arXiv subjects

Yixin Zhang

Publications and source records attributed to Yixin Zhang.

83 records · Page 5Linked to original sources

arXiv2022

Low-confidence Samples Matter for Domain Adaptation

Domain adaptation (DA) aims to transfer knowledge from a label-rich source domain to a related but label-scarce target domain. The conventional DA strategy is to align the feature distributions of the two domains. Recently, increasing researches have focused on self-training or other semi-supervised algorithms to explore the data structure of the target domain. However, the bulk of them depend largely on confident samples in order to build reliable pseudo labels, prototypes or cluster centers. Representing the target data structure in such a way would overlook the huge low-confidence samples, resulting in sub-optimal transferability that is biased towards the samples similar to the source domain. To overcome this issue, we propose a novel contrastive learning method by processing low-confidence samples, which encourages the model to make use of the target data structure through the instance discrimination process. To be specific, we create positive and negative pairs only using low-confidence samples, and then re-represent the original features with the classifier weights rather than directly utilizing them, which can better encode the task-specific semantic information. Furthermore, we combine cross-domain mixup to augment the proposed contrastive loss. Consequently, the domain gap can be well bridged through contrastive learning of intermediate representations across domains. We evaluate the proposed method in both unsupervised and semi-supervised DA settings, and extensive experimental results on benchmarks reveal that our method is effective and achieves state-of-the-art performance. The code can be found in https://github.com/zhyx12/MixLRCo.

cs.CV↗

arXiv2021

5th Place Solution for VSPW 2021 Challenge

In this article, we introduce the solution we used in the VSPW 2021 Challenge. Our experiments are based on two baseline models, Swin Transformer and MaskFormer. To further boost performance, we adopt stochastic weight averaging technique and design hierarchical ensemble strategy. Without using any external semantic segmentation dataset, our solution ranked the 5th place in the private leaderboard. Besides, we have some interesting attempts to tackle long-tail recognition and overfitting issues, which achieves improvement on val subset. Maybe due to distribution difference, these attempts don't work on test subset. We will also introduce these attempts and hope to inspire other researchers.

cs.CV↗

arXiv2021

Relaxation of Thermal Capillary Waves for Nanoscale Liquid Films on Anisotropic-slip Substrates

The relaxation dynamics of thermal capillary waves for nanoscale liquid films on anisotropic-slip substrates are investigated, using both molecular dynamics (MD) simulations and a Langevin model. The anisotropy of slip on substrates is achieved using a specific lattice plane of a face-centred cubic lattice. This surface's anisotropy breaks the simple scalar proportionality between slip velocity and wall shear stress and requires the introduction of a slip-coefficient tensor. The Langevin equation can describe both the growth of capillary wave spectra and the relaxation of capillary wave correlations, with the former providing a time scale for the surface to reach thermal equilibrium. Temporal correlations of interfacial Fourier modes, measured at thermal equilibrium in MD, demonstrate that (i) larger slip lengths lead to a faster decay in wave correlations, and (ii) unlike on isotropic-slip substrates, the time correlations of waves on anisotropic-slip substrates are wave-direction dependent. These findings emerge naturally from the proposed Langevin equation, which becomes wave-direction dependent, agrees well with MD results, and allows to produce experimentally-verifiable predictions.

physics.flu-dyn↗

arXiv2021

Deep Reinforcement Learning for Adaptive Exploration of Unknown Environments

Performing autonomous exploration is essential for unmanned aerial vehicles (UAVs) operating in unknown environments. Often, these missions start with building a map for the environment via pure exploration and subsequently using (i.e. exploiting) the generated map for downstream navigation tasks. Accomplishing these navigation tasks in two separate steps is not always possible or even disadvantageous for UAVs deployed in outdoor and dynamically changing environments. Current exploration approaches either use a priori human-generated maps or use heuristics such as frontier-based exploration. Other approaches use learning but focus only on learning policies for specific tasks by either using sample inefficient random exploration or by making impractical assumptions about full map availability. In this paper, we develop an adaptive exploration approach to trade off between exploration and exploitation in one single step for UAVs searching for areas of interest (AoIs) in unknown environments using Deep Reinforcement Learning (DRL). The proposed approach uses a map segmentation technique to decompose the environment map into smaller, tractable maps. Then, a simple information gain function is repeatedly computed to determine the best target region to search during each iteration of the process. DDQN and A2C algorithms are extended with a stack of LSTM layers and trained to generate optimal policies for the exploration and exploitation, respectively. We tested our approach in 3 different tasks against 4 baselines. The results demonstrate that our proposed approach is capable of navigating through randomly generated environments and covering more AoI in less time steps compared to the baselines.

cs.LG↗

arXiv2021

Thermal Capillary Wave Growth and Surface Roughening of Nanoscale Liquid Films

The well-known thermal capillary wave theory, which describes the capillary spectrum of the free surface of a liquid film, does not reveal the transient dynamics of surface waves, e.g., the process through which a smooth surface becomes rough. Here, a Langevin model is proposed that can capture these dynamics, goes beyond the long-wave paradigm which can be inaccurate at the nanoscale, and is validated using molecular dynamics simulations for nanoscale films on both planar and cylindrical substrates. We show that a scaling relation exists for surface roughening of a planar film and the scaling exponents belong to a specific universality class. The capillary spectra of planar films are found to advance towards a static spectrum, with the roughness of the surface $W$ increasing as a power law of time $W\sim t^{1/8}$ before saturation. However, the spectra of an annular film (with outer radius $h_0$) are unbounded for dimensionless wavenumber $qh_0<1$ due to the Rayleigh-Plateau instability.

