arXiv subjects

Yixiao Zhang

Publications and source records attributed to Yixiao Zhang.

62 records · Page 4Linked to original sources

arXiv2021

Calibrating Concepts and Operations: Towards Symbolic Reasoning on Real Images

While neural symbolic methods demonstrate impressive performance in visual question answering on synthetic images, their performance suffers on real images. We identify that the long-tail distribution of visual concepts and unequal importance of reasoning steps in real data are the two key obstacles that limit the models' real-world potentials. To address these challenges, we propose a new paradigm, Calibrating Concepts and Operations (CCO), which enables neural symbolic models to capture underlying data characteristics and to reason with hierarchical importance. Specifically, we introduce an executor with learnable concept embedding magnitudes for handling distribution imbalance, and an operation calibrator for highlighting important operations and suppressing redundant ones. Our experiments show CCO substantially boosts the performance of neural symbolic methods on real images. By evaluating models on the real world dataset GQA, CCO helps the neural symbolic method NSCL outperforms its vanilla counterpart by 9.1% (from 47.0% to 56.1%); this result also largely reduces the performance gap between symbolic and non-symbolic methods. Additionally, we create a perturbed test set for better understanding and analyzing model performance on real images. Code is available at https://github.com/Lizw14/CaliCO.git .

cs.CV↗

arXiv2020

How does Background Air Pressure Influence the Inner Edge of the Habitable Zone for Tidally Locked Planets in a 3D View?

We examine the effect of varying background N2 surface pressure (labelled as pN2) on the inner edge of the habitable zone for 1:1 tidally locked planets around M dwarfs, using the three-dimensional (3D) atmospheric general circulation model (AGCM) ExoCAM. In our experiments, the rotation period is fixed when varying the stellar flux, in order to more clearly isolate the role of pN2. We find that the stellar flux threshold for the runaway greenhouse is a non-monotonous function of pN2. This is due to the competing effects of five processes: pressure broadening, heat capacity, lapse rate, relative humidity, and clouds. These competing processes increase the complexity in predicting the location of the inner edge of the habitable zone. For a slow rotation orbit of 60 Earth days, the critical stellar flux for the runaway greenhouse onset is 1700--1750, 1900--1950, and 1750--1800 W m$^{-2}$ under 0.25, 1.0, and 4.0 bar of pN2, respectively, suggesting that the magnitude of the effect of pN2 is within ~13%. For a rapid rotation orbit, the effect of varying pN2 on the inner edge is smaller, within a range of ~7%. Moreover, we show that Rayleigh scattering effect as varying pN2 is unimportant for the inner edge due to the masking effect of cloud scattering and to the strong shortwave absorption by water vapor under hot climates. Future work using AGCMs having different cloud and convection schemes and cloud-resolving models having explicit cloud and convection are required to revise this problem.

astro-ph.EP↗

arXiv2020

PIANOTREE VAE: Structured Representation Learning for Polyphonic Music

The dominant approach for music representation learning involves the deep unsupervised model family variational autoencoder (VAE). However, most, if not all, viable attempts on this problem have largely been limited to monophonic music. Normally composed of richer modality and more complex musical structures, the polyphonic counterpart has yet to be addressed in the context of music representation learning. In this work, we propose the PianoTree VAE, a novel tree-structure extension upon VAE aiming to fit the polyphonic music learning. The experiments prove the validity of the PianoTree VAE via (i)-semantically meaningful latent code for polyphonic segments; (ii)-more satisfiable reconstruction aside of decent geometry learned in the latent space; (iii)-this model's benefits to the variety of the downstream music generation.

eess.AS↗

arXiv2020

Learning Interpretable Representation for Controllable Polyphonic Music Generation

While deep generative models have become the leading methods for algorithmic composition, it remains a challenging problem to control the generation process because the latent variables of most deep-learning models lack good interpretability. Inspired by the content-style disentanglement idea, we design a novel architecture, under the VAE framework, that effectively learns two interpretable latent factors of polyphonic music: chord and texture. The current model focuses on learning 8-beat long piano composition segments. We show that such chord-texture disentanglement provides a controllable generation pathway leading to a wide spectrum of applications, including compositional style transfer, texture variation, and accompaniment arrangement. Both objective and subjective evaluations show that our method achieves a successful disentanglement and high quality controlled music generation.

cs.SD↗

arXiv2020

Small Sensitivity of the Simulated Climate of Tidally Locked Aquaplanets to Model Resolution

