arXiv subjects

Xin Wang

Publications and source records attributed to Xin Wang.

At least 1,657 records · Page 92Linked to original sources

arXiv2018

XL-NBT: A Cross-lingual Neural Belief Tracking Framework

Task-oriented dialog systems are becoming pervasive, and many companies heavily rely on them to complement human agents for customer service in call centers. With globalization, the need for providing cross-lingual customer support becomes more urgent than ever. However, cross-lingual support poses great challenges---it requires a large amount of additional annotated data from native speakers. In order to bypass the expensive human annotation and achieve the first step towards the ultimate goal of building a universal dialog system, we set out to build a cross-lingual state tracking framework. Specifically, we assume that there exists a source language with dialog belief tracking annotations while the target languages have no annotated dialog data of any form. Then, we pre-train a state tracker for the source language as a teacher, which is able to exploit easy-to-access parallel data. We then distill and transfer its own knowledge to the student state tracker in target languages. We specifically discuss two types of common parallel resources: bilingual corpus and bilingual dictionary, and design different transfer learning strategies accordingly. Experimentally, we successfully use English state tracker as the teacher to transfer its knowledge to both Italian and German trackers and achieve promising results.

cs.CL↗

arXiv2018

Some intriguing upper bounds for separating hash families

An $N\times n$ matrix on $q$ symbols is called $\{w_1,\ldots,w_t\}$-separating if for arbitrary $t$ pairwise disjoint column sets $C_1,\ldots,C_t$ with $|C_i|=w_i$ for $1\le i\le t$, there exists a row $f$ such that $f(C_1),\ldots,f(C_t)$ are also pairwise disjoint, where $f(C_i)$ denotes the collection of components of $C_i$ restricted to row $f$. Given integers $N,q$ and $w_1,\ldots,w_t$, denote by $C(N,q,\{w_1,\ldots,w_t\})$ the maximal $n$ such that a corresponding matrix does exist. The determination of $C(N,q,\{w_1,\ldots,w_t\})$ has received remarkable attentions during the recent years. The main purpose of this paper is to introduce two novel methodologies to attack the upper bound of $C(N,q,\{w_1,\ldots,w_t\})$. The first one is a combination of the famous graph removal lemma in extremal graph theory and a Johnson-type recursive inequality in coding theory, and the second one is the probabilistic method. As a consequence, we obtain several intriguing upper bounds for some parameters of $C(N,q,\{w_1,\ldots,w_t\})$, which significantly improve the previously known results.

math.CO↗

arXiv2018

Two-color electromagnetically induced transparency via modulated coupling between a mechanical resonator and a qubit

We discuss level splitting and sideband transitions induced by a modulated coupling between a superconducting quantum circuit and a nanomechanical resonator. First, we show how to achieve an unconventional time-dependent longitudinal coupling between a flux (transmon) qubit and the resonator. Considering a sinusoidal modulation of the coupling strength, we find that a first-order sideband transition can be split into two. Moreover, under the driving of a red-detuned field, we discuss the optical response of the qubit for a resonant probe field. We show that level splitting induced by modulating this longitudinal coupling can enable two-color electromagnetically induced transparency (EIT), in addition to single-color EIT. In contrast to standard predictions of two-color EIT in atomic systems, we apply here only a single drive (control) field. The monochromatic modulation of the coupling strength is equivalent to employing two eigenfrequency-tunable mechanical resonators. Both drive-probe detuning for single-color EIT and the distance between transparent windows for two-color EIT, can be adjusted by tuning the modulation frequency of the coupling.

quant-ph↗

arXiv2018

S3D: Single Shot multi-Span Detector via Fully 3D Convolutional Networks

In this paper, we present a novel Single Shot multi-Span Detector for temporal activity detection in long, untrimmed videos using a simple end-to-end fully three-dimensional convolutional (Conv3D) network. Our architecture, named S3D, encodes the entire video stream and discretizes the output space of temporal activity spans into a set of default spans over different temporal locations and scales. At prediction time, S3D predicts scores for the presence of activity categories in each default span and produces temporal adjustments relative to the span location to predict the precise activity duration. Unlike many state-of-the-art systems that require a separate proposal and classification stage, our S3D is intrinsically simple and dedicatedly designed for single-shot, end-to-end temporal activity detection. When evaluating on THUMOS'14 detection benchmark, S3D achieves state-of-the-art performance and is very efficient and can operate at 1271 FPS.

