arXiv subjects

Qian Li

Publications and source records attributed to Qian Li.

At least 235 records · Page 13Linked to original sources

arXiv2022

CoSCL: Cooperation of Small Continual Learners is Stronger than a Big One

Continual learning requires incremental compatibility with a sequence of tasks. However, the design of model architecture remains an open question: In general, learning all tasks with a shared set of parameters suffers from severe interference between tasks; while learning each task with a dedicated parameter subspace is limited by scalability. In this work, we theoretically analyze the generalization errors for learning plasticity and memory stability in continual learning, which can be uniformly upper-bounded by (1) discrepancy between task distributions, (2) flatness of loss landscape and (3) cover of parameter space. Then, inspired by the robust biological learning system that processes sequential experiences with multiple parallel compartments, we propose Cooperation of Small Continual Learners (CoSCL) as a general strategy for continual learning. Specifically, we present an architecture with a fixed number of narrower sub-networks to learn all incremental tasks in parallel, which can naturally reduce the two errors through improving the three components of the upper bound. To strengthen this advantage, we encourage to cooperate these sub-networks by penalizing the difference of predictions made by their feature representations. With a fixed parameter budget, CoSCL can improve a variety of representative continual learning approaches by a large margin (e.g., up to 10.64% on CIFAR-100-SC, 9.33% on CIFAR-100-RS, 11.45% on CUB-200-2011 and 6.72% on Tiny-ImageNet) and achieve the new state-of-the-art performance.

cs.LG↗

arXiv2022

Atomically engineered cobaltite layers for robust ferromagnetism

Emergent phenomena at heterointerfaces are directly associated with the bonding geometry of adjacent layers. Effective control of accessible parameters, such as the bond length and bonding angles, offers an elegant method to tailor competing energies of the electronic and magnetic ground states. In this study, we construct unit thick syntactic layers of cobaltites within a strongly tilted octahedral matrix via atomically precise synthesis. The octahedral tilt patterns of adjacent layers propagate into cobaltites, leading to a continuation of octahedral tilting while maintaining significant misfit tensile strain. These effects induce severe rumpling within an atomic plane of neighboring layers triggers the electronic reconstruction between the splitting orbitals. First-principles calculations reveal that the cobalt ions transits to a higher spin state level upon octahedral tilting, resulting in robust ferromagnetism in ultrathin cobaltites. This work demonstrates a design methodology for fine-tuning the lattice and spin degrees of freedom in correlated quantum heterostructures by exploiting epitaxial geometric engineering.

cond-mat.mtrl-sci↗

arXiv2022

How Does Knowledge Graph Embedding Extrapolate to Unseen Data: A Semantic Evidence View

Knowledge Graph Embedding (KGE) aims to learn representations for entities and relations. Most KGE models have gained great success, especially on extrapolation scenarios. Specifically, given an unseen triple (h, r, t), a trained model can still correctly predict t from (h, r, ?), or h from (?, r, t), such extrapolation ability is impressive. However, most existing KGE works focus on the design of delicate triple modeling function, which mainly tells us how to measure the plausibility of observed triples, but offers limited explanation of why the methods can extrapolate to unseen data, and what are the important factors to help KGE extrapolate. Therefore in this work, we attempt to study the KGE extrapolation of two problems: 1. How does KGE extrapolate to unseen data? 2. How to design the KGE model with better extrapolation ability? For the problem 1, we first discuss the impact factors for extrapolation and from relation, entity and triple level respectively, propose three Semantic Evidences (SEs), which can be observed from train set and provide important semantic information for extrapolation. Then we verify the effectiveness of SEs through extensive experiments on several typical KGE methods. For the problem 2, to make better use of the three levels of SE, we propose a novel GNN-based KGE model, called Semantic Evidence aware Graph Neural Network (SE-GNN). In SE-GNN, each level of SE is modeled explicitly by the corresponding neighbor pattern, and merged sufficiently by the multi-layer aggregation, which contributes to obtaining more extrapolative knowledge representation. Finally, through extensive experiments on FB15k-237 and WN18RR datasets, we show that SE-GNN achieves state-of-the-art performance on Knowledge Graph Completion task and performs a better extrapolation ability. Our code is available at https://github.com/renli1024/SE-GNN.

