arXiv subjects

Rui Zhu

Publications and source records attributed to Rui Zhu.

At least 145 records · Page 8Linked to original sources

arXiv2021

Imitating Interactive Intelligence

A common vision from science fiction is that robots will one day inhabit our physical spaces, sense the world as we do, assist our physical labours, and communicate with us through natural language. Here we study how to design artificial agents that can interact naturally with humans using the simplification of a virtual environment. This setting nevertheless integrates a number of the central challenges of artificial intelligence (AI) research: complex visual perception and goal-directed physical control, grounded language comprehension and production, and multi-agent social interaction. To build agents that can robustly interact with humans, we would ideally train them while they interact with humans. However, this is presently impractical. Therefore, we approximate the role of the human with another learned agent, and use ideas from inverse reinforcement learning to reduce the disparities between human-human and agent-agent interactive behaviour. Rigorously evaluating our agents poses a great challenge, so we develop a variety of behavioural tests, including evaluation by humans who watch videos of agents or interact directly with them. These evaluations convincingly demonstrate that interactive training and auxiliary losses improve agent behaviour beyond what is achieved by supervised learning of actions alone. Further, we demonstrate that agent capabilities generalise beyond literal experiences in the dataset. Finally, we train evaluation models whose ratings of agents agree well with human judgement, thus permitting the evaluation of new agent models without additional effort. Taken together, our results in this virtual environment provide evidence that large-scale human behavioural imitation is a promising tool to create intelligent, interactive agents, and the challenge of reliably evaluating such agents is possible to surmount.

cs.LG↗

arXiv2020

Topological phase transitions driven by polarity change and next-nearest-neighbor hopping in skyrmion crystals

By considering the bulk-edge correspondence of the skyrmion crystal this work focused on the relation of the topological properties to the polarity ($Q_{\rm{sk}}$), next-nearest-neighbor hopping ($t'$), and exchange coupling ($J$). We found that by continually increasing polarity of the skyrmions from 1 to 2, the monopole-lattice phase and the dipole-lattice phase both span a wide range in the parameter space demonstrating topological robustness. While the next-nearest-neighbor-hopping of the electrons coexists with the nearest-neighbor hopping the flux pattern of the emergent magnetic field traversing the two-dimensional SkX is reshaped giving rise to different topological Hall effect. By increasing $t'$ from 0 to $t$ with the polarity fixed to be 1, we found Chern numbers of the electronic bands change from unity to indefinite values demonstrating a phase transition. We also considered the effect of finite $J$ and found the results obtained in the adiabatic limit hold to around $J=t$.

cond-mat.mes-hall↗

arXiv2020

Deep Keypoint-Based Camera Pose Estimation with Geometric Constraints

Estimating relative camera poses from consecutive frames is a fundamental problem in visual odometry (VO) and simultaneous localization and mapping (SLAM), where classic methods consisting of hand-crafted features and sampling-based outlier rejection have been a dominant choice for over a decade. Although multiple works propose to replace these modules with learning-based counterparts, most have not yet been as accurate, robust and generalizable as conventional methods. In this paper, we design an end-to-end trainable framework consisting of learnable modules for detection, feature extraction, matching and outlier rejection, while directly optimizing for the geometric pose objective. We show both quantitatively and qualitatively that pose estimation performance may be achieved on par with the classic pipeline. Moreover, we are able to show by end-to-end training, the key components of the pipeline could be significantly improved, which leads to better generalizability to unseen datasets compared to existing learning-based methods.

cs.CV↗

arXiv2020

Bounding The Number of Linear Regions in Local Area for Neural Networks with ReLU Activations

The number of linear regions is one of the distinct properties of the neural networks using piecewise linear activation functions such as ReLU, comparing with those conventional ones using other activation functions. Previous studies showed this property reflected the expressivity of a neural network family ([14]); as a result, it can be used to characterize how the structural complexity of a neural network model affects the function it aims to compute. Nonetheless, it is challenging to directly compute the number of linear regions; therefore, many researchers focus on estimating the bounds (in particular the upper bound) of the number of linear regions for deep neural networks using ReLU. These methods, however, attempted to estimate the upper bound in the entire input space. The theoretical methods are still lacking to estimate the number of linear regions within a specific area of the input space, e.g., a sphere centered at a training data point such as an adversarial example or a backdoor trigger. In this paper, we present the first method to estimate the upper bound of the number of linear regions in any sphere in the input space of a given ReLU neural network. We implemented the method, and computed the bounds in deep neural networks using the piece-wise linear active function. Our experiments showed that, while training a neural network, the boundaries of the linear regions tend to move away from the training data points. In addition, we observe that the spheres centered at the training data points tend to contain more linear regions than any arbitrary points in the input space. To the best of our knowledge, this is the first study of bounding linear regions around a specific data point. We consider our work as a first step toward the investigation of the structural complexity of deep neural networks in a specific input area.