cs.CV↗

arXiv2018

Investigating accuracy of pitch-accent annotations in neural network-based speech synthesis and denoising effects

We investigated the impact of noisy linguistic features on the performance of a Japanese speech synthesis system based on neural network that uses WaveNet vocoder. We compared an ideal system that uses manually corrected linguistic features including phoneme and prosodic information in training and test sets against a few other systems that use corrupted linguistic features. Both subjective and objective results demonstrate that corrupted linguistic features, especially those in the test set, affected the ideal system's performance significantly in a statistical sense due to a mismatched condition between the training and test sets. Interestingly, while an utterance-level Turing test showed that listeners had a difficult time differentiating synthetic speech from natural speech, it further indicated that adding noise to the linguistic features in the training set can partially reduce the effect of the mismatch, regularize the model, and help the system perform better when linguistic features of the test set are noisy.

eess.AS↗

arXiv2018

Probabilistic distillation of quantum coherence

The ability to distill quantum coherence is pivotal for optimizing the performance of quantum technologies; however, such a task cannot always be accomplished with certainty. Here we develop a general framework of probabilistic distillation of quantum coherence in a one-shot setting, establishing fundamental limitations for different classes of free operations. We first provide a geometric interpretation for the maximal success probability, showing that under maximally incoherent operations (MIO) and dephasing-covariant incoherent operations (DIO) the problem can be simplified into efficiently computable semidefinite programs. Exploiting these results, we find that DIO and its subset of strictly incoherent operations (SIO) have equal power in probabilistic distillation of coherence from pure input states, while MIO are strictly stronger. We then prove a fundamental no-go result: distilling coherence from any full-rank state is impossible even probabilistically. We further find that in some conditions the maximal success probability can vanish suddenly beyond a certain threshold in the distillation fidelity. Finally, we consider probabilistic coherence distillation assisted by a catalyst and demonstrate, with specific examples, its superiority to the unassisted and deterministic cases.

quant-ph↗

arXiv2018

PSDF Fusion: Probabilistic Signed Distance Function for On-the-fly 3D Data Fusion and Scene Reconstruction

We propose a novel 3D spatial representation for data fusion and scene reconstruction. Probabilistic Signed Distance Function (Probabilistic SDF, PSDF) is proposed to depict uncertainties in the 3D space. It is modeled by a joint distribution describing SDF value and its inlier probability, reflecting input data quality and surface geometry. A hybrid data structure involving voxel, surfel, and mesh is designed to fully exploit the advantages of various prevalent 3D representations. Connected by PSDF, these components reasonably cooperate in a consistent frame- work. Given sequential depth measurements, PSDF can be incrementally refined with less ad hoc parametric Bayesian updating. Supported by PSDF and the efficient 3D data representation, high-quality surfaces can be extracted on-the-fly, and in return contribute to reliable data fu- sion using the geometry information. Experiments demonstrate that our system reconstructs scenes with higher model quality and lower redundancy, and runs faster than existing online mesh generation systems.

cs.RO↗

arXiv2018

CUBE: An Information-optimized parallel Cosmological $N$-body Algorithm

Cosmological large scale structure $N$-body simulations are computation-light, memory-heavy problems in supercomputing. The considerable amount of memory is usually dominated by an inefficient way of storing more than sufficient phase space information of particles. We present a new parallel, information-optimized, particle-mesh-based $N$-body code CUBE, in which information-efficiency and memory-efficiency are increased by nearly an order of magnitude. This is accomplished by storing particle's relative phase space coordinates instead of global values, and in the format of fixed point as light as 1 byte. The remaining information is given by complementary density and velocity fields (negligible in memory space) and proper ordering of particles (no extra memory). Our numerical experiments show that this information-optimized $N$-body algorithm provides accurate results within the error of the particle-mesh algorithm. This significant lowering of the memory-to-computation ratio breaks the bottleneck of scaling up and speeding up large cosmological $N$-body simulations on multi-core and heterogeneous computing systems.

astro-ph.CO↗

arXiv2018

Look Before You Leap: Bridging Model-Free and Model-Based Reinforcement Learning for Planned-Ahead Vision-and-Language Navigation

