conference paper
published journal article
This study explores the growth of electronic home shopping in terms of likely transportation and communication interactions. Although opportunities exist to shop from home today, most consumers initiate travel trips to stores or markets. Widespread use of automobiles has facilitated the retailing configurations we know today but the development of new electronic networks could change this. This study establishes a baseline to explore shopping activities using two-day travel activity data from a large U.S. metropolitan area. It is found that people who telework from home today spend more time engaged in shopping activities than other workers. Potentially, their saved work travel is converted into new trips. In the future, saved shopping travel might be converted into other types of travel, and modelling results show that for busy working women, there is a latent demand for maintenance-related activities. The study results suggest that electronic home shopping will bring into play complex interactions between communications and transportation.
conference paper
published journal article
published journal article
published journal article
Background: Wildfire smoke is a leading driver of acute exposure to PM2.5 in the American West and a significant contributor to chronic pollution exposure in immediately impacted and further downwind areas. Exposure to wildfire smoke is linked to acute respiratory morbidity and all-cause mortality, yet little is known about chronic effects of repeated, elevated exposures. Inclusion of wildfire smoke in air quality models for health effects research is important for improving accuracy of the overall models and understanding the specific and independent effects of wildfire smoke relative to the entire pollution mixture. However, the nature of smoke, including high spatial variability and the three-dimensional structure of smoke plumes, presents challenges for the accurate representation of wildfire smoke in health research applications. In this presentation, we will discuss exposure and health research applications that use explicit representations of wildfire smoke to improve exposure estimates. Methods:We used dispersion modeling of wildfire smoke to predict ground-level concentrations and support a deep learning ensemble model of PM2.5 over California for 2008-2017. Smoke emissions were modeled using satellite detections of wildfires and a database of emissions related to fire radiative energy. Emissions were dispersed using a fine-scale meteorological data set. We assessed the ability of the model to reproduce spatial and temporal patterns of wildfire smoke using visual satellite imagery of smoke and correlations with ground-based monitoring, respectively. Results: Our results show that the inclusion of smoke dispersion surfaces produces accurate predictions of PM2.5 concentrations in wildfire smoke conditions. Conclusions: This work highlights the importance of incorporating wildfire smoke data sources into exposure assessments, and it indicates new directions for use of wildfire smoke data in health research. The extension of our smoke modeling through the high-smoke years between 2018 and 2021 will support important additional research on wildfire smoke exposure health impacts. Keywords: Wildfire
published journal article
published journal article
published journal article
conference paper
Emerging multi-model workloads with heavy models like recent large language models significantly increased the compute and memory demands on hardware. To address such increasing demands, designing a scalable hardware architecture became a key problem. Among recent solutions, the 2.5D silicon interposer multi-chip module (MCM)-based AI accelerator has been actively explored as a promising scalable solution due to their significant benefits in the low engineering cost and composability. However, previous MCM accelerators are based on homogeneous architectures with fixed dataflow, which encounter major challenges from highly heterogeneous multi-model work-loads due to their limited workload adaptivity. Therefore, in this work, we explore the opportunity in the heterogeneous dataflow MCM AI accelerators. We identify the scheduling of multi-model workload on heterogeneous dataflow MCM AI accelerator is an important and challenging problem due to its significance and scale, which reaches mathbfO(10^56) even for a two-model workload on 6×6 chiplets. We develop a set of heuristics to navigate the huge scheduling space and codify them into a scheduler, SCAR, with advanced techniques such as inter-chiplet pipelining. Our evaluation on ten multi-model workload scenarios for datacenter multitenancy and AR/VR use-cases has shown the efficacy of our approach, achieving on average 27.6% and 29.6% less energy-delay product (EDP) for the respective applications settings compared to homogeneous baselines.