cs.LG↗

arXiv2020

Slow propagation of 2 GHz acoustical waves in a suspended GaAs on insulator phononic waveguide

Optoelectronic oscillators have dominated the scene of microwave oscillators in the last few years thanks to their great performances regarding frequency stability and phase noise. However, miniaturization of such a device is an up to date challenge. Recently, devices based on phonon-photon interaction gather a lot of interest thanks to their extreme compactness and working frequency directly in the GHz. In this frame, a still missing element to obtain long-term frequency stability performances is an on-chip delay within the feedback loop. Here, we experimentally show filtering and slow propagation of 2 GHz acoustic waves on a Gallium Arsenide membrane heterogeneously integrated on silicon wafer. By engineering the dispersion of an acoustical waveguide, we evidence a group velocity below 1000 m/s for the mode able to propagate. Thus, an integrated delay implementation is at reach for potential improvement of opto-acoustic devices such as optomechanical oscillators or wireless applications.

physics.app-ph↗

arXiv2020

Microwave diagnostics of magnetic field strengths in solar flaring loops

We have performed microwave diagnostics of the magnetic field strengths in solar flare loops based on the theory of gyrosynchrotron emission. From Nobeyama Radioheliograph observations of three flare events at 17 and 34 GHz, we obtained the degree of circular polarization and the spectral index of microwave flux density, which were then used to map the magnetic field strengths in post-flare loops. Our results show that the magnetic field strength typically decreases from ~800 G near the loop footpoints to ~100 G at a height of 10--25 Mm. Comparison of our results with magnetic field modeling using a flux rope insertion method is also discussed. Our study demonstrates the potential of microwave imaging observations, even at only two frequencies, in diagnosing the coronal magnetic field of flaring regions.

astro-ph.SR↗

arXiv2020

Compression Algorithm Based on Irregular Sequence

The paper introduces a new lossless, highly robust compression algorithm that similar with LZW algorithm, yet the algorithm discards dictionary processing and uses irregular sequences with massive, random information instead. Then the paper found the ineffectiveness of the algorithm due to limited computing ability of hardware and made a few improvements to the algorithm. The algorithm is recommended to be applied in interplanetary communications between a high-compute-ability device and a low-compute-ability receiving device, whose signal would be easily interfered by cosmos rays.

eess.SP↗

arXiv2020

Impacts of COVID-19 control measures on tropospheric NO$_2$ over China, South Korea and Italy

Tropospheric nitrogen dioxide (NO$_2$) concentrations are strongly affected by anthropogenic activities. Using space-based measurements of tropospheric NO$_2$, here we investigate the responses of tropospheric NO$_2$ to the 2019 novel coronavirus (COVID-19) over China, South Korea, and Italy. We find noticeable reductions of tropospheric NO$_2$ columns due to the COVID-19 controls by more than 40% over E. China, South Korea, and N. Italy. The 40% reductions of tropospheric NO$_2$ are coincident with intensive lockdown events as well as up to 20% reductions in anthropogenic nitrogen oxides (NO$_x$) emissions. The perturbations in tropospheric NO$_2$ diminished accompanied with the mitigation of COVID-19 pandemic, and finally disappeared within around 50-70 days after the starts of control measures over all three nations, providing indications for the start, maximum, and mitigation of intensive controls. This work exhibits significant influences of lockdown measures on atmospheric environment, highlighting the importance of satellite observations to monitor anthropogenic activity changes.

physics.ao-ph↗

arXiv2020

SE-KGE: A Location-Aware Knowledge Graph Embedding Model for Geographic Question Answering and Spatial Semantic Lifting

Learning knowledge graph (KG) embeddings is an emerging technique for a variety of downstream tasks such as summarization, link prediction, information retrieval, and question answering. However, most existing KG embedding models neglect space and, therefore, do not perform well when applied to (geo)spatial data and tasks. For those models that consider space, most of them primarily rely on some notions of distance. These models suffer from higher computational complexity during training while still losing information beyond the relative distance between entities. In this work, we propose a location-aware KG embedding model called SE-KGE. It directly encodes spatial information such as point coordinates or bounding boxes of geographic entities into the KG embedding space. The resulting model is capable of handling different types of spatial reasoning. We also construct a geographic knowledge graph as well as a set of geographic query-answer pairs called DBGeo to evaluate the performance of SE-KGE in comparison to multiple baselines. Evaluation results show that SE-KGE outperforms these baselines on the DBGeo dataset for geographic logic query answering task. This demonstrates the effectiveness of our spatially-explicit model and the importance of considering the scale of different geographic entities. Finally, we introduce a novel downstream task called spatial semantic lifting which links an arbitrary location in the study area to entities in the KG via some relations. Evaluation on DBGeo shows that our model outperforms the baseline by a substantial margin.

