Searcharxiv⌕ Search

arXiv subjects

Kai Zhao

Publications and source records attributed to Kai Zhao.

At least 217 records · Page 12Linked to original sources
arXiv2019

Translate-to-Recognize Networks for RGB-D Scene Recognition

Cross-modal transfer is helpful to enhance modality-specific discriminative power for scene recognition. To this end, this paper presents a unified framework to integrate the tasks of cross-modal translation and modality-specific recognition, termed as Translate-to-Recognize Network (TRecgNet). Specifically, both translation and recognition tasks share the same encoder network, which allows to explicitly regularize the training of recognition task with the help of translation, and thus improve its final generalization ability. For translation task, we place a decoder module on top of the encoder network and it is optimized with a new layer-wise semantic loss, while for recognition task, we use a linear classifier based on the feature embedding from encoder and its training is guided by the standard cross-entropy loss. In addition, our TRecgNet allows to exploit large numbers of unlabeled RGB-D data to train the translation task and thus improve the representation power of encoder network. Empirically, we verify that this new semi-supervised setting is able to further enhance the performance of recognition network. We perform experiments on two RGB-D scene recognition benchmarks: NYU Depth v2 and SUN RGB-D, demonstrating that TRecgNet achieves superior performance to the existing state-of-the-art methods, especially for recognition solely based on a single modality.

cs.CV↗
arXiv2018

When to Finish? Optimal Beam Search for Neural Text Generation (modulo beam size)

In neural text generation such as neural machine translation, summarization, and image captioning, beam search is widely used to improve the output text quality. However, in the neural generation setting, hypotheses can finish in different steps, which makes it difficult to decide when to end beam search to ensure optimality. We propose a provably optimal beam search algorithm that will always return the optimal-score complete hypothesis (modulo beam size), and finish as soon as the optimality is established (finishing no later than the baseline). To counter neural generation's tendency for shorter hypotheses, we also introduce a bounded length reward mechanism which allows a modified version of our beam search algorithm to remain optimal. Experiments on neural machine translation demonstrate that our principled beam search algorithm leads to improvement in BLEU score over previously proposed alternatives.

cs.CL↗
arXiv2018

Vanishing contact structure problem and convergence of the viscosity solutions

This paper is devoted to study the vanishing contact structure problem which is a generalization of the vanishing discount problem. Let $H^λ(x,p,u)$ be a family of Hamiltonians of contact type with parameter $λ>0$ and converges to $G(x,p)$. For the contact type Hamilton-Jacobi equation with respect to $H^λ$, we prove that, under mild assumptions, the associated viscosity solution $u^λ$ converges to a specific viscosity solution $u^0$ of the vanished contact equation. As applications, we give some convergence results for the nonlinear vanishing discount problem.

math.AP↗
arXiv2018

Hi-Fi: Hierarchical Feature Integration for Skeleton Detection

In natural images, the scales (thickness) of object skeletons may dramatically vary among objects and object parts, making object skeleton detection a challenging problem. We present a new convolutional neural network (CNN) architecture by introducing a novel hierarchical feature integration mechanism, named Hi-Fi, to address the skeleton detection problem. The proposed CNN-based approach has a powerful multi-scale feature integration ability that intrinsically captures high-level semantics from deeper layers as well as low-level details from shallower layers. % By hierarchically integrating different CNN feature levels with bidirectional guidance, our approach (1) enables mutual refinement across features of different levels, and (2) possesses the strong ability to capture both rich object context and high-resolution details. Experimental results show that our method significantly outperforms the state-of-the-art methods in terms of effectively fusing features from very different scales, as evidenced by a considerable performance improvement on several benchmarks.

cs.CV↗
arXiv2018

High Efficiency and Low Distortion Photoacoustic Effect in 3D Graphene Sponge

The conversion of light in sound plays a crucial role in spectroscopy, applied physics, and technology. In this paper, light sound conversion in 3D graphene sponge through a photothermoacoustic mechanism is reported. It is shown that the unique combination of mechanical, optical, and thermodynamic properties of graphene assembled in a 3D sponge structure allows an unprecedented high efficiency conversion independent of light wavelength from infrared to ultraviolet. As a first application of this effect, a photothermal based graphene sponge loudspeaker is demonstrated, providing a full digital operation for frequencies from acoustic to ultrasound. The present results suggest a new pathway for light generation and control of sound and ultrasound signals potentially usable in a variety of new technological applications from high fidelity loudspeaker and radiation detectors to medical devices.