physics.flu-dyn↗

arXiv2019

Blockchain of Signature Material Combining Cryptographic Hash Function and DNA Steganography

An ideal signature material and method, which can be used to prove the authenticity of a physical item and against forgery, should be immune to the fast developments in digital and engineering technologies. Herein, the design of signature material combining cryptographic hash function and DNA steganography is proposed. The encrypting materials are used to construct a series of time-stamped records (blockchain) associated with published hash values, while each DNA-encrypted block is associated with a set of DNA keys. The decrypted DNA information, as digital keys, can be validated through a hash function to compare with the published hash values. The blocks can also be cross-referenced among different related signatures. While both digital cryptography and DNA steganography can have large key size, automated brutal force search is far more labor intensive and expensive for DNA steganography with wet lab experiments, as compared to its digital counterpart. Moreover, the time-stamped blockchain structure permits the incorporation of new cryptographic functions and DNA steganographies over time, thus can evolve over time without losing the continuous history line.

q-bio.BM↗

arXiv2019

Identifying vital nodes based on reverse greedy method

The identification of vital nodes that maintain the network connectivity is a long-standing challenge in network science. In this paper, we propose a so-called reverse greedy method where the least important nodes are preferentially chosen to make the size of the largest component in the corresponding induced subgraph as small as possible. Accordingly, the nodes being chosen later are more important in maintaining the connectivity. Empirical analyses on ten real networks show that the reverse greedy method performs remarkably better than well-known state-of-the-art methods.

physics.soc-ph↗

arXiv2018

Underwater Anchor-AUV Localization Geometries with an Isogradient Sound Speed Profile: A CRLB-Based Optimality Analysis

Existing works have explored the anchor deployment for autonomous underwater vehicles (AUVs) localization under the assumption that the sound propagates straightly underwater at a constant speed. Considering that the underwater acoustic waves propagate along bent curves at varying speeds in practice, it becomes much more challenging to determine a proper anchor deployment configuration. In this paper, taking the practical variability of underwater sound speed into account, we investigate the anchor-AUV geometry problem in a 3-D time-of-flight (ToF) based underwater scenario from the perspective of localization accuracy. To address this problem, we first rigorously derive the Jacobian matrix of measurement errors to quantify the Cramer-Rao lower bound (CRLB) with a widely-adopted isogradient sound speed profile (SSP). We then formulate an optimization problem that minimizes the trace of the CRLB subject to the angle and range constraints to figure out the anchor-AUV geometry, which is multivariate and nonlinear and thus generally hard to handle. For mathematical tractability, by adopting tools from the estimation theory, we interestingly find that this problem can be equivalently transformed into a more explicit univariate optimization problem. By this, we obtain an easy-to-implement anchor-AUV geometry that yields satisfactory localization performance, referred to as the uniform sea-surface circumference (USC) deployment. Extensive simulation results validate our theoretical analysis and show that our proposed USC scheme outperforms both the cube and the random deployment schemes in terms of localization accuracy under the same parameter settings.

cs.IT↗

arXiv2017

Universal decay of quantumness for photonic qubits carrying orbital angular momentum through atmospheric turbulence

We investigate the decay properties of the quantumness including quantum entanglement, quantum discord and quantum coherence for two photonic qubits, which are partially entangled in their orbital angular momenta, through Kolmogorov turbulent atmosphere. It is found that the decay of quantum coherence and quantum discord may be qualitatively different from that of quantum entanglement when the initial state of two photons is not maximally entangled. We also derive two universal decay laws for quantum coherence and quantum discord, respectively, and show that the decay of quantum coherence is more robust than nonclassical correlations.

quant-ph↗

arXiv2014

Sudden transitions of trace distance discord of dipole-dipole coupled two qubits

We investigate the exact dynamics of trace distance discord by considering two qubits under dephasing whose states belong to a class of $X$ states beyond Bell diagonal form. The necessary condition for the occurrence of freezing trace distance discord is found and compared with that of entropic discord. For an illustration, we consider two interacting qubits coupled to independent reservoirs and demonstrate these dynamical properties of trace distance discord. It is interesting to find that the freezing trace distance discord exists even for $X$ states without maximally mixed marginals and can be tuned by dipole-dipole coupling of two qubits. Moreover, we consider the initial extended Werner-like states and investigate the differences between trace distance discord and entanglement. The influences of initial state and the dipole-dipole coupling of the two qubits on the dynamics of nonclassical correlations are explored.

quant-ph↗

arXiv2014

Negativity of quantumness and non-Markovianity in a qubit coupled to a thermal Ising spin bath system

We propose a scheme to characterize the non-Markovian dynamics and quantify the non-Markovianity via the non-classicality measured by the negativity of quantumness. By considering a qubit in contact with a critical Ising spin bath and introducing an ancilla, we show that revivals of negativity of quantumness indicate the non-Markovian dynamics. Furthermore, a normalized measure of non-Markovianity based on the negativity of quantumness is introduced and the influences of bath criticality, bath temperature and bath size on the non-Markovianity are discussed. It is shown that, at the critical point, the decay of non-Markovianity versus the size of spin bath is fastest and the non-Markovianity is exactly zero only in the thermodynamic limit. Besides, non-trivial behaviours of negativity of quantumness such as sudden change, double sudden changes and keeping constant are found for different relations between parameters of the initial state. Finally, how the non-classicality of the system is affected by a series of bang-bang pulses is also examined.

quant-ph↗