Existing research studies on vision and language grounding for robot navigation focus on improving model-free deep reinforcement learning (DRL) models in synthetic environments. However, model-free DRL models do not consider the dynamics in the real-world environments, and they often fail to generalize to new scenes. In this paper, we take a radical approach to bridge the gap between synthetic studies and real-world practices---We propose a novel, planned-ahead hybrid reinforcement learning model that combines model-free and model-based reinforcement learning to solve a real-world vision-language navigation task. Our look-ahead module tightly integrates a look-ahead policy model with an environment model that predicts the next state and the reward. Experimental results suggest that our proposed method significantly outperforms the baselines and achieves the best on the real-world Room-to-Room dataset. Moreover, our scalable method is more generalizable when transferring to unseen environments.

cs.CV↗

arXiv2018

SkipNet: Learning Dynamic Routing in Convolutional Networks

While deeper convolutional networks are needed to achieve maximum accuracy in visual perception tasks, for many inputs shallower networks are sufficient. We exploit this observation by learning to skip convolutional layers on a per-input basis. We introduce SkipNet, a modified residual network, that uses a gating network to selectively skip convolutional blocks based on the activations of the previous layer. We formulate the dynamic skipping problem in the context of sequential decision making and propose a hybrid learning algorithm that combines supervised learning and reinforcement learning to address the challenges of non-differentiable skipping decisions. We show SkipNet reduces computation by 30-90% while preserving the accuracy of the original model on four benchmark datasets and outperforms the state-of-the-art dynamic networks and static compression methods. We also qualitatively evaluate the gating policy to reveal a relationship between image scale and saliency and the number of layers skipped.

cs.CV↗

arXiv2018

Understanding the Reconstruction of the Biased Tracer

Recent development in the reconstruction of the large-scale structure (LSS) has seen significant improvement in restoring the linear baryonic acoustic oscillation (BAO) from at least the non-linear matter field. This outstanding performance is achieved by iteratively solving the Monge-Ampere equation of the mass conservation. However, this technique also relies on several assumptions that are not valid in reality, namely the longitudinal displacement, the absence of shell-crossing and homogeneous initial condition. In particular, the conservation equation of the tracers comprises the biasing information that breaks down the last assumption. Consequently, direct reconstruction would entangle the non-linear displacement with complicated bias parameters and further affect the BAO. In this paper, we formulate a theoretical model describing the reconstructed biased map by matching the tracer overdensity with an auxiliary fluid with vanishing initial perturbation. Regarding the performance of the reconstruction algorithm, we show that even though the shot noise is still the most significant limiting factors in a realistic survey, inappropriate treatment of the bias could also shift the reconstructed frame and therefore broaden the BAO peak. We suggest that, in principle, this bias-related BAO smearing effect could be used to independently self-calibrate the bias parameters.

astro-ph.CO↗

arXiv2018

Semidefinite programming strong converse bounds for classical capacity

We investigate the classical communication over quantum channels when assisted by no-signaling (NS) and positive-partial-transpose-preserving (PPT) codes, for which both the optimal success probability of a given transmission rate and the one-shot $ε$-error capacity are formalized as semidefinite programs (SDPs). Based on this, we obtain improved SDP finite blocklength converse bounds of general quantum channels for entanglement-assisted codes and unassisted codes. Furthermore, we derive two SDP strong converse bounds for the classical capacity of general quantum channels: for any code with a rate exceeding either of the two bounds of the channel, the success probability vanishes exponentially fast as the number of channel uses increases. In particular, applying our efficiently computable bounds, we derive an improved upper bound on the classical capacity of the amplitude damping channel. We also establish the strong converse property for the classical and private capacities of a new class of quantum channels. We finally study the zero-error setting and provide efficiently computable upper bounds on the one-shot zero-error capacity of a general quantum channel.

quant-ph↗

arXiv2018

No Metrics Are Perfect: Adversarial Reward Learning for Visual Storytelling

Though impressive results have been achieved in visual captioning, the task of generating abstract stories from photo streams is still a little-tapped problem. Different from captions, stories have more expressive language styles and contain many imaginary concepts that do not appear in the images. Thus it poses challenges to behavioral cloning algorithms. Furthermore, due to the limitations of automatic metrics on evaluating story quality, reinforcement learning methods with hand-crafted rewards also face difficulties in gaining an overall performance boost. Therefore, we propose an Adversarial REward Learning (AREL) framework to learn an implicit reward function from human demonstrations, and then optimize policy search with the learned reward function. Though automatic eval- uation indicates slight performance boost over state-of-the-art (SOTA) methods in cloning expert behaviors, human evaluation shows that our approach achieves significant improvement in generating more human-like stories than SOTA systems.