cs.CL↗

arXiv2022

Coexistence of extended flat band and Kekulé order in Li-intercalated graphene

Doping graphene near the 1/4 filling to shift the extended flat band and van Hove singularity below E$_F$ has been highly desirable. Here we report the experimental observation of an extended flat band below E$_F$ in Li-intercalated graphene. Strong electron-phonon interaction is clearly identified by notable kinks in the band dispersion. Moreover, the evolution of the band structure upon Li intercalation shows that the extended flat band and the Kekulé order emerge simultaneously. Our work provides opportunities for investigating flat band related instabilities and its interplay with the Kekulé order

cond-mat.mes-hall↗

arXiv2022

Room-temperature printing of ultrathin Quasi-2D GaN semiconductor via liquid metal gallium surface confined nitridation reaction

Outstanding wide-bandgap semiconductor material such as gallium nitride (GaN) has been extensively utilized in power electronics, radiofrequency amplifiers, and harsh environment devices. Due to its quantum confinement effect in enabling desired deep-ultraviolet emission, excitonic impact, and electronic transport features, two-dimensional (2D) or ultrathin quasi-2D GaN semiconductors have been one of the most remarkable candidates for future growth of microelectronic devices. Here, for the first time, we reported a large area, wide bandgap, and room-temperature quasi-2D GaN synthesis and printing strategy through introducing the plasma medicated liquid metal gallium surface-confined nitridation reaction mechanism. The developed direct fabrication and compositional process is consistent with various electronics manufacturing approaches and thus opens an easy going way for cost-effective growth of the third-generation semiconductor. In particular, the fully printed field-effect transistors relying on the GaN thus made show p-type switching with an on/off ratio greater than 105, maximum field-effect hole mobility of 53 cm2/(V*s), and a small sub-threshold swing. As it was demonstrated, the present method allows to produce at room temperature the GaN with thickness spanning from 1nm to nanometers. This basic method can be further extended, generalized, and utilized for making various electronic and photoelectronic devices in the coming time.

physics.app-ph↗

arXiv2022

Forestry digital twin with machine learning in Landsat 7 data

Modeling forests using historical data allows for more accurately evolution analysis, thus providing an important basis for other studies. As a recognized and effective tool, remote sensing plays an important role in forestry analysis. We can use it to derive information about the forest, including tree type, coverage and canopy density. There are many forest time series modeling studies using statistic values, but few using remote sensing images. Image prediction digital twin is an implementation of digital twin, which aims to predict future images bases on historical data. In this paper, we propose an LSTM-based digital twin approach for forest modeling, using Landsat 7 remote sensing image within 20 years. The experimental results show that the prediction twin method in this paper can effectively predict the future images of study area.

cs.LG↗

arXiv2022

New Distinguishers for Negation-Limited Weak Pseudorandom Functions

We show how to distinguish circuits with $\log k$ negations (a.k.a $k$-monotone functions) from uniformly random functions in $\exp\left(\tilde{O}\left(n^{1/3}k^{2/3}\right)\right)$ time using random samples. The previous best distinguisher, due to the learning algorithm by Blais, Cannone, Oliveira, Servedio, and Tan (RANDOM'15), requires $\exp\big(\tilde{O}(n^{1/2} k)\big)$ time. Our distinguishers are based on Fourier analysis on \emph{slices of the Boolean cube}. We show that some "middle" slices of negation-limited circuits have strong low-degree Fourier concentration and then we apply a variation of the classic Linial, Mansour, and Nisan "Low-Degree algorithm" (JACM'93) on slices. Our techniques also lead to a slightly improved weak learner for negation limited circuits under the uniform distribution.

