arXiv subjects

Tao Li

Publications and source records attributed to Tao Li.

At least 451 records · Page 25Linked to original sources

arXiv2020

Causality and Batch Reinforcement Learning: Complementary Approaches To Planning In Unknown Domains

Reinforcement learning algorithms have had tremendous successes in online learning settings. However, these successes have relied on low-stakes interactions between the algorithmic agent and its environment. In many settings where RL could be of use, such as health care and autonomous driving, the mistakes made by most online RL algorithms during early training come with unacceptable costs. These settings require developing reinforcement learning algorithms that can operate in the so-called batch setting, where the algorithms must learn from set of data that is fixed, finite, and generated from some (possibly unknown) policy. Evaluating policies different from the one that collected the data is called off-policy evaluation, and naturally poses counter-factual questions. In this project we show how off-policy evaluation and the estimation of treatment effects in causal inference are two approaches to the same problem, and compare recent progress in these two areas.

cs.LG↗

arXiv2020

Structured Tuning for Semantic Role Labeling

Recent neural network-driven semantic role labeling (SRL) systems have shown impressive improvements in F1 scores. These improvements are due to expressive input representations, which, at least at the surface, are orthogonal to knowledge-rich constrained decoding mechanisms that helped linear SRL models. Introducing the benefits of structure to inform neural models presents a methodological challenge. In this paper, we present a structured tuning framework to improve models using softened constraints only at training time. Our framework leverages the expressiveness of neural networks and provides supervision with structured loss components. We start with a strong baseline (RoBERTa) to validate the impact of our approach, and show that our framework outperforms the baseline by learning to comply with declarative constraints. Additionally, our experiments with smaller training sizes show that we can achieve consistent improvements under low-resource scenarios.

cs.CL↗

arXiv2020

Self-Learning with Rectification Strategy for Human Parsing

In this paper, we solve the sample shortage problem in the human parsing task. We begin with the self-learning strategy, which generates pseudo-labels for unlabeled data to retrain the model. However, directly using noisy pseudo-labels will cause error amplification and accumulation. Considering the topology structure of human body, we propose a trainable graph reasoning method that establishes internal structural connections between graph nodes to correct two typical errors in the pseudo-labels, i.e., the global structural error and the local consistency error. For the global error, we first transform category-wise features into a high-level graph model with coarse-grained structural information, and then decouple the high-level graph to reconstruct the category features. The reconstructed features have a stronger ability to represent the topology structure of the human body. Enlarging the receptive field of features can effectively reducing the local error. We first project feature pixels into a local graph model to capture pixel-wise relations in a hierarchical graph manner, then reverse the relation information back to the pixels. With the global structural and local consistency modules, these errors are rectified and confident pseudo-labels are generated for retraining. Extensive experiments on the LIP and the ATR datasets demonstrate the effectiveness of our global and local rectification modules. Our method outperforms other state-of-the-art methods in supervised human parsing tasks.

cs.CV↗

arXiv2020

Theory of the anomalous spin dynamics of spin-$\frac{1}{2}$ triangular lattice Heisenberg antiferromagnet and its application to Ba$_3$CoSb$_2$O$_9$

Although it is well accepted that the spin-$\frac{1}{2}$ triangular lattice Heisenberg antiferromagnet(TLHAF) has a long range ordered ground state, a thorough understanding of its spin dynamics is still missing. While the linear spin wave theory(LSWT) predicts three branches of magnon mode in the magnetic Brillouin zone(MBZ), the 1/S expansion at the next order is found to break down in a large portion of the MBZ centered around the M point, leaving the fate of the magnon modes there undecided. Recent neutron scattering measurement on Ba$_3$CoSb$_2$O$_9$, an ideal realization of the spin-$\frac{1}{2}$ TLHAF, provides a surprising answer to this issue. Two, rather than three branches of magnon mode are observed around the M point, whose dispersion are strongly renormalized with respect to the LSWT prediction and exhibit pronounced roton-like minimum at the M point. This is accompanied by a strong spin fluctuation continuum at higher energy, inside which two strong and broad spectral peaks of unknown origin are observed. In this work, we propose a simple picture for these spectral anomalies by invoking the resonating valence bond(RVB) physics in the description of the ground state of the system. We find that the roton-like minimum in the magnon dispersion can be explained by the coupling between the collective spin fluctuation and the continuum of Dirac spinon excitation moving in a $π$-flux background. We also propose that the two broad peaks in the continuum can be understood respectively as the Landau damped third magnon mode and the Landau damped longitudinal mode. Such a picture can be verified by studying the polarization character of the various spectral features.