cs.DB↗

arXiv2020

Semantically-Enriched Search Engine for Geoportals: A Case Study with ArcGIS Online

Many geoportals such as ArcGIS Online are established with the goal of improving geospatial data reusability and achieving intelligent knowledge discovery. However, according to previous research, most of the existing geoportals adopt Lucene-based techniques to achieve their core search functionality, which has a limited ability to capture the user's search intentions. To better understand a user's search intention, query expansion can be used to enrich the user's query by adding semantically similar terms. In the context of geoportals and geographic information retrieval, we advocate the idea of semantically enriching a user's query from both geospatial and thematic perspectives. In the geospatial aspect, we propose to enrich a query by using both place partonomy and distance decay. In terms of the thematic aspect, concept expansion and embedding-based document similarity are used to infer the implicit information hidden in a user's query. This semantic query expansion 1 2 G. Mai et al. framework is implemented as a semantically-enriched search engine using ArcGIS Online as a case study. A benchmark dataset is constructed to evaluate the proposed framework. Our evaluation results show that the proposed semantic query expansion framework is very effective in capturing a user's search intention and significantly outperforms a well-established baseline-Lucene's practical scoring function-with more than 3.0 increments in DCG@K (K=3,5,10).

cs.IR↗

arXiv2020

Multi-Scale Representation Learning for Spatial Feature Distributions using Grid Cells

Unsupervised text encoding models have recently fueled substantial progress in NLP. The key idea is to use neural networks to convert words in texts to vector space representations based on word positions in a sentence and their contexts, which are suitable for end-to-end training of downstream tasks. We see a strikingly similar situation in spatial analysis, which focuses on incorporating both absolute positions and spatial contexts of geographic objects such as POIs into models. A general-purpose representation model for space is valuable for a multitude of tasks. However, no such general model exists to date beyond simply applying discretization or feed-forward nets to coordinates, and little effort has been put into jointly modeling distributions with vastly different characteristics, which commonly emerges from GIS data. Meanwhile, Nobel Prize-winning Neuroscience research shows that grid cells in mammals provide a multi-scale periodic representation that functions as a metric for location encoding and is critical for recognizing places and for path-integration. Therefore, we propose a representation learning model called Space2Vec to encode the absolute positions and spatial relationships of places. We conduct experiments on two real-world geographic data for two different tasks: 1) predicting types of POIs given their positions and context, 2) image classification leveraging their geo-locations. Results show that because of its multi-scale representations, Space2Vec outperforms well-established ML approaches such as RBF kernels, multi-layer feed-forward nets, and tile embedding approaches for location modeling and image classification tasks. Detailed analysis shows that all baselines can at most well handle distribution at one scale but show poor performances in other scales. In contrast, Space2Vec's multi-scale representation can handle distributions at different scales.

cs.CV↗

arXiv2019

A note on the asymptotic behavior of the height for a birth-and-death process

This paper focuses on the asymptotic behaviors of the {\it height} for a birth-and-death process which related to a mean-field model \cite{FFS}(or the Anick-Mitra-Sondhi model \cite{DDM}). Recently, the asymptotic mean value of the height for the model is given in \cite{LAV}. In this paper, first, the asymptotic variance of the height is given, and as a consequence, a weak Law of Large Number for the height is obtained. Second, the centered and normalized height is proved to converge in distribution to a degenerate distribution, this indicates that the desired Central Limit Theorem fails.

math.PR↗

arXiv2019

3D Objectness Estimation via Bottom-up Regret Grouping

3D objectness estimation, namely discovering semantic objects from 3D scene, is a challenging and significant task in 3D understanding. In this paper, we propose a 3D objectness method working in a bottom-up manner. Beginning with over-segmented 3D segments, we iteratively group them into object proposals by learning an ingenious grouping predictor to determine whether two 3D segments can be grouped or not. To enhance robustness, a novel regret mechanism is presented to withdraw incorrect grouping operations. Hence the irreparable consequences brought by mistaken grouping in prior bottom-up works can be greatly reduced. Our experiments show that our method outperforms state-of-the-art 3D objectness methods with a small number of proposals in two difficult datasets, GMU-kitchen and CTD. Further ablation study also demonstrates the effectiveness of our grouping predictor and regret mechanism.

cs.CV↗

arXiv2019

Unconventional topological Hall effect in high-topological-number skyrmion crystals

