skip to main content

Search for: All records

Creators/Authors contains: "Dong, Yue"

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. Free, publicly-accessible full text available November 22, 2022
  2. Free, publicly-accessible full text available December 28, 2022
  3. Free, publicly-accessible full text available September 1, 2022
  4. Despite considerable advancements with deep neural language models (LMs), neural text generation still suffers from degeneration: the generated text is repetitive, generic, selfcontradictory, and often lacks commonsense. Our analyses on sentence-level attention patterns in LMs reveal that neural degeneration may be associated with insufficient learning of task-specific characteristics by the attention mechanism. This finding motivates onthe-fly attention modulation1– a simple but effective method that enables the injection of priors into attention computation during inference. Automatic and human evaluation results on three text generation benchmarks demonstrate that attention modulation helps LMs generate text with enhanced fluency, creativity, and commonsense reasoning, inmore »addition to significantly reduce sentence-level repetition.« less
  5. Abstract Radiative feedbacks depend on the spatial patterns of sea surface temperature (SST) and thus can change over time as SST patterns evolve—the so-called pattern effect. This study investigates intermodel differences in the magnitude of the pattern effect and how these differences contribute to the spread in effective equilibrium climate sensitivity (ECS) within CMIP5 and CMIP6 models. Effective ECS in CMIP5 estimated from 150-yr-long abrupt4×CO2 simulations is on average 10% higher than that estimated from the early portion (first 50 years) of those simulations, which serves as an analog for historical warming; this difference is reduced to 7% on averagemore »in CMIP6. The (negative) net radiative feedback weakens over the course of the abrupt4×CO2 simulations in the vast majority of CMIP5 and CMIP6 models, but this weakening is less dramatic on average in CMIP6. For both ensembles, the total variance in the effective ECS is found to be dominated by the spread in radiative response on fast time scales, rather than the spread in feedback changes. Using Green’s functions derived from two AGCMs shows that the spread in feedbacks on fast time scales may be primarily due to differences in atmospheric model physics, whereas the spread in feedback evolution is primarily governed by differences in SST patterns. Intermodel spread in feedback evolution is well explained by differences in the relative warming in the west Pacific warm-pool regions for the CMIP5 models, but this relation fails to explain differences across the CMIP6 models, suggesting that a stronger sensitivity of extratropical clouds to surface warming may also contribute to feedback changes in CMIP6.« less
  6. We present a new weakly supervised learning-based method for generating novel category-specific 3D shapes from unoccluded image collections. Our method is weakly supervised and only requires silhouette annotations from unoccluded, category-specific objects. Our method does not require access to the object's 3D shape, multiple observations per object from different views, intra-image pixel correspondences, or any view annotations. Key to our method is a novel multi-projection generative adversarial network (MP-GAN) that trains a 3D shape generator to be consistent with multiple 2D projections of the 3D shapes, and without direct access to these 3D shapes. This is achieved through multiple discriminatorsmore »that encode the distribution of 2D projections of the 3D shapes seen from a different views. Additionally, to determine the view information for each silhouette image, we also train a view prediction network on visualizations of 3D shapes synthesized by the generator. We iteratively alternate between training the generator and training the view prediction network. We validate our multi-projection GAN on both synthetic and real image datasets. Furthermore, we also show that multi-projection GANs can aid in learning other high-dimensional distributions from lower dimensional training datasets, such as material-class specific spatially varying reflectance properties from images.« less