cond-mat.str-el↗

arXiv2020

A Deep Learning Method for Complex Human Activity Recognition Using Virtual Wearable Sensors

Sensor-based human activity recognition (HAR) is now a research hotspot in multiple application areas. With the rise of smart wearable devices equipped with inertial measurement units (IMUs), researchers begin to utilize IMU data for HAR. By employing machine learning algorithms, early IMU-based research for HAR can achieve accurate classification results on traditional classical HAR datasets, containing only simple and repetitive daily activities. However, these datasets rarely display a rich diversity of information in real-scene. In this paper, we propose a novel method based on deep learning for complex HAR in the real-scene. Specially, in the off-line training stage, the AMASS dataset, containing abundant human poses and virtual IMU data, is innovatively adopted for enhancing the variety and diversity. Moreover, a deep convolutional neural network with an unsupervised penalty is proposed to automatically extract the features of AMASS and improve the robustness. In the on-line testing stage, by leveraging advantages of the transfer learning, we obtain the final result by fine-tuning the partial neural network (optimizing the parameters in the fully-connected layers) using the real IMU data. The experimental results show that the proposed method can surprisingly converge in a few iterations and achieve an accuracy of 91.15% on a real IMU dataset, demonstrating the efficiency and effectiveness of the proposed method.

cs.CV↗

arXiv2020

Detecting EPR steering via two classes of local measurements

We study the Einstein-Podolsky-Rosen (EPR) steering and present steerability criteria for arbitrary qubit-qudit (qudit-qubit) systems based on mutually unbiased measurements (MUMs) and general symmetric informationally complete measurements (general SIC-POVMs). Avoiding the usual complicated steering inequalities, these criteria can be more operational than some existing criteria and implemented experimentally. Detailed examples are given to illustrate the efficiency of the criteria in both computation and experimental implementation.

quant-ph↗

arXiv2019

Pressure-induced superconductivity in topological type II Dirac semimetal NiTe2

Very recently, NiTe2 has been reported to be a type II Dirac semimetal with Dirac nodes near the Fermi surface. Furthermore, it is unveiled that NiTe2 presents the Hall Effect, which is ascribed to orbital magnetoresistance. The physical properties behavior of NiTe2 under high pressure attracts us. In this paper, we investigate the electrical properties of polycrystalline NiTe2 by application of pressure ranging from 3.4GPa to 54.45Gpa. Superconductivity emerges at critical pressure 12GPa with a transition temperature of 3.7K, and Tc reaches its maximum, 6.4 K, at the pressure of 52.8GPa. Comparing with the superconductivity in MoP, we purposed the possibility of topological superconductivity in NiTe2. Two superconductivity transitions are observed with pressure increasing in single crystal.

cond-mat.supr-con↗

arXiv2019

Mutually unbiased measurement based entanglement witnesses

We study entanglement witness and present a construction of entanglement witnesses in terms of the mutually unbiased measurements (MUMs). These witnesses include the entanglement witnesses constructed from mutually unbiased bases (MUBs) as a special case. Comparing with the dimension dependence of MUBs, the witnesses can be always constructed from a complete set of $d+1$ MUMs for any dimension $d$. We show that our witness can detect entanglement better than previous separability criterion given also by MUMs. And the approach can be experimentally implemented.

quant-ph↗

arXiv2019

On Measuring and Mitigating Biased Inferences of Word Embeddings

Word embeddings carry stereotypical connotations from the text they are trained on, which can lead to invalid inferences in downstream models that rely on them. We use this observation to design a mechanism for measuring stereotypes using the task of natural language inference. We demonstrate a reduction in invalid inferences via bias mitigation strategies on static word embeddings (GloVe). Further, we show that for gender bias, these techniques extend to contextualized embeddings when applied selectively only to the static components of contextualized embeddings (ELMo, BERT).

cs.CL↗

arXiv2019

What does the giant thermal Hall effect observed in the high temperature superconductors imply?

