IBM

Person search (PS) is a computer vision problem that joins the two tasks of person detection and person re-identification (ReID). Previous works handle PS problem with either two-step or one-step approaches and have attained much attention due to complex challenges in the scene such as appearance variations, background clutter, and deformation. These approaches achieve significant performance but are still prone to performance degradation under complex scenes which may jeopardize the accuracy of person search methods. In this paper, we propose a novel Part-based Signal Modulation module for Person Search (PSM-PS) within a faster R-CNN-based person search framework. The proposed PSM module transforms the person parts, represented as part tokens, in a wave-like manner, where amplitude indicates the real part and phase shows the imaginary part in a complex domain. The proposed PSM module modulates the pedestrian part tokens such that it enhances the feature representation where the close parts of the person have a close phase compared to others. The experiments are performed over the two prominent person search datasets: CUHK-SYSU [23] and PRW [26]. The extensive experimental study demonstrates the effectiveness of our method and shows the state-of-the-art performance compared to other methods.

During the 2020 US presidential election, conspiracy theories about large-scale voter fraud were widely circulated on social media platforms. Given their scale, persistence, and impact, it is critically important to understand the mechanisms that caused these theories to spread. The aim of this preregistered study was to investigate whether retweet frequencies among proponents of voter fraud conspiracy theories on Twitter during the 2020 US election are consistent with frequency bias and/or content bias. To do this, we conducted generative inference using an agent-based model of cultural transmission on Twitter and the VoterFraud2020 dataset. The results show that the observed retweet distribution is consistent with a strong content bias causing users to preferentially retweet tweets with negative emotional valence. Frequency information appears to be largely irrelevant to future retweet count. Follower count strongly predicts retweet count in a simpler linear model but does not appear to drive the overall retweet distribution after temporal dynamics are accounted for. Future studies could apply our methodology in a comparative framework to assess whether content bias for emotional valence in conspiracy theory messages differs from other forms of information on social media.

In the electronic environment of today, the sophistication of a user cannot be anticipated to be at any particular advanced level. As a matter of fact, users range from first graders, to individuals with advanced knowledge, to elders with little or no technical background. As a result, a much greater burden is placed on software developers to ensure the product they release operates correctly under any conceivable key entry from a user. Anything less can be a disaster for a product, or set of products, due to customer dissatisfaction and complaints.

Background The COVID‐19 pandemic may influence delivery outcomes through direct effects of infection or indirect effects of disruptions in prenatal care. We examined early pandemic‐related changes in birth outcomes for pregnant women with and without a COVID‐19 diagnosis at delivery. Methods We compared four delivery outcomes—preterm delivery (PTD), severe maternal morbidity (SMM), stillbirth, and cesarean birth—between 2017 and 2019 (prepandemic) and between April and December 2020 (early‐pandemic) using interrupted time series models on 11.8 million deliveries, stratified by COVID‐19 infection status at birth with entropy weighting for historical controls, from the Healthcare Cost and Utilization Project across 43 states and the District of Columbia. Results Relative to 2017–2019, women without COVID‐19 at delivery in 2020 had lower odds of PTD (OR = 0.93; 95% CI = 0.92–0.94) and SMM (OR = 0.88; 95% CI = 0.85–0.91) but increased odds of stillbirth (OR = 1.04; 95% CI = 1.01–1.08). Absolute effects were small across race/ethnicity groups. Deliveries with COVID‐19 had an excess of each outcome, by factors of 1.07–1.46 for outcomes except SMM at 4.21. The effect for SMM was more pronounced for Asian/Pacific Islander non‐Hispanic (API; OR = 10.51; 95% CI = 5.49–20.14) and Hispanic (OR = 5.09; 95% CI = 4.29–6.03) pregnant women than for White non‐Hispanic (OR = 3.28; 95% CI = 2.65–4.06) women. Discussion Decreasing rates of PTD and SMM and increasing rates of stillbirth among deliveries without COVID‐19 were small but suggest indirect effects of the pandemic on maternal outcomes. Among pregnant women with COVID‐19 at delivery, adverse effects, particularly SMM for API and Hispanic women, underscore the importance of addressing health disparities.

Fine-grained activity recognition enables explainable analysis of procedures for skill assessment, autonomy, and error detection in robot-assisted surgery. However, existing recognition models suffer from the limited availability of annotated datasets with both kinematic and video data and an inability to generalize to unseen subjects and tasks. Kinematic data from the surgical robot is particularly critical for safety monitoring and autonomy, as it is unaffected by common camera issues such as occlusions and lens contamination. We leverage an aggregated dataset of six dry-lab surgical tasks from a total of 28 subjects to train activity recognition models at the gesture and motion primitive (MP) levels and for separate robotic arms using only kinematic data. The models are evaluated using the LOUO (Leave-One-User-Out) and our proposed LOTO (Leave-One-Task-Out) cross validation methods to assess their ability to generalize to unseen users and tasks respectively. Gesture recognition models achieve higher accuracies and edit scores than MP recognition models. But, using MPs enables the training of models that can generalize better to unseen tasks. Also, higher MP recognition accuracy can be achieved by training separate models for the left and right robot arms. For task-generalization, MP recognition models perform best if trained on similar tasks and/or tasks from the same dataset.

Masks have remained an important mitigation strategy in the fight against COVID-19 due to their ability to prevent the transmission of respiratory droplets between individuals. In this work, we provide a comprehensive quantitative analysis of the impact of mask-wearing. To this end, we propose a novel agent-based model of viral spread on networks where agents may either wear no mask or wear one of several types of masks with different properties (e.g., cloth or surgical). We derive analytical expressions for three key epidemiological quantities: The probability of emergence, the epidemic threshold, and the expected epidemic size. In particular, we show how the aforementioned quantities depend on the structure of the contact network, viral transmission dynamics, and the distribution of the different types of masks within the population. Through extensive simulations, we then investigate the impact of different allocations of masks within the population and tradeoffs between the outward efficiency and inward efficiency of the masks. Interestingly, we find that masks with high outward efficiency and low inward efficiency are most useful for controlling the spread in the early stages of an epidemic, while masks with high inward efficiency but low outward efficiency are most useful in reducing the size of an already large spread. Last, we study whether degree-based mask allocation is more effective in reducing the probability of epidemic as well as epidemic size compared to random allocation. The result echoes the previous findings that mitigation strategies should differ based on the stage of the spreading process, focusing on source control before the epidemic emerges and on self-protection after the emergence.