arXiv subjects

Ke Chen

Publications and source records attributed to Ke Chen.

At least 325 records · Page 18Linked to original sources

arXiv2018

Decoupling between thermodynamics and dynamics during rejuvenation in colloidal glasses

We rejuvenate well-aged quasi-2D binary colloidal glasses by thermal cycling, and systematically measure both the statistical responses and particle-level structural evolutions during rejuvenation. While the moduli and boson peak are continuously rejuvenated with increasing number of cycles, the mean square displacement (MSD) fluctuates significantly between different groups of thermal cycles. The decoupling between the thermodynamical and dynamical evolutions suggests different microscopic origins for different bulk properties of glasses. We find that a small fraction of structural rearrangements triggered by thermal cycling could alter the whole elastic continuum and lead to the significant thermodynamic rejuvenation, while localized defects could be activated and deactivated at the positions close to the rearrangements with significantly high mobility change and hence result in the fluctuated dynamics even with only about 10% of particles as fast regions. Our results offer a comprehensive picture for the microscopic mechanisms underlying bulk glass rejuvenation, which could be readily used to refine glass properties or to formulate further statistical theories in glassy systems.

cond-mat.soft↗

arXiv2018

Intervention Aided Reinforcement Learning for Safe and Practical Policy Optimization in Navigation

Combining deep neural networks with reinforcement learning has shown great potential in the next-generation intelligent control. However, there are challenges in terms of safety and cost in practical applications. In this paper, we propose the Intervention Aided Reinforcement Learning (IARL) framework, which utilizes human intervened robot-environment interaction to improve the policy. We used the Unmanned Aerial Vehicle (UAV) as the test platform. We built neural networks as our policy to map sensor readings to control signals on the UAV. Our experiment scenarios cover both simulation and reality. We show that our approach substantially reduces the human intervention and improves the performance in autonomous navigation, at the same time it ensures safety and keeps training cost acceptable.

cs.RO↗

arXiv2018

A Convex Geodesic Selective Model for Image Segmentation

Selective segmentation is an important application of image processing. In contrast to global segmentation in which all objects are segmented, selective segmentation is used to isolate specific objects in an image and is of particular interest in medical imaging -- permitting segmentation and review of a single organ. An important consideration is to minimise the amount of user input to obtain the segmentation; this differs from interactive segmentation in which more user input is allowed than selective segmentation. To achieve selection, we propose a selective segmentation model which uses the edge-weighted geodesic distance from a marker set as a penalty term. It is demonstrated that this edge-weighted geodesic penalty term improves on previous selective penalty terms. A convex formulation of the model is also presented, allowing arbitrary initialisation. It is shown that the proposed model is less parameter dependent and requires less user input than previous models. Further modifications are made to the edge-weighted geodesic distance term to ensure segmentation robustness to noise and blur. We can show that the overall Euler-Lagrange equation admits a unique viscosity solution. Numerical results show that the result is robust to user input and permits selective segmentations that are not possible with other models.

math.NA↗

arXiv2018

Multigrid Algorithm Based on Hybrid Smoothers for Variational and Selective Segmentation Models

Automatic segmentation of an image to identify all meaningful parts is one of the most challenging as well as useful tasks in a number of application areas. This is widely studied. Selective segmentation, less studied, aims to use limited user specified information to extract one or more interesting objects (instead of all objects). Constructing a fast solver remains a challenge for both classes of model. However our primary concern is on selective segmentation. In this work, we develop an effective multigrid algorithm, based on a new non-standard smoother to deal with non-smooth coefficients, to solve the underlying partial differential equations (PDEs) of a class of variational segmentation models in the level set formulation. For such models, non-smoothness (or jumps) is typical as segmentation is only possible if edges (jumps) are present. In comparison with previous multigrid methods which were shown to produce an acceptable {\it mean} smoothing rate for related models, the new algorithm can ensure a small and {\it global} smoothing rate that is a sufficient condition for convergence. Our rate analysis is by Local Fourier Analysis and, with it, we design the corresponding iterative solver, improving on an ineffective line smoother. Numerical tests show that the new algorithm outperforms multigrid methods based on competing smoothers.