A giant negative thermal Hall signal is discovered recently in the pseudogap phase of the high temperature superconductors\cite{Taillefer}. The Wiedemann-Franz law relating the charge and thermal transport is found to be strongly violated as the thermal Hall signal increases monotonically with decreasing doping in the pseudogap phase when the Hall response of the system is seen to be strongly suppressed. In particular, the thermal Hall signal is the strongest in the parent compounds, which are antiferromagnetic insulators. People believe that such an observation may challenge our understanding of the parent compounds as a spin-$\frac{1}{2}$ Heisenberg antiferromagnet on the square lattice, since it is forbidden by a well known no-go "theorem" for quantum magnets with an edge-sharing lattice. Here we show that the observed thermal Hall signal can be naturally understood as the orbital magnetic response of the quantum Heisenberg model on square lattice. We show that the thermal Hall signal is in fact a measure of the spin chirality fluctuation in the system. The universal observation of the giant negative thermal Hall signal in and only in the pseudogap phase implies that the antiferromagnetic correlation between local spins is at the root of the pseudogap phenomena. It also implies that the pseudogap phenomena in the cuprates is always accompanied by strong spin chirality fluctuation, which is suppressed only when the local spin fluctuation at low energy is totally replaced by itinerant quasiparticle behavior at sufficiently large doping.

cond-mat.str-el↗

arXiv2019

Resource-efficient analyzer of Bell and Greenberger-Horne-Zeilinger states of multiphoton systems

We propose a resource-efficient error-rejecting entangled-state analyzer for polarization-encoded multiphoton systems. Our analyzer is based on two single-photon quantum-nondemolition detectors, where each of them is implemented with a four-level emitter (e.g., a quantum dot) coupled to a one-dimensional system (such as a micropillar cavity or a photonic nanocrystal waveguide). The analyzer works in a passive way and can completely distinguish $2^n$ Greenberger-Horne-Zeilinger~(GHZ) states of $n$ photons without using any active operation or fast switching. The efficiency and fidelity of the GHZ-state analysis can, in principle, be close to unity, when an ideal single-photon scattering condition is fulfilled. For a nonideal scattering, which typically reduces the fidelity of a GHZ-state analysis, we introduce a passively error-rejecting circuit to enable a near-perfect fidelity at the expense of a slight decrease of its efficiency. Furthermore, the protocol can be directly used to perform a two-photon Bell-state analysis. This passive, resource-efficient, and error-rejecting protocol can, therefore, be useful for practical quantum networks.

quant-ph↗

arXiv2019

Periodic orbits of linear Filippov systems with a line of discontinuity

In this paper we consider periodic orbits of planar linear Filippov systems with a line of discontinuity. Unlike many publications researching only the maximum number of crossing periodic orbits, we investigate not only the number and configuration of sliding periodic orbits, but also the coexistence of sliding periodic orbits and crossing ones. Firstly, we prove that the number of sliding periodic orbits is at most 2, and give all possible configurations of one or two sliding periodic orbits. Secondly, we prove that two sliding periodic orbits coexist with at most one crossing periodic orbit, and one sliding periodic orbit can coexist with two crossing ones.

math.DS↗

arXiv2019

Component Attention Guided Face Super-Resolution Network: CAGFace

To make the best use of the underlying structure of faces, the collective information through face datasets and the intermediate estimates during the upsampling process, here we introduce a fully convolutional multi-stage neural network for 4$\times$ super-resolution for face images. We implicitly impose facial component-wise attention maps using a segmentation network to allow our network to focus on face-inherent patterns. Each stage of our network is composed of a stem layer, a residual backbone, and spatial upsampling layers. We recurrently apply stages to reconstruct an intermediate image, and then reuse its space-to-depth converted versions to bootstrap and enhance image quality progressively. Our experiments show that our face super-resolution method achieves quantitatively superior and perceptually pleasing results in comparison to state of the art.