cs.CC↗

arXiv2022

Universal Segmentation of 33 Anatomies

In the paper, we present an approach for learning a single model that universally segments 33 anatomical structures, including vertebrae, pelvic bones, and abdominal organs. Our model building has to address the following challenges. Firstly, while it is ideal to learn such a model from a large-scale, fully-annotated dataset, it is practically hard to curate such a dataset. Thus, we resort to learn from a union of multiple datasets, with each dataset containing the images that are partially labeled. Secondly, along the line of partial labelling, we contribute an open-source, large-scale vertebra segmentation dataset for the benefit of spine analysis community, CTSpine1K, boasting over 1,000 3D volumes and over 11K annotated vertebrae. Thirdly, in a 3D medical image segmentation task, due to the limitation of GPU memory, we always train a model using cropped patches as inputs instead a whole 3D volume, which limits the amount of contextual information to be learned. To this, we propose a cross-patch transformer module to fuse more information in adjacent patches, which enlarges the aggregated receptive field for improved segmentation performance. This is especially important for segmenting, say, the elongated spine. Based on 7 partially labeled datasets that collectively contain about 2,800 3D volumes, we successfully learn such a universal model. Finally, we evaluate the universal model on multiple open-source datasets, proving that our model has a good generalization performance and can potentially serve as a solid foundation for downstream tasks.

eess.IV↗

arXiv2022

Magnetization of the spin-1/2 Heisenberg antiferromagnet on the triangular lattice

After decades of debate, now there is a rough consensus that at zero temperature the spin-$1/2$ Heisenberg antiferromagnet on the triangular lattice is three-sublattice $120^\circ$ magnetically ordered, in contrast to a quantum spin liquid as originally proposed. However, there remains considerable discrepancy in the magnetization reported among various methods. To resolve this issue, in this work we revisit this model by the tensor-network state algorithm. The ground-state energy per bond $E_b$ and magnetization per spin $M_0$ in the thermodynamic limit are obtained with high precision. The former is estimated to be $E_b = -0.18334(10)$. This value agrees well with that from the series expansion. The three-sublattice magnetic order is firmly confirmed and the magnetization is determined as $M_0 = 0.161(5)$. It is about $32\%$ of its classical value and slightly below the lower bound from the series expansion. In comparison with the best estimated value by Monte Carlo and density-matrix renormalization group, our result is about $20\%$ smaller. This magnetic order is consistent with further analysis of the three-body correlation. Our work thus provides new benchmark results for this prototypical model.

cond-mat.str-el↗

arXiv2022

Universal classical optical computing inspired by quantum information process

Quantum computing has attracted much attention in recent decades, since it is believed to solve certain problems substantially faster than traditional computing methods. Theoretically, such an advance can be obtained by networks of the quantum operators in universal gate sets, one famous example of which is formed by CNOT gate and single qubit gates. However, realizing a device that performs practical quantum computing is tricky. This is because it requires a scalable qubit system with long coherence time and good controls, which is harsh for most current platforms. Here, we demonstrate that the information process based on a relatively stable system -- classical optical system, can be considered as an analogy of universal quantum computing. By encoding the information via the polarization state of classical beams, the optical computing elements that corresponds to the universal gate set are presented and their combination for a general information process are theoretically illustrated. Taking the analogy of two-qubit processor as an example, we experimentally verify that our proposal works well. Considering the potential of optical system for reliable and low-energy-consuming computation, our results open a new way towards advanced information processing with high quality and efficiency.

physics.optics↗

arXiv2022

Causal Disentanglement for Semantics-Aware Intent Learning in Recommendation

Traditional recommendation models trained on observational interaction data have generated large impacts in a wide range of applications, it faces bias problems that cover users' true intent and thus deteriorate the recommendation effectiveness. Existing methods tracks this problem as eliminating bias for the robust recommendation, e.g., by re-weighting training samples or learning disentangled representation. The disentangled representation methods as the state-of-the-art eliminate bias through revealing cause-effect of the bias generation. However, how to design the semantics-aware and unbiased representation for users true intents is largely unexplored. To bridge the gap, we are the first to propose an unbiased and semantics-aware disentanglement learning called CaDSI (Causal Disentanglement for Semantics-Aware Intent Learning) from a causal perspective. Particularly, CaDSI explicitly models the causal relations underlying recommendation task, and thus produces semantics-aware representations via disentangling users true intents aware of specific item context. Moreover, the causal intervention mechanism is designed to eliminate confounding bias stemmed from context information, which further to align the semantics-aware representation with users true intent. Extensive experiments and case studies both validate the robustness and interpretability of our proposed model.