cs.CL↗

arXiv2018

One-shot coherence distillation

We characterize the distillation of quantum coherence in the one-shot setting, that is, the conversion of general quantum states into maximally coherent states under different classes of quantum operations. We show that the maximally incoherent operations (MIO) and the dephasing-covariant incoherent operations (DIO) have the same power in the task of one-shot coherence distillation. We establish that the one-shot distillable coherence under MIO and DIO is efficiently computable with a semidefinite program, which we show to correspond to a quantum hypothesis testing problem. Further, we introduce a family of coherence monotones generalizing the robustness of coherence as well as the modified trace distance of coherence, and show that they admit an operational interpretation in characterizing the fidelity of distillation under different classes of operations. By providing an explicit formula for these quantities for pure states, we show that the one-shot distillable coherence under MIO, DIO, strictly incoherent operations (SIO), and incoherent operations (IO) is equal for all pure states.

quant-ph↗

arXiv2018

Geometric ergodicity of Polya-Gamma Gibbs sampler for Bayesian logistic regression with a flat prior

The logistic regression model is the most popular model for analyzing binary data. In the absence of any prior information, an improper flat prior is often used for the regression coefficients in Bayesian logistic regression models. The resulting intractable posterior density can be explored by running Polson et al.'s (2013) data augmentation (DA) algorithm. In this paper, we establish that the Markov chain underlying Polson et al.'s (2013) DA algorithm is geometrically ergodic. Proving this theoretical result is practically important as it ensures the existence of central limit theorems (CLTs) for sample averages under a finite second moment condition. The CLT in turn allows users of the DA algorithm to calculate standard errors for posterior estimates.

math.ST↗

arXiv2018

Constructions of Augmented Orthogonal Arrays

Augmented orthogonal arrays (AOAs) were introduced by Stinson, who showed the equivalence between ideal ramp schemes and augmented orthogonal arrays (Discrete Math. 341 (2018), 299-307). In this paper, we show that there is an AOA$(s,t,k,v)$ if and only if there is an OA$(t,k,v)$ which can be partitioned into $v^{t-s}$ subarrays, each being an OA$(s,k,v)$, and that there is a linear AOA$(s,t,k,q)$ if and only if there is a linear maximum distance separable (MDS) code of length $k$ and dimension $t$ over $\mathbb{F}_q$ which contains a linear MDS subcode of length $k$ and dimension $s$ over $\mathbb{F}_q$. Some constructions for AOAs and some new infinite classes of AOAs are also given.

math.CO↗

arXiv2018

IDK Cascades: Fast Deep Learning by Learning not to Overthink

Advances in deep learning have led to substantial increases in prediction accuracy but have been accompanied by increases in the cost of rendering predictions. We conjecture that fora majority of real-world inputs, the recent advances in deep learning have created models that effectively "overthink" on simple inputs. In this paper, we revisit the classic question of building model cascades that primarily leverage class asymmetry to reduce cost. We introduce the "I Don't Know"(IDK) prediction cascades framework, a general framework to systematically compose a set of pre-trained models to accelerate inference without a loss in prediction accuracy. We propose two search based methods for constructing cascades as well as a new cost-aware objective within this framework. The proposed IDK cascade framework can be easily adopted in the existing model serving systems without additional model re-training. We evaluate the proposed techniques on a range of benchmarks to demonstrate the effectiveness of the proposed framework.

cs.CV↗

arXiv2018

Plasmonic multiple exciton generation

We show that bi-exciton formation can be highly efficient in a solar cell with the semiconductor absorber filled with an array of metallic nanoparticles having plasmonic resonance tuned to the semiconductor gap energy. This process can be viewed as plasmon-enhanced multiple exciton generation (PMEG), with the resulting cell efficiency exceeding the Shockley-Queisser limit. We demonstrate, that efficiency of the PMEG process, increases with decreasing of the semiconductor gap size, and illustrate that by considering in detail three systems with gradually decreasing gap size: GaAs, Si and Ge.

cond-mat.mtrl-sci↗