published journal article
conference paper
conference paper
published journal article
By analyzing seven-day travel logs from Los Angeles during 2011–2013, we contribute to the understanding of intrapersonal day-to-day travel variability (IDTV) in relation to socio-demographic and land-use characteristics and the implication of travel survey duration for travel parameter estimates. Our main sample included 2,395 person-days from 352 individual participants in 219 households. Our analytical methods included linear regressions and random sampling experiments. Our Feasible Generalized Least Squares (FGLS) regression models revealed that many factors significantly influenced IDTV, such as gender, age, income, and household type. However, the observed socio-demographic and land-use characteristics could only explain a small portion of IDTV. The random sampling experiments enabled us to contrast travel variables measured from the seven-day master sample with those from subsamples of a shorter period (one to six days). The “optimal” duration for a travel survey may depend on the specific travel variables measured, and we provide evidence that studies of transit and non-motorized travel will require longer surveys than studies of car travel. In conclusion, the conventional one-day approach is likely to produce imprecise parameter estimates due to the intrapersonal day-to-day travel variability. We recommend that transportation professionals and policy makers consider shifting from the conventional one-day approach toward a multi-day approach. Surveys that focus on the modes of walking, biking, and transit should consider data collection for at least seven days.
Preprint Journal Article
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 workloads 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 O(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.
published journal article
This paper separately analyzes the impacts of weather and seasonality on commute mode choice using a stated-preference (SP) survey. The studyâ??s twin objectives include quantifying the relative attractiveness of a flexible transit mode under adverse weather and analyzing the heterogenous impacts of seasonality and weather on different segments of the population. The survey was implemented during summer and winter months to analyze the impact of seasonality on commute mode choice. To analyze the impact of weather on commute mode choice, the survey randomly assigns respondents to a good or bad weather scenario. The three SP mode choice options are fixed-route transit, personal car, and a hypothetical flexible transit mode. The flexible transit mode allows travelers to wait for a transit vehicle at their origin point â?? an especially attractive feature on cold and/or snowy/rainy days. However, like with existing dial-a-ride services, travelers using the flexible transit service may experience detours between their pickup and drop-off locations. Several panel mixed-logit choice models that incorporate weather and seasonality were estimated using the SP survey data. The panel mixed-logit model form captures taste heterogeneity across respondents and correlation across multiple SP choice experiments from a single respondent.Model estimation results indicate that relative to good weather, summer days, respondents were significantly more likely to choose traditional and flexible transit on good weather, winter days, and flexible transit on bad weather, winter days. Furthermore, the results show that the impacts of weather and seasonality on commute mode choice vary across the population. Most of the respondents had a lower (higher) propensity to choose the car (flexible transit) mode under bad weather than good weather. Conversely, car commuters, bicycle commuters, and baby boomers had a much higher propensity to choose the car mode under bad weather than good weather. Additionally, the results show that millennials had a much lower propensity to choose the car mode in the winter than the summer, whereas seasonality had little impact on the propensity of non-millennials to choose the car mode. The paper discusses the implications of these findings on the design of transit, flexible transit, and emerging mobility services, as well as the management of transportation systems and dissemination of information regarding alternative transport modes during adverse weather events.
published journal article
This paper analyzes an irreversible “where-and-when” investment decision, in which a government must decide not only when to invest in income-increasing infrastructure but also where to make the investment, doing so under imperfect observability of the investment gains. The two models considered in the paper differ in the source of the imperfection. In the signal model, the imperfection comes from imperfect observability of initial income gains from the investment, while in the option model, it comes from the stochastic nature of the income gains in the second period. In addition to providing the first treatment of this type of problem, the analysis shows that the influences of underlying parameters on whether or not the government waits to invest are similar in the two models. (C) 2015 Published by Elsevier B.V.
conference paper
conference paper
published journal article
Research suggests that gender influences attitudes toward both the environment and safety. While pro-environmental attitudes might encourage transit use, safety concerns might discourage transit use if the transit environment is perceived as unsafe. To quantitatively examine how gender, environmental beliefs, and safety concerns jointly affect transit use, results are analyzed from a longitudinal quasi-experimental study which conducted pre- and post-opening travel surveys near a new light rail transit service in Los Angeles. It is found that the influence of safety concerns on transit use is more prominent than that of environmental attitudes, particularly for women. Living closer to a new light rail transit station correlates with an increase in train ridership. This effect, however, is significantly lower for women. The results suggest that to foster transit use, reducing personal safety concerns related to transit may be more effective than increasing public awareness of transportation-related environmental issues, especially for attracting female riders.