cs.IR↗

arXiv2022

Epitaxial stabilization of an orthorhombic Mg-Ti-O superconductor

The family of titanium oxide superconductors exhibits many intriguing phenomena comparable to cuprates and iron pnictides/chalcogenides, and thus provides an ideal platform to contrastively study the unconventional pairing mechanism of high-temperature superconductors. Here, we successfully deposit superconducting Mg-Ti-O films on MgAl$_2$O$_4$ substrates with three principal orientations by ablating a MgTi$_2$O$_4$ target. Particularly, it is striking to observed that a single-crystalline film of an unintended structure has been grown on the (011)-oriented substrate, with the highest zero resistance transition temperature ($T_{\mathrm{c}0}$) of 5.0 K among them. The film has a highly reduced Mg/Ti ratio and an orthorhombic Ti$_9$O$_{10}$-like structure (denoted as Mg: Ti$_9$O$_{10}$), demonstrated by further characterizations of chemical composition and structure. Such a structure is unstable in bulk but favorable to be epitaxially stabilized on the (011)-surface of MgAl$_2$O$_4$ due to a relatively small strain at the formed interface. An isotropic upper critical field ($B_{\mathrm{c}2}$) up to 13.7 T that breaks the Pauli limit is observed in the Mg: Ti$_9$O$_{10}$ film, analogous to other superconducting titanium oxides. The similarity points to a common origin for the superconductivity in the family, which will provide valuable opinions for the mechanism of unconventional superconductivity in transition metal compounds.

cond-mat.supr-con↗

arXiv2021

Deep Treatment-Adaptive Network for Causal Inference

Causal inference is capable of estimating the treatment effect (i.e., the causal effect of treatment on the outcome) to benefit the decision making in various domains. One fundamental challenge in this research is that the treatment assignment bias in observational data. To increase the validity of observational studies on causal inference, representation based methods as the state-of-the-art have demonstrated the superior performance of treatment effect estimation. Most representation based methods assume all observed covariates are pre-treatment (i.e., not affected by the treatment), and learn a balanced representation from these observed covariates for estimating treatment effect. Unfortunately, this assumption is often too strict a requirement in practice, as some covariates are changed by doing an intervention on treatment (i.e., post-treatment). By contrast, the balanced representation learned from unchanged covariates thus biases the treatment effect estimation.

cs.LG↗

arXiv2021

A Survey on Text Classification: From Shallow to Deep Learning

Text classification is the most fundamental and essential task in natural language processing. The last decade has seen a surge of research in this area due to the unprecedented success of deep learning. Numerous methods, datasets, and evaluation metrics have been proposed in the literature, raising the need for a comprehensive and updated survey. This paper fills the gap by reviewing the state-of-the-art approaches from 1961 to 2021, focusing on models from traditional models to deep learning. We create a taxonomy for text classification according to the text involved and the models used for feature extraction and classification. We then discuss each of these categories in detail, dealing with both the technical developments and benchmark datasets that support tests of predictions. A comprehensive comparison between different techniques, as well as identifying the pros and cons of various evaluation metrics are also provided in this survey. Finally, we conclude by summarizing key implications, future research directions, and the challenges facing the research area.

cs.CL↗

arXiv2021

Reinforcement Learning-based Dialogue Guided Event Extraction to Exploit Argument Relations

Event extraction is a fundamental task for natural language processing. Finding the roles of event arguments like event participants is essential for event extraction. However, doing so for real-life event descriptions is challenging because an argument's role often varies in different contexts. While the relationship and interactions between multiple arguments are useful for settling the argument roles, such information is largely ignored by existing approaches. This paper presents a better approach for event extraction by explicitly utilizing the relationships of event arguments. We achieve this through a carefully designed task-oriented dialogue system. To model the argument relation, we employ reinforcement learning and incremental learning to extract multiple arguments via a multi-turned, iterative process. Our approach leverages knowledge of the already extracted arguments of the same sentence to determine the role of arguments that would be difficult to decide individually. It then uses the newly obtained information to improve the decisions of previously extracted arguments. This two-way feedback process allows us to exploit the argument relations to effectively settle argument roles, leading to better sentence understanding and event extraction. Experimental results show that our approach consistently outperforms seven state-of-the-art event extraction methods for the classification of events and argument role and argument identification.