physics.app-ph↗
arXiv2017

Urban Pulse: Capturing the Rhythm of Cities

Cities are inherently dynamic. Interesting patterns of behavior typically manifest at several key areas of a city over multiple temporal resolutions. Studying these patterns can greatly help a variety of experts ranging from city planners and architects to human behavioral experts. Recent technological innovations have enabled the collection of enormous amounts of data that can help in these studies. However, techniques using these data sets typically focus on understanding the data in the context of the city, thus failing to capture the dynamic aspects of the city. The goal of this work is to instead understand the city in the context of multiple urban data sets. To do so, we define the concept of an "urban pulse" which captures the spatio-temporal activity in a city across multiple temporal resolutions. The prominent pulses in a city are obtained using the topology of the data sets, and are characterized as a set of beats. The beats are then used to analyze and compare different pulses. We also design a visual exploration framework that allows users to explore the pulses within and across multiple cities under different conditions. Finally, we present three case studies carried out by experts from two different domains that demonstrate the utility of our framework.

cs.CY↗
arXiv2017

Deep Regression Forests for Age Estimation

Age estimation from facial images is typically cast as a nonlinear regression problem. The main challenge of this problem is the facial feature space w.r.t. ages is heterogeneous, due to the large variation in facial appearance across different persons of the same age and the non-stationary property of aging patterns. In this paper, we propose Deep Regression Forests (DRFs), an end-to-end model, for age estimation. DRFs connect the split nodes to a fully connected layer of a convolutional neural network (CNN) and deal with heterogeneous data by jointly learning input-dependant data partitions at the split nodes and data abstractions at the leaf nodes. This joint learning follows an alternating strategy: First, by fixing the leaf nodes, the split nodes as well as the CNN parameters are optimized by Back-propagation; Then, by fixing the split nodes, the leaf nodes are optimized by iterating a step-size free and fast-converging update rule derived from Variational Bounding. We verify the proposed DRFs on three standard age estimation benchmarks and achieve state-of-the-art results on all of them.

cs.CV↗
arXiv2017

OSU Multimodal Machine Translation System Report

This paper describes Oregon State University's submissions to the shared WMT'17 task "multimodal translation task I". In this task, all the sentence pairs are image captions in different languages. The key difference between this task and conventional machine translation is that we have corresponding images as additional information for each sentence pair. In this paper, we introduce a simple but effective system which takes an image shared between different languages, feeding it into the both encoding and decoding side. We report our system's performance for English-French and English-German with Flickr30K (in-domain) and MSCOCO (out-of-domain) datasets. Our system achieves the best performance in TER for English-German for MSCOCO dataset.

cs.CL↗
arXiv2017

A machine learning approach to drug repositioning based on drug expression profiles: Applications to schizophrenia and depression/anxiety disorders

Development of new medications is a very lengthy and costly process. Finding novel indications for existing drugs, or drug repositioning, can serve as a useful strategy to shorten the development cycle. In this study, we present an approach to drug discovery or repositioning by predicting indication for a particular disease based on expression profiles of drugs, with a focus on applications in psychiatry. Drugs that are not originally indicated for the disease but with high predicted probabilities serve as good candidates for repurposing. This framework is widely applicable to any chemicals or drugs with expression profiles measured, even if the drug targets are unknown. It is also highly flexible as virtually any supervised learning algorithms can be used. We applied this approach to identify repositioning opportunities for schizophrenia as well as depression and anxiety disorders. We applied various state-of-the-art machine learning (ML) approaches for prediction, including deep neural networks, support vector machines (SVM), elastic net, random forest and gradient boosted machines. The performance of the five approaches did not differ substantially, with SVM slightly outperformed the others. However, methods with lower predictive accuracy can still reveal literature-supported candidates that are of different mechanisms of actions. As a further validation, we showed that the repositioning hits are enriched for psychiatric medications considered in clinical trials. Notably, many top repositioning hits are supported by previous preclinical or clinical studies. Finally, we propose that ML approaches may provide a new avenue to explore drug mechanisms via examining the variable importance of gene features.

