally leads to generalizations of the statistics to covariance and higher-order spectra. These in turn provide a theoretical foundation for assumptions underlying the widespread Bayesian inference framework for three-dimensional refinement and for explaining the limitations of autocorrelation-based methods in cryo-EM.

Domain Generalization under Conditional and Label Shifts via Variational Bayesian Inference

In this work, we propose a domain generalization (DG) approach to learn on several labeled source domains and transfer knowledge to a target domain that is inaccessible in training. Considering the inherent conditional and label shifts, we would expect the alignment of p(x|y) and p(y). However, the widely used domain invariant feature learning (IFL) methods relies on aligning the marginal concept shift w.r.t. p(x), which rests on an unrealistic assumption that p(y) is invariant across domains. We thereby propose a novel variational Bayesian inference framework to enforce the conditional distribution alignment w.r.t. p(x|y) via the prior distribution matching in a latent space, which also takes the marginal label shift w.r.t. p(y) into consideration with the posterior alignment. Extensive experiments on various benchmarks demonstrate that our framework is robust to the label shift and the cross-domain accuracy is significantly improved, thereby achieving superior performance over the conventional IFL counterparts.