Spatio-Temporal Inference Transformer Network for Video Inpainting

Video inpainting aims to complete in a visually pleasing way the missing regions in video frames. Video inpainting is an exciting task due to the variety of motions across different frames. The existing methods usually use attention models to inpaint videos by seeking the damaged content from other frames. Nevertheless, these methods suffer due to irregular attention weight from spatio-temporal dimensions, thus giving rise to artifacts in the inpainted video. To overcome the above problem, Spatio-Temporal Inference Transformer Network (STITN) has been proposed. The STITN aligns the frames to be inpainted and concurrently inpaints all the frames, and a spatio-temporal adversarial loss function improves the STITN. Our method performs considerably better than the existing deep learning approaches in quantitative and qualitative evaluation.

Generalized Temporal Inference via Planning

This paper introduces the Temporal Inference Problem (TIP), a general formulation for a family of inference problems that reason about the past, present or future state of some observed agent. A TIP builds on the models of an actor and of an observer. Observations of the actor are gathered at arbitrary times and a TIP encodes hypothesis on unobserved segments of the actor's trajectory. Regarding the last observation as the present time, a TIP enables to hypothesize about the past trajectory, future trajectory or current state of the actor. We use LTL as a language for expressing hypotheses and reduce a TIP to a planning problem which is solved with an off-the-shelf classical planner. The output of the TIP is the most likely hypothesis, the minimal cost trajectory under the assumption that the actor is rational. Our proposal is evaluated on a wide range of TIP instances defined over different planning domains.

Inference from longitudinal laboratory tests characterizes temporal evolution of COVID-19-associated coagulopathy (CAC)

Temporal inference from laboratory testing results and triangulation with clinical outcomes extracted from unstructured electronic health record (EHR) provider notes is integral to advancing precision medicine. Here, we studied 246 SARS-CoV-2 PCR-positive (COVIDpos) patients and propensity-matched 2460 SARS-CoV-2 PCR-negative (COVIDneg) patients subjected to around 700,000 lab tests cumulatively across 194 assays. Compared to COVIDneg patients at the time of diagnostic testing, COVIDpos patients tended to have higher plasma fibrinogen levels and lower platelet counts. However, as the infection evolves, COVIDpos patients distinctively show declining fibrinogen, increasing platelet counts, and lower white blood cell counts. Augmented curation of EHRs suggests that only a minority of COVIDpos patients develop thromboembolism, and rarely, disseminated intravascular coagulopathy (DIC), with patients generally not displaying platelet reductions typical of consumptive coagulopathies. These temporal trends provide fine-grained resolution into COVID-19 associated coagulopathy (CAC) and set the stage for personalizing thromboprophylaxis.

Internet of Things (IoT)-Enabled Elderly Fall Verification, Exploiting Temporal Inference Models in Smart Homes

Everyday life of the elderly and impaired population living in smart homes is challenging because of possible accidents that may occur due to daily activities. In such activities, persons often lean over (to reach something) and, if they not cautious, are prone to falling. To identify fall incidents, which could stochastically cause serious injuries or even death, we propose specific temporal inference models; namely, CM-I and CM-II. These models can infer a fall incident based on classification methods by exploiting wearable Internet of Things (IoT) altimeter sensors adopted by seniors. We analyzed real and synthetic data of fall and lean over incidents to test the proposed models. The results are promising for incorporating such inference models to assist healthcare for fall verification of seniors in smart homes. Specifically, the CM-II model achieved a prediction accuracy of 0.98, which is the highest accuracy when compared to other models in the literature under the McNemar’s test criterion. These models could be incorporated in wearable IoT devices to provide early warning and prediction of fall incidents to clinical doctors.