q-bio.GN↗
arXiv2017

Label Distribution Learning Forests

Label distribution learning (LDL) is a general learning framework, which assigns to an instance a distribution over a set of labels rather than a single label or multiple labels. Current LDL methods have either restricted assumptions on the expression form of the label distribution or limitations in representation learning, e.g., to learn deep features in an end-to-end manner. This paper presents label distribution learning forests (LDLFs) - a novel label distribution learning algorithm based on differentiable decision trees, which have several advantages: 1) Decision trees have the potential to model any general form of label distributions by a mixture of leaf node predictions. 2) The learning of differentiable decision trees can be combined with representation learning. We define a distribution-based loss function for a forest, enabling all the trees to be learned jointly, and show that an update function for leaf node predictions, which guarantees a strict decrease of the loss function, can be derived by variational bounding. The effectiveness of the proposed LDLFs is verified on several LDL tasks and a computer vision application, showing significant improvements to the state-of-the-art LDL methods.

cs.LG↗
arXiv2017

Jointly Trained Sequential Labeling and Classification by Sparse Attention Neural Networks

Sentence-level classification and sequential labeling are two fundamental tasks in language understanding. While these two tasks are usually modeled separately, in reality, they are often correlated, for example in intent classification and slot filling, or in topic classification and named-entity recognition. In order to utilize the potential benefits from their correlations, we propose a jointly trained model for learning the two tasks simultaneously via Long Short-Term Memory (LSTM) networks. This model predicts the sentence-level category and the word-level label sequence from the stepwise output hidden representations of LSTM. We also introduce a novel mechanism of "sparse attention" to weigh words differently based on their semantic relevance to sentence-level classification. The proposed method outperforms baseline models on ATIS and TREC datasets.

cs.CL↗
arXiv2017

Joint Syntacto-Discourse Parsing and the Syntacto-Discourse Treebank

Discourse parsing has long been treated as a stand-alone problem independent from constituency or dependency parsing. Most attempts at this problem are pipelined rather than end-to-end, sophisticated, and not self-contained: they assume gold-standard text segmentations (Elementary Discourse Units), and use external parsers for syntactic features. In this paper we propose the first end-to-end discourse parser that jointly parses in both syntax and discourse levels, as well as the first syntacto-discourse treebank by integrating the Penn Treebank with the RST Treebank. Built upon our recent span-based constituency parser, this joint syntacto-discourse parser requires no preprocessing whatsoever (such as segmentation or feature extraction), achieves the state-of-the-art end-to-end discourse parsing accuracy.

cs.CL↗
arXiv2017

DeepSkeleton: Learning Multi-task Scale-associated Deep Side Outputs for Object Skeleton Extraction in Natural Images

Object skeletons are useful for object representation and object detection. They are complementary to the object contour, and provide extra information, such as how object scale (thickness) varies among object parts. But object skeleton extraction from natural images is very challenging, because it requires the extractor to be able to capture both local and non-local image context in order to determine the scale of each skeleton pixel. In this paper, we present a novel fully convolutional network with multiple scale-associated side outputs to address this problem. By observing the relationship between the receptive field sizes of the different layers in the network and the skeleton scales they can capture, we introduce two scale-associated side outputs to each stage of the network. The network is trained by multi-task learning, where one task is skeleton localization to classify whether a pixel is a skeleton pixel or not, and the other is skeleton scale prediction to regress the scale of each skeleton pixel. Supervision is imposed at different stages by guiding the scale-associated side outputs toward the groundtruth skeletons at the appropriate scales. The responses of the multiple scale-associated side outputs are then fused in a scale-specific way to detect skeleton pixels using multiple scales effectively. Our method achieves promising results on two skeleton extraction datasets, and significantly outperforms other competitors. Additionally, the usefulness of the obtained skeletons and scales (thickness) are verified on two object detection applications: Foreground object segmentation and object proposal detection.

cs.CV↗
arXiv2017

A Time Localization System in Smart Home Using Hierarchical Structure and Dynamic Frequency