cs.CV↗

arXiv2019

Crossing limit cycles of nonsmooth Liénard systems and applications

Continuing the investigation for the number of crossing limit cycles of nonsmooth Liénard systems in [Nonlinearity 21(2008), 2121-2142] for the case of a unique equilibrium, in this paper we consider the case of any number of equilibria. We give results about the existence and uniqueness of crossing limit cycles, which hold not only for a unique equilibrium but also for multiple equilibria. Moreover, we find a sufficient condition for the nonexistence of crossing limit cycles. Finally, applying our results we prove the uniqueness of crossing limit cycles for planar piecewise linear systems with a line of discontinuity and without sliding sets.

math.DS↗

arXiv2019

BTS-DSN: Deeply Supervised Neural Network with Short Connections for Retinal Vessel Segmentation

Background and Objective: The condition of vessel of the human eye is an important factor for the diagnosis of ophthalmological diseases. Vessel segmentation in fundus images is a challenging task due to complex vessel structure, the presence of similar structures such as microaneurysms and hemorrhages, micro-vessel with only one to several pixels wide, and requirements for finer results. Methods:In this paper, we present a multi-scale deeply supervised network with short connections (BTS-DSN) for vessel segmentation. We used short connections to transfer semantic information between side-output layers. Bottom-top short connections pass low level semantic information to high level for refining results in high-level side-outputs, and top-bottom short connection passes much structural information to low level for reducing noises in low-level side-outputs. In addition, we employ cross-training to show that our model is suitable for real world fundus images. Results: The proposed BTS-DSN has been verified on DRIVE, STARE and CHASE_DB1 datasets, and showed competitive performance over other state-of-the-art methods. Specially, with patch level input, the network achieved 0.7891/0.8212 sensitivity, 0.9804/0.9843 specificity, 0.9806/0.9859 AUC, and 0.8249/0.8421 F1-score on DRIVE and STARE, respectively. Moreover, our model behaves better than other methods in cross-training experiments. Conclusions: BTS-DSN achieves competitive performance in vessel segmentation task on three public datasets. It is suitable for vessel segmentation. The source code of our method is available at https://github.com/guomugong/BTS-DSN.

cs.CV↗

arXiv2019

A Logic-Driven Framework for Consistency of Neural Models

While neural models show remarkable accuracy on individual predictions, their internal beliefs can be inconsistent across examples. In this paper, we formalize such inconsistency as a generalization of prediction error. We propose a learning framework for constraining models using logic rules to regularize them away from inconsistency. Our framework can leverage both labeled and unlabeled examples and is directly compatible with off-the-shelf learning schemes without model redesign. We instantiate our framework on natural language inference, where experiments show that enforcing invariants stated in logic can help make the predictions of neural models both accurate and consistent.

cs.AI↗

arXiv2019

Entanglement enhanced distinguishability of coherence-breaking channels

Originated from the superposition principle in quantum mechanics, coherence has been extensively studied as a kind important resource in quantum information processing. We investigate the distinguishability of coherence-breaking channels with the help of quantum entanglement. By explicitly computing the minimal error probability of channel discrimination, it is shown that entanglement can enhance the capacity of coherence-breaking channel distinguishability with same types for some cases while cannot enhanced for some other cases. For coherence-breaking channels with different types, the channel distinguishability cannot be enhanced via entanglement.

quant-ph↗

arXiv2019

On Convergence Rate of Adaptive Multiscale Value Function Approximation For Reinforcement Learning

In this paper, we propose a generic framework for devising an adaptive approximation scheme for value function approximation in reinforcement learning, which introduces multiscale approximation. The two basic ingredients are multiresolution analysis as well as tree approximation. Starting from simple refinable functions, multiresolution analysis enables us to construct a wavelet system from which the basis functions are selected adaptively, resulting in a tree structure. Furthermore, we present the convergence rate of our multiscale approximation which does not depend on the regularity of basis functions.

cs.LG↗