math.NA↗

arXiv2018

Orientational Order in Disordered Colloidal Suspensions

Exploring structural order in disordered systems including liquids and glasses is an intriguing but challenging issue in condensed matter physics. Here we construct a new parameter based on the angular distribution function of particles and show that this new orientational order has significantly higher correlation with dynamic heterogeneity compared to a translational parameter based on the radial distribution functions in colloidal glasses. The gradual development of orientational and translational order in supercooled liquids shows that the higher correlation between orientational order and dynamics comes from the onset of glass transition and the orientational order would dominate the glassy dynamics after a simple liquid is considerably supercooled. Our results suggest that orientational order reflects the formation of amorphous order during glass transition while translational order is mainly a result of density increase.

cond-mat.soft↗

arXiv2018

Online learning in optical tomography: a stochastic approach

We study the inverse problem of radiative transfer equation (RTE) using stochastic gradient descent method (SGD) in this paper. Mathematically, optical tomography amounts to recovering the optical parameters in RTE using the incoming-outgoing pair of light intensity. We formulate it as a PDE-constraint optimization problem, where the mismatch of computed and measured outgoing data is minimized with same initial data and RTE constraint. The memory and computation cost it requires, however, is typically prohibitive, especially in high dimensional space. Smart iterative solvers that only use partial information in each step is called for thereafter. Stochastic gradient descent method is an online learning algorithm that randomly selects data for minimizing the mismatch. It requires minimum memory and computation, and advances fast, therefore perfectly serves the purpose. In this paper we formulate the problem, in both nonlinear and its linearized setting, apply SGD algorithm and analyze the convergence performance.

math.OC↗

arXiv2018

Learning to Play General Video-Games via an Object Embedding Network

Deep reinforcement learning (DRL) has proven to be an effective tool for creating general video-game AI. However most current DRL video-game agents learn end-to-end from the video-output of the game, which is superfluous for many applications and creates a number of additional problems. More importantly, directly working on pixel-based raw video data is substantially distinct from what a human player does.In this paper, we present a novel method which enables DRL agents to learn directly from object information. This is obtained via use of an object embedding network (OEN) that compresses a set of object feature vectors of different lengths into a single fixed-length unified feature vector representing the current game-state and fulfills the DRL simultaneously. We evaluate our OEN-based DRL agent by comparing to several state-of-the-art approaches on a selection of games from the GVG-AI Competition. Experimental results suggest that our object-based DRL agent yields performance comparable to that of those approaches used in our comparative study.

cs.LG↗

arXiv2018

Comparison between Grating Imaging and Transient Grating Techniques on Measuring Carrier Diffusion in Semiconductor

Optical grating technique, where optical gratings are generated via light inference, has been widely used to measure charge carrier and phonon transport in semiconductors. In this paper, compared are three types of transient optical grating techniques: transient grating diffraction, transient grating heterodyne, and grating imaging, by utilizing them to measure carrier diffusion coefficient in a GaAs/AlAs superlattice. Theoretical models are constructed for each technique to extract the carrier diffusion coefficient, and the results from all three techniques are consistent. Our main findings are: (1) the transient transmission change obtained from transient grating heterodyne and grating imaging techniques are identical, even these two techniques originate from different detection principles; and (2) By adopting detection of transmission change (heterodyne amplification) instead of pure diffraction, the grating imaging technique (transient grating heterodyne) has overwhelming advantage in signal intensity than the transient grating diffraction, with a signal intensity ratio of 315:1 (157:1).

cond-mat.mtrl-sci↗

arXiv2018

Accelerated Carrier Recombination by Grain Boundary/Edge Defects in MBE Grown Transition Metal Dichalcogenides

Defect-carrier interaction in transition metal dichalcogenides (TMDs) play important roles in carrier relaxation dynamics and carrier transport, which determines the performance of electronic devices. With femtosecond laser time-resolved spectroscopy, we investigated the effect of grain boundary/edge defects on the ultrafast dynamics of photoexcited carrier in MBE grown MoTe2 and MoSe2. We found that, comparing with exfoliated samples, carrier recombination rate in MBE grown samples accelerates by about 50 times. We attribute this striking difference to the existence of abundant grain boundary/edge defects in MBE grown samples, which can serve as effective recombination centers for the photoexcited carriers. We also observed coherent acoustic phonons in both exfoliated and MBE grown MoTe2, indicating strong electron-phonon coupling in this materials. Our measured sound velocity agrees well with previously reported result of theoretical calculation. Our findings provide useful reference for the fundamental parameters: carrier lifetime and sound velocity, reveal the undiscovered carrier recombination effect of grain boundary/edge defects, both of which will facilitate the defect engineering in TMD materials for high speed opto-electronics.