Both GPS and WiFi based localization have been exploited in recent years, yet most researches focus on localizing at home without environment context. Besides, the near home or workplace area is complex and has little attention in smart home or IOT. Therefore, after exploring the realistic route in and out of building, we conducted a time localization system (TLS) based on off-the-shelf smart phones with WiFi identification. TLS can identify the received signal strength indication (RSSI) of home and construct radio map of users' time route without site survey. As designed to service the smart devices in home, TLS applies the time interval as the distance of positions and as the variables of WiFi environment to mark time points. Experimental results with real users show that TLS as a service system for timeline localization achieves a median accuracy of 70 seconds and is more robust compared with nearest neighbor localization approach.

cs.NI↗
arXiv2017

Textual Entailment with Structured Attentions and Composition

Deep learning techniques are increasingly popular in the textual entailment task, overcoming the fragility of traditional discrete models with hard alignments and logics. In particular, the recently proposed attention models (Rocktäschel et al., 2015; Wang and Jiang, 2015) achieves state-of-the-art accuracy by computing soft word alignments between the premise and hypothesis sentences. However, there remains a major limitation: this line of work completely ignores syntax and recursion, which is helpful in many traditional efforts. We show that it is beneficial to extend the attention model to tree nodes between premise and hypothesis. More importantly, this subtree-level attention reveals information about entailment relation. We study the recursive composition of this subtree-level entailment relation, which can be viewed as a soft version of the Natural Logic framework (MacCartney and Manning, 2009). Experiments show that our structured attention and entailment composition model can correctly identify and infer entailment relations from the bottom up, and bring significant improvements in accuracy.

cs.CL↗
arXiv2016

Network Analysis of Urban Traffic with Big Bus Data

Urban traffic analysis is crucial for traffic forecasting systems, urban planning and, more recently, various mobile and network applications. In this paper, we analyse urban traffic with network and statistical methods. Our analysis is based on one big bus dataset containing 45 million bus arrival samples in Helsinki. We mainly address following questions: 1. How can we identify the areas that cause most of the traffic in the city? 2. Why there is a urban traffic? Is bus traffic a key cause of the urban traffic? 3. How can we improve the urban traffic systems? To answer these questions, first, the betweenness is used to identify the most import areas that cause most traffics. Second, we find that bus traffic is not an important cause of urban traffic using statistical methods. We differentiate the urban traffic and the bus traffic in a city. We use bus delay as an identification of the urban traffic, and the number of bus as an identification of the bus traffic. Third, we give our solutions on how to improve urban traffic by the traffic simulation on road networks. We show that adding more buses during the peak time and providing better bus schedule plan in the hot areas like railway station, metro station, shopping malls etc. will reduce the urban traffic.

physics.soc-ph↗
arXiv2016

The production of unknown neutron-rich isotopes in $^{238}$U+$^{238}$U collisions at near-barrier energy

The production cross sections for primary and residual fragments with charge number from $Z$=70 to 120 produced in the collision of $^{238}$U+$^{238}$U at 7.0 MeV/nucleon are calculated by the improved quantum molecular dynamics (ImQMD) model incorporated with the statistical evaporation model (HIVAP code). The calculation results predict that about sixty unknown neutron-rich isotopes from element Ra ($Z$=88) to Db ($Z$=105) can be produced with the production cross sections above the lower bound of $10^{-8}$ mb in this reaction. And almost all of unknown neutron-rich isotopes are emitted at the laboratory angles $θ_{lab}\leq$ 60$^\circ$. Two cases, i.e. the production of the unknown uranium isotopes with $A\geq$ 244 and that of rutherfordium with $A\geq$ 269 are investigated for understanding the production mechanism of unknown neutron-rich isotopes. It is found that for the former case the collision time between two uranium nuclei is shorter and the primary fragments producing the residues have smaller excitation energies of $\leq$ 30 MeV and the outgoing angles of those residues cover a range of 30$^\circ$-60$^\circ$. For the later case, the longer collision time is needed for a large number of nucleons being transferred and thus it results in the higher excitation energies and smaller outgoing angles of primary fragments, and eventually results in a very small production cross section for the residues of Rf with $A\geq$ 269 which have a small interval of outgoing angles of $θ_{lab}$=40$^\circ$-50$^\circ$.

nucl-th↗