cs.CL↗

arXiv2021

Population inversion and Dirac fermion cooling in 3D Dirac semimetal Cd$_3$As$_2$

Revealing the ultrafast dynamics of three-dimensional (3D) Dirac fermions upon photoexcitation is critical for both fundamental science and device applications. So far, how the cooling of 3D Dirac fermions differs from that of two-dimensional (2D) Dirac fermions and whether there is population inversion are fundamental questions that remain to be answered. Here we reveal the ultrafast dynamics of Dirac fermions in a model 3D Dirac semimetal Cd$_3$As$_2$ by ultrafast time- and angle-resolved photoemission spectroscopy (TrARPES) with a tunable probe photon energy from 5.3 - 6.9 eV. The energy- and momentum-resolved relaxation rate shows a linear dependence on the energy, suggesting Dirac fermion cooling through intraband relaxation. Moreover, a population inversion is reported based on the observation of accumulated photoexcited carriers in the conduction band with a lifetime of $τ_n$ = 3.0 ps. Our work provides direct experimental evidence for a long-lived population inversion in a 3D Dirac semimetal, which is in contrast to 2D graphene where the interband relaxation occurs on a much faster timescale.

cond-mat.mes-hall↗

arXiv2021

Room-temperature ferromagnetism at an oxide/nitride interface

Heterointerfaces have led to the discovery of novel electronic and magnetic states because of their strongly entangled electronic degrees of freedom. Single-phase chromium compounds always exhibit antiferromagnetism following the prediction of Goodenough-Kanamori rules. So far, exchange coupling between chromium ions via hetero-anions has not been explored and the associated quantum states is unknown. Here we report the successful epitaxial synthesis and characterizations of chromium oxide (Cr2O3)-chromium nitride (CrN) superlattices. Room-temperature ferromagnetic spin ordering is achieved at the interfaces between these two antiferromagnets, and the magnitude of the effect decays with increasing layer thickness. First-principles calculations indicate that robust ferromagnetic spin interaction between Cr3+ ions via anion-hybridizations across the interface yields the lowest total energy. This work opens the door to fundamental understanding of the unexpected and exceptional properties of oxide-nitride interfaces and provides access to hidden phases at low-dimensional quantum heterostructures.

cond-mat.mtrl-sci↗

arXiv2021

DuDoTrans: Dual-Domain Transformer Provides More Attention for Sinogram Restoration in Sparse-View CT Reconstruction

While Computed Tomography (CT) reconstruction from X-ray sinograms is necessary for clinical diagnosis, iodine radiation in the imaging process induces irreversible injury, thereby driving researchers to study sparse-view CT reconstruction, that is, recovering a high-quality CT image from a sparse set of sinogram views. Iterative models are proposed to alleviate the appeared artifacts in sparse-view CT images, but the computation cost is too expensive. Then deep-learning-based methods have gained prevalence due to the excellent performances and lower computation. However, these methods ignore the mismatch between the CNN's \textbf{local} feature extraction capability and the sinogram's \textbf{global} characteristics. To overcome the problem, we propose \textbf{Du}al-\textbf{Do}main \textbf{Trans}former (\textbf{DuDoTrans}) to simultaneously restore informative sinograms via the long-range dependency modeling capability of Transformer and reconstruct CT image with both the enhanced and raw sinograms. With such a novel design, reconstruction performance on the NIH-AAPM dataset and COVID-19 dataset experimentally confirms the effectiveness and generalizability of DuDoTrans with fewer involved parameters. Extensive experiments also demonstrate its robustness with different noise-level scenarios for sparse-view CT reconstruction. The code and models are publicly available at https://github.com/DuDoTrans/CODE

eess.IV↗