cond-mat.mtrl-sci↗

arXiv2017

The Oort conjecture for Shimura curves of small unitary rank

We prove that a Shimura curve in the Siegel modular variety is not generically contained in the open Torelli locus as long as the rank of unitary part in its canonical Higgs bundle satisfies a numerical upper bound. As an application we show that the Coleman-Oort conjecture holds for Shimura curves associated to partial corestriction upon a suitable choice of parameters, which generalizes a construction due to Mumford.

math.AG↗

arXiv2017

Three-Stream Convolutional Networks for Video-based Person Re-Identification

This paper aims to develop a new architecture that can make full use of the feature maps of convolutional networks. To this end, we study a number of methods for video-based person re-identification and make the following findings: 1) Max-pooling only focuses on the maximum value of a receptive field, wasting a lot of information. 2) Networks with different streams even including the one with the worst performance work better than networks with same streams, where each one has the best performance alone. 3) A full connection layer at the end of convolutional networks is not necessary. Based on these studies, we propose a new convolutional architecture termed Three-Stream Convolutional Networks (TSCN). It first uses different streams to learn different aspects of feature maps for attentive spatio-temporal fusion of video, and then merges them together to study some union features. To further utilize the feature maps, two architectures are designed by using the strategies of multi-scale and upsampling. Comparative experiments on iLIDS-VID, PRID-2011 and MARS datasets illustrate that the proposed architectures are significantly better for feature extraction than the state-of-the-art models.

cs.CV↗

arXiv2017

Convolutional Low-Resolution Fine-Grained Classification

Successful fine-grained image classification methods learn subtle details between visually similar (sub-)classes, but the problem becomes significantly more challenging if the details are missing due to low resolution. Encouraged by the recent success of Convolutional Neural Network (CNN) architectures in image classification, we propose a novel resolution-aware deep model which combines convolutional image super-resolution and convolutional fine-grained classification into a single model in an end-to-end manner. Extensive experiments on the Stanford Cars and Caltech-UCSD Birds 200-2011 benchmarks demonstrate that the proposed model consistently performs better than conventional convolutional net on classifying fine-grained object classes in low-resolution images.

cs.CV↗

arXiv2017

Carrier Trapping by Oxygen Impurities in Molybdenum Diselenide

Understanding defect effect on carrier dynamics is essential for both fundamental physics and potential applications of transition metal dichalcogenides. Here, the phenomenon of oxygen impurities trapping photo-excited carriers has been studied with ultrafast pump-probe spectroscopy. Oxygen impurities are intentionally created in exfoliated multilayer MoSe2 with Ar+ plasma irradiation and air exposure. After plasma treatment, the signal of transient absorption first increases and then decreases, which is a signature of defect capturing carriers. With larger density of oxygen defects, the trapping effect becomes more prominent. The trapping defect densities are estimated from the transient absorption signal, and its increasing trend in the longer-irradiated sample agrees with the results from X-ray photoelectron spectroscopy. First principle calculations with density functional theory reveal that oxygen atoms occupying Mo vacancies create mid-gap defect states, which are responsible for the carrier trapping. Our findings shed light on the important role of oxygen defects as carrier trappers in transition metal dichalcogenides, and facilitates defect engineering in relevant material and device applications.

cond-mat.mtrl-sci↗

arXiv2017

Stability of Stationary Inverse Transport Equation in Diffusion Scaling

We consider the inverse problem of reconstructing the optical parameters for stationary radiative transfer equation (RTE) from velocity-averaged measurement. The RTE often contains multiple scales characterized by the magnitude of a dimensionless parameter---the Knudsen number ($K_n$). In the diffusive scaling ($K_n \ll 1$), the stationary RTE is well approximated by an elliptic equation in the forward setting. However, the inverse problem for the elliptic equation is acknowledged to be severely ill-posed as compared to the well-posedness of inverse transport equation, which raises the question of how uniqueness being lost as $K_n \rightarrow 0$. We tackle this problem by examining the stability of inverse problem with varying $K_n$. We show that, the discrepancy in two measurements is amplified in the reconstructed parameters at the order of $K_n^p~ (p = 1\text{ or} ~2)$, and as a result lead to ill-posedness in the zero limit of $K_n$. Our results apply to both continuous and discrete settings. Some numerical tests are performed in the end to validate these theoretical findings.