Tidally locked terrestrial planets around low-mass stars are the prime targets of finding potentially habitable exoplanets. Several atmospheric general circulation models have been employed to simulate their possible climates, however, model intercomparisons showed that there are large differences in the results of the models even when they are forced with the same boundary conditions. In this paper, we examine whether model resolution contributes to the differences. Using the atmospheric general circulation model ExoCAM coupled to a 50-m slab ocean, we examine three different horizontal resolutions (440 km * 550 km, 210 km * 280 km, and 50 km * 70 km in latitude and longitude) and three different vertical resolutions (26, 51, and 74 levels) under the same dynamical core and the same schemes of radiation, convection and clouds. Among the experiments, the differences are within 5 K in global-mean surface temperature and within 0.007 in planetary albedo. These differences are from cloud feedback, water vapor feedback, and the decreasing trend of relative humidity with increasing resolution. Relatively small-scale downdrafts between upwelling columns over the substellar region are better resolved and the mixing between dry and wet air parcels and between anvil clouds and their environment are enhanced as the resolution is increased. These reduce atmospheric relative humidity and high-level cloud fraction, causing a lower clear-sky greenhouse effect, a weaker cloud longwave radiation effect, and subsequently a cooler climate with increasing model resolution. Overall, the sensitivity of the simulated climate of tidally locked aquaplanets to model resolution is small.

astro-ph.EP↗

arXiv2020

C2FNAS: Coarse-to-Fine Neural Architecture Search for 3D Medical Image Segmentation

3D convolution neural networks (CNN) have been proved very successful in parsing organs or tumours in 3D medical images, but it remains sophisticated and time-consuming to choose or design proper 3D networks given different task contexts. Recently, Neural Architecture Search (NAS) is proposed to solve this problem by searching for the best network architecture automatically. However, the inconsistency between search stage and deployment stage often exists in NAS algorithms due to memory constraints and large search space, which could become more serious when applying NAS to some memory and time consuming tasks, such as 3D medical image segmentation. In this paper, we propose coarse-to-fine neural architecture search (C2FNAS) to automatically search a 3D segmentation network from scratch without inconsistency on network size or input size. Specifically, we divide the search procedure into two stages: 1) the coarse stage, where we search the macro-level topology of the network, i.e. how each convolution module is connected to other modules; 2) the fine stage, where we search at micro-level for operations in each cell based on previous searched macro-level topology. The coarse-to-fine manner divides the search procedure into two consecutive stages and meanwhile resolves the inconsistency. We evaluate our method on 10 public datasets from Medical Segmentation Decalthon (MSD) challenge, and achieve state-of-the-art performance with the network searched using one dataset, which demonstrates the effectiveness and generalization of our searched models.

cs.CV↗

arXiv2020

When Radiology Report Generation Meets Knowledge Graph

Automatic radiology report generation has been an attracting research problem towards computer-aided diagnosis to alleviate the workload of doctors in recent years. Deep learning techniques for natural image captioning are successfully adapted to generating radiology reports. However, radiology image reporting is different from the natural image captioning task in two aspects: 1) the accuracy of positive disease keyword mentions is critical in radiology image reporting in comparison to the equivalent importance of every single word in a natural image caption; 2) the evaluation of reporting quality should focus more on matching the disease keywords and their associated attributes instead of counting the occurrence of N-gram. Based on these concerns, we propose to utilize a pre-constructed graph embedding module (modeled with a graph convolutional neural network) on multiple disease findings to assist the generation of reports in this work. The incorporation of knowledge graph allows for dedicated feature learning for each disease finding and the relationship modeling between them. In addition, we proposed a new evaluation metric for radiology image reporting with the assistance of the same composed graph. Experimental results demonstrate the superior performance of the methods integrated with the proposed graph embedding module on a publicly accessible dataset (IU-RR) of chest radiographs compared with previous approaches using both the conventional evaluation metrics commonly adopted for image captioning and our proposed ones.

cs.CV↗

arXiv2019

Resonant charge-transfer in grazing collisions of H$^-$ with vicinal nanosurfaces on Cu(111), Au(100) and Pd(111) substrates: A comparative study

We compare the electron dynamics at monocrystalline Cu(111), Au(100) and Pd(111) precursor substrates with vicinal nanosteps. The unoccupied bands of a surface superlattice are populated \textit{via} the resonant charge transfer (RCT) between the surface and a H$^-$ ion that flies by at grazing angles. A quantum mechanical wave packet propagation approach is utilized to simulate the motion of the active electron where time-evolved wave packet densities are used to visualize the dynamics through the superlattice. The survived ion fraction in the reflected beam generally exhibits modulations as a function of the vicinal terrace size and shows peaks at those energies that access the image state subband dispersions. However, differences in magnitudes of the ion-survival as a function of the particular substrate selection as well as the ion-surface interaction time based on the choice of two ion-trajectories are examined. A square well model producing standing waves between the steps on the surface explains well the energies of the maxima in the ion survival probability for all the metals considered, indicating that the primary process of confinement induced subband formation is rather robust. The work may motivate measurements and applications of shallow-angle ion-scattering spectroscopy to access electronic substructures in periodically nanostructured surfaces.

physics.atom-ph↗