Search for: All records

We consider transferability estimation, the prob- lem of estimating how well deep learning models transfer from a source to a target task. We focus on regression tasks, which received little previous attention, and propose two simple and computa- tionally efficient approaches that estimate trans- ferability based on the negative regularized mean squared error of a linear regression model. We prove novel theoretical results connecting our ap- proaches to the actual transferability of the optimal target models obtained from the transfer learning process. Despite their simplicity, our approaches significantly outperform existing state-of-the-art regression transferability estimators in both accu- racy and efficiency. On two large-scale keypoint re- gression benchmarks, our approaches yield 12% to 36% better results on average while being at least 27% faster than previous state-of-the-art methods. 

Many globally distributed data stores need to replicate data across large geographic distances. Since synchronously replicating data across such distances is slow, those systems with high consistency requirements often geo-partition data and direct all linearizable requests to the primary region of the accessed data. This significantly improves performance for workloads where most transactions access data close to where they originate from. However, supporting serializable multi-geo-partition transactions is a challenge, and they often degrade the performance of the whole system. This becomes even more challenging when they conflict with single-partition requests, where optimistic protocols lead to high numbers of aborts, and pessimistic protocols lead to high numbers of distributed deadlocks. In this paper, we describe the design of concurrency control and deadlock resolution protocols, built within a practical, complete implementation of a geographically replicated database system called Detock, that enables processing strictly-serializable multi-region transactions with near-zero performance degradation at extremely high conflict and order of magnitude higher throughput relative to state-of-the art geo-replication approaches, while improving latency by up to a factor of 5. 

Abstract A new method that automatically determines the modality of an observed particle size distribution (PSD) and the representation of each mode as a gamma function was used to characterize data obtained during the High Altitude Ice Crystals and High Ice Water Content (HAIC-HIWC) project based out of Cayenne, French Guiana, in 2015. PSDs measured by a 2D stereo probe and a precipitation imaging probe for particles with maximum dimension ( D max ) > 55 μ m were used to show how the gamma parameters varied with environmental conditions, including temperature ( T ) and convective properties such as cloud type, mesoscale convective system (MCS) age, distance away from the nearest convective peak, and underlying surface characteristics. Four kinds of modality PSDs were observed: unimodal PSDs and three types of multimodal PSDs (Bimodal1 with breakpoints 100 ± 20 μ m between modes, Bimodal2 with breakpoints 1000 ± 300 μ m, and Trimodal PSDs with two breakpoints). The T and ice water content (IWC) are the most important factors influencing the modality of PSDs, with the frequency of multimodal PSDs increasing with increasing T and IWC. An ellipsoid of equally plausible solutions in ( N o – λ–μ ) phase space is defined for each mode of the observed PSDs for different environmental conditions. The percentage overlap between ellipsoids was used to quantify the differences between overlapping ellipsoids for varying conditions. The volumes of the ellipsoid decrease with increasing IWC for most cases, and ( N o – λ–μ ) vary with environmental conditions related to distribution of IWC. HIWC regions are dominated by small irregular ice crystals and columns. The parameters ( N o – λ–μ ) in each mode exhibit mutual dependence. 

Abstract During near-0°C surface conditions, diverse precipitation types (p-types) are possible, including rain, drizzle, freezing rain, freezing drizzle, ice pellets, wet snow, snow, and snow pellets. Near-0°C precipitation affects wide swaths of the United States and Canada, impacting aviation, road transportation, power generation and distribution, winter recreation, ecology, and hydrology. Fundamental challenges remain in observing, diagnosing, simulating, and forecasting near-0°C p-types, particularly during transitions and within complex terrain. Motivated by these challenges, the field phase of the Winter Precipitation Type Research Multi-scale Experiment (WINTRE-MIX) was conducted from 1 February – 15 March 2022 to better understand how multiscale processes influence the variability and predictability of p-type and amount under near-0°C surface conditions. WINTRE-MIX took place near the US / Canadian border, in northern New York and southern Quebec, a region with plentiful near-0°C precipitation influenced by terrain. During WINTRE-MIX, existing advanced mesonets in New York and Quebec were complemented by deployment of: (1) surface instruments, (2) the National Research Council Convair-580 research aircraft with W- and X-band Doppler radars and in situ cloud and aerosol instrumentation, (3) two X-band dual-polarization Doppler radars and a C-band dual-polarization Doppler radar from University of Illinois, and (4) teams collecting manual hydrometeor observations and radiosonde measurements. Eleven intensive observing periods (IOPs) were coordinated. Analysis of these WINTRE-MIX IOPs is illuminating how synoptic dynamics, mesoscale dynamics, and microscale processes combine to determine p-type and its predictability under near-0°C conditions. WINTRE-MIX research will contribute to improving nowcasts and forecasts of near-0°C precipitation through evaluation and refinement of observational diagnostics and numerical forecast models.