math.NA↗

arXiv2017

Chiral Topological Excitons in a Chern Band Insulator

A family of semiconductors called as Chern band insulator are shown to host exciton bands with non-zero topological Chern integers and chiral exciton edge modes. Using a prototypical two-band Chern insulator model, we calculate a cross-correlation function to obtain the exciton bands and their Chern integers. The lowest exciton band acquires Chern integers such as $\pm 1$ and $\pm 2$ in electronic Chern insulator phase. The non-trivial topology can be experimentally observed both by non-local optoelectronic response of exciton edge modes and by a phase shift in the cross-correlation response due to the bulk mode. Our result suggests that magnetically doped HgTe, InAs/GaSb quantum wells and $\text{(Bi,Sb)}_{2} \text{Te}_{3}$ thin film are promising candidates for a platform of topological excitonics.

cond-mat.mes-hall↗

arXiv2017

On the Use of Sparse Filtering for Covariate Shift Adaptation

In this paper we formally analyse the use of sparse filtering algorithms to perform covariate shift adaptation. We provide a theoretical analysis of sparse filtering by evaluating the conditions required to perform covariate shift adaptation. We prove that sparse filtering can perform adaptation only if the conditional distribution of the labels has a structure explained by a cosine metric. To overcome this limitation, we propose a new algorithm, named periodic sparse filtering, and carry out the same theoretical analysis regarding covariate shift adaptation. We show that periodic sparse filtering can perform adaptation under the looser and more realistic requirement that the conditional distribution of the labels has a periodic structure, which may be satisfied, for instance, by user-dependent data sets. We experimentally validate our theoretical results on synthetic data. Moreover, we apply periodic sparse filtering to real-world data sets to demonstrate that this simple and computationally efficient algorithm is able to achieve competitive performances.

cs.LG↗

arXiv2017

Generalized Multi-view Embedding for Visual Recognition and Cross-modal Retrieval

In this paper, the problem of multi-view embedding from different visual cues and modalities is considered. We propose a unified solution for subspace learning methods using the Rayleigh quotient, which is extensible for multiple views, supervised learning, and non-linear embeddings. Numerous methods including Canonical Correlation Analysis, Partial Least Sqaure regression and Linear Discriminant Analysis are studied using specific intrinsic and penalty graphs within the same framework. Non-linear extensions based on kernels and (deep) neural networks are derived, achieving better performance than the linear ones. Moreover, a novel Multi-view Modular Discriminant Analysis (MvMDA) is proposed by taking the view difference into consideration. We demonstrate the effectiveness of the proposed multi-view embedding methods on visual object recognition and cross-modal image retrieval, and obtain superior results in both applications compared to related methods.

cs.CV↗

arXiv2017

Giant Thermal Conductivity Enhancement in Multilayer MoS2 under Highly Compressive Strain

Multilayer MoS2 possesses highly anisotropic thermal conductivities along in-plane and cross-plane directions that could hamper heat dissipation in electronics. With about 9% cross-plane compressive strain created by hydrostatic pressure in a diamond anvil cell, we observed about 12 times increase in the cross-plane thermal conductivity of multilayer MoS2. Our experimental and theoretical studies reveal that this drastic change arises from the greatly strengthened interlayer interaction and heavily modified phonon dispersions along cross-plane direction, with negligible contribution from electronic thermal conductivity, despite its enhancement of 4 orders of magnitude. The anisotropic thermal conductivity in the multilayer MoS2 at ambient environment becomes almost isotropic under highly compressive strain, effectively transitioning from 2D to 3D heat dissipation. This strain tuning approach also makes possible parallel tuning of structural, thermal and electrical properties, and can be extended to the whole family of 2D Van der Waals solids, down to two layer systems.

cond-mat.mtrl-sci↗