Browsing by Author "Qiu, Q"

Now showing 1 - 3 of 3
  • Thumbnail Image
    ItemOpen Access
    DCFNet: Deep Neural Network with Decomposed Convolutional Filters
    (35th International Conference on Machine Learning, ICML 2018, 2018-01-01) Qiu, Q; Cheng, X; Calderbank, R; Sapiro, G
    ©35th International Conference on Machine Learning, ICML 2018.All Rights Reserved. Filters in a Convolutional Neural Network (CNN) contain model parameters learned from enormous amounts of data. In this paper, we suggest to decompose convolutional filters in CNN as a truncated expansion with pre-fixed bases, namely the Decomposed Convolutional Filters network (DCFNet), where the expansion coefficients remain learned from data. Such a structure not only reduces the number of trainable parameters and computation, but also imposes filter regularity by bases truncation. Through extensive experiments, we consistently observe that DCFNet maintains accuracy for image classification tasks with a significant reduction of model parameters, particularly with Fourier-Bessel (FB) bases, and even with random bases. Theoretically, we analyze the representation stability of DCFNet with respect to input variations, and prove representation stability under generic assumptions on the expansion coefficients. The analysis is consistent with the empirical observations.
  • Thumbnail Image
    ItemOpen Access
    Postseismic coastal development in Aceh, Indonesia - Field observations and numerical modeling
    (Marine Geology, 2017-10-01) Monecke, K; Meilianda, E; Walstra, DJ; Hill, EM; McAdoo, BG; Qiu, Q; Storms, JEA; Masputri, AS; Mayasari, CD; Nasir, M; Riandi, I; Setiawan, A; Templeton, CK
    We model postseismic changes to the shoreline of West Aceh, Indonesia, a region largely affected by the December 2004 Sumatra-Andaman earthquake and ensuing Indian Ocean tsunami, using a cross-shore morphodynamic model. Subsidence of 0.5–1.0 m and tsunami scouring during the 2004 event caused the complete destruction of the beach and the landward displacement of the western coast of Aceh by an average of 110 m. Comparing a series of satellite images and topographic surveys, we reconstruct the build-up of a new beach ridge along a 6 km long stretch of coastline in the years following the event. We then use the cross-shore model UNIBEST-TC developed for a wave-dominated sandy shoreline to determine the controlling factors of shoreline recovery. Input parameters include bathymetric data measured in 2015, grain size characteristics of offshore sediment samples, modeled wave data, tidal elevations from a nearby tide-gauge station as well as measured and modeled postseismic uplift data. After establishing a cross-shore profile in equilibrium with the prevailing hydrodynamic conditions, we simulate the post-tsunami recovery, the effect of the monsoon seasons, as well as the influence of postseismic land level changes for up to 10 years and compare them to the observed coastal development. Our modeling results indicate that the recovery of the western Acehnese shoreline after the 2004 tsunami was quick with littoral sediment transport normalizing to pre-tsunami conditions within two to four years following the event. However, field data shows that the shoreline stabilized 50–90 m landward of its pre-2004 tsunami position, most likely due to the build-up of a prominent higher beach ridge in response to coseismic subsidence. Observed variability in shoreline position in the order of a few tens of meters since 2009 can be attributed to seasonal wave climate variability related to the monsoon cycle. The effect of postseismic uplift on shoreline position is small and in the order of only a few meters over 10 years, which is 3 to 5 times smaller than long-term coastal progradation rates that are driven by abundant sediment supply to the littoral zone. This overall progradational trend will promote preservation of seismically modified beach ridges, which can serve as paleoseismic indicators.
  • Thumbnail Image
    ItemOpen Access
    RotDCF: Decomposition of Convolutional Filters for Rotation-Equivariant Deep Networks.
    (CoRR, 2018) Cheng, X; Qiu, Q; Calderbank, R; Sapiro, G
    Explicit encoding of group actions in deep features makes it possible for convolutional neural networks (CNNs) to handle global deformations of images, which is critical to success in many vision tasks. This paper proposes to decompose the convolutional filters over joint steerable bases across the space and the group geometry simultaneously, namely a rotation-equivariant CNN with decomposed convolutional filters (RotDCF). This decomposition facilitates computing the joint convolution, which is proved to be necessary for the group equivariance. It significantly reduces the model size and computational complexity while preserving performance, and truncation of the bases expansion serves implicitly to regularize the filters. On datasets involving in-plane and out-of-plane object rotations, RotDCF deep features demonstrate greater robustness and interpretability than regular CNNs. The stability of the equivariant representation to input variations is also proved theoretically under generic assumptions on the filters in the decomposed form. The RotDCF framework can be extended to groups other than rotations, providing a general approach which achieves both group equivariance and representation stability at a reduced model size.