skip to main content
Explore Scholarly Publications and Datasets in the NSF-PAR

  • Home
  • Search Results
  • Page 1 of 7

Search for: All records

Creators/Authors contains: "Wu, C."

Note: When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external site maintained by the publisher. Some full text articles may not yet be available without a charge during the embargo (administrative interval).
What is a DOI Number?

Some links on this page may take you to non-federal websites. Their policies may differ from this site.

  1. ; ; ; ; ( , The Tenth International Conference on Learning Representations)
    In the Mixup training paradigm, a model is trained using convex combinations of data points and their associated labels. Despite seeing very few true data points during training, models trained using Mixup seem to still minimize the original empirical risk and exhibit better generalization and robustness on various tasks when compared to standard training. In this paper, we investigate how these benefits of Mixup training rely on properties of the data in the context of classification. For minimizing the original empirical risk, we compute a closed form for the Mixup-optimal classification, which allows us to construct a simple dataset onmore »which minimizing the Mixup loss can provably lead to learning a classifier that does not minimize the empirical loss on the data. On the other hand, we also give sufficient conditions for Mixup training to also minimize the original empirical risk. For generalization, we characterize the margin of a Mixup classifier, and use this to understand why the decision boundary of a Mixup classifier can adapt better to the full structure of the training data when compared to standard training. In contrast, we also show that, for a large class of linear models and linearly separable datasets, Mixup training leads to learning the same classifier as standard training.« less
    Free, publicly-accessible full text available July 1, 2023
  2. ; ; ; ; ; ; ; ; ; ; et al ( , Physical Review C)
    Free, publicly-accessible full text available February 1, 2023
  3. ; ; ; ; ; ; ; ; ( , Physical Review D)
    Free, publicly-accessible full text available September 1, 2022
  4. ; ( , IEEE Conference on Computer Vision and Pattern Recognition)
    Accepted Manuscript Full Text Available
  5. ; ; ; ; ( , International Conference on Learning Representations (ICLR))
    Accepted Manuscript Full Text Available
  6. ; ; ; ; ; ( , HotMobile)
    Autonomous vehicles are predicted to dominate the transportation industry in the foreseeable future. Safety is one of the major chal- lenges to the early deployment of self-driving systems. To ensure safety, self-driving vehicles must sense and detect humans, other vehicles, and road infrastructure accurately, robustly, and timely. However, existing sensing techniques used by self-driving vehicles may not be absolutely reliable. In this paper, we design REITS, a system to improve the reliability of RF-based sensing modules for autonomous vehicles. We conduct theoretical analysis on possible failures of existing RF-based sensing systems. Based on the analysis, REITS adopts a multi-antenna design,more »which enables constructive blind beamforming to return an enhanced radar signal in the incident direction. REITS can also let the existing radar system sense identifi- cation information by switching between constructive beamforming state and destructive beamforming state. Preliminary results show that REITS improves the detection distance of a self-driving car radar by a factor of 3.63.« less
    Full Text Available 
  7. ; ; ; ( , Proceedings of the 38th International Conference on Machine Learning)
    Accepted Manuscript Full Text Available
  8. ; ; ; ; ( , Journal of the Meteorological Society of Japan)
    This study investigates the synoptic-scale flows associated with extreme rainfall systems over the Asian–Australian monsoon region (90 – 160°E and 12°S – 27°N). On the basis of the statistics of the 17-year Precipitation Radar observations from Tropical Rainfall Measurement Mission, a total of 916 extreme systems, with both the horizontal size and maximum rainfall intensity exceeding the 99.9th percentiles of the tropical rainfall systems, are identified over this region. The synoptic wind pattern and rainfall distribution surrounding each system are classified into four major types: vortex, coastal, coastal with vortex, and none of above, with each accounting for 44, 29, 7, and 20 %,more »respectively. The vortex type occurs mainly over the off-equatorial areas in boreal summer. The coast-related types show significant seasonal variations in their occurrence, with high frequency in the Bay of Bengal in boreal summer and on the west side of Borneo and Sumatra in boreal winter. The none-of-the-above type occurs mostly over the open ocean and in boreal winter; these events are mainly associated with the cold surge events. The environment analysis shows that coast-related extremes in the warm season are found within the areas where high total water vapor and low-level vertical wind shear occur frequently. Despite the different synoptic environments, these extremes show a similar internal structure, with broad stratiform and wide convective core (WCC) rain. Furthermore, the maximum rain rate is located mostly over the convective area, near the convective–stratiform boundary in the system. Our results highlight the critical role of the strength and direction of synoptic flows in the generation of extreme rainfall systems near coastal areas. With the enhancement of the lowlevel vertical wind shear and moisture by the synoptic flow, the coastal convection triggered diurnally has a higher chance to organize into mesoscale convective systems and hence a higher probability to produce extreme rainfall.« less
    Accepted Manuscript Full Text Available
  9. ; ; ; ; ; ; ; ; ; ; et al ( , Physical Review C)
    Full Text Available 
  10. ; ; ; ; ; ; ; ; ; ; et al ( , Physical Review C)
    Full Text Available