Skyrmions with the topological number $Q$ equal an integer larger than 1 are called high-topological-number skyrmions or high-$Q$ skyrmions. In this work, we theoretically study the topological Hall effect in square-lattice high-$Q$ skyrmion crystals (SkX) with $Q=2$ and $Q=3$. As a result of the emergent magnetic field, Landau-level-like electronic band structure gives rise to quantized Hall conductivity when the Fermi energy is within the gaps between adjacent single band or multiple bands intertwined. We found that different from conventional ($Q=1$) SkX the Hall quantization number increases by $1/Q$ in average when the elevating Fermi energy crosses each band. We attribute the result to the fact that the Berry phase ${\cal{C}}$ is measured in the momentum space and the topological number of a single skyrmion $Q$ is measured in the real space. The reciprocality does not affect the conventional SkX because $Q=1=1/Q$.

cond-mat.mes-hall↗

arXiv2019

TransGCN:Coupling Transformation Assumptions with Graph Convolutional Networks for Link Prediction

Link prediction is an important and frequently studied task that contributes to an understanding of the structure of knowledge graphs (KGs) in statistical relational learning. Inspired by the success of graph convolutional networks (GCN) in modeling graph data, we propose a unified GCN framework, named TransGCN, to address this task, in which relation and entity embeddings are learned simultaneously. To handle heterogeneous relations in KGs, we introduce a novel way of representing heterogeneous neighborhood by introducing transformation assumptions on the relationship between the subject, the relation, and the object of a triple. Specifically, a relation is treated as a transformation operator transforming a head entity to a tail entity. Both translation assumption in TransE and rotation assumption in RotatE are explored in our framework. Additionally, instead of only learning entity embeddings in the convolution-based encoder while learning relation embeddings in the decoder as done by the state-of-art models, e.g., R-GCN, the TransGCN framework trains relation embeddings and entity embeddings simultaneously during the graph convolution operation, thus having fewer parameters compared with R-GCN. Experiments show that our models outperform the-state-of-arts methods on both FB15K-237 and WN18RR.

cs.LG↗

arXiv2019

Contextual Graph Attention for Answering Logical Queries over Incomplete Knowledge Graphs

Recently, several studies have explored methods for using KG embedding to answer logical queries. These approaches either treat embedding learning and query answering as two separated learning tasks, or fail to deal with the variability of contributions from different query paths. We proposed to leverage a graph attention mechanism to handle the unequal contribution of different query paths. However, commonly used graph attention assumes that the center node embedding is provided, which is unavailable in this task since the center node is to be predicted. To solve this problem we propose a multi-head attention-based end-to-end logical query answering model, called Contextual Graph Attention model(CGA), which uses an initial neighborhood aggregation layer to generate the center embedding, and the whole model is trained jointly on the original KG structure as well as the sampled query-answer pairs. We also introduce two new datasets, DB18 and WikiGeo19, which are rather large in size compared to the existing datasets and contain many more relation types, and use them to evaluate the performance of the proposed model. Our result shows that the proposed CGA with fewer learnable parameters consistently outperforms the baseline models on both datasets as well as Bio dataset.

cs.LG↗

arXiv2019

Android Malware Detection based on Factorization Machine

As the popularity of Android smart phones has increased in recent years, so too has the number of malicious applications. Due to the potential for data theft mobile phone users face, the detection of malware on Android devices has become an increasingly important issue in cyber security. Traditional methods like signature-based routines are unable to protect users from the ever-increasing sophistication and rapid behavior changes in new types of Android malware. Therefore, a great deal of effort has been made recently to use machine learning models and methods to characterize and generalize the malicious behavior patterns of mobile apps for malware detection. In this paper, we propose a novel and highly reliable classifier for Android Malware detection based on a Factorization Machine architecture and the extraction of Android app features from manifest files and source code. Our results indicate that the numerical feature representation of an app typically results in a long and highly sparse vector and that the interactions among different features are critical to revealing malicious behavior patterns. After performing an extensive performance evaluation, our proposed method achieved a test result of 100.00% precision score on the DREBIN dataset and 99.22% precision score with only 1.10% false positive rate on the AMD dataset. These metrics match the performance of state-of-the-art machine-learning-based Android malware detection methods and several commercial antivirus engines with the benefit of training up to 50 times faster.

cs.CR↗

arXiv2019

MMDetection: Open MMLab Detection Toolbox and Benchmark

We present MMDetection, an object detection toolbox that contains a rich set of object detection and instance segmentation methods as well as related components and modules. The toolbox started from a codebase of MMDet team who won the detection track of COCO Challenge 2018. It gradually evolves into a unified platform that covers many popular detection methods and contemporary modules. It not only includes training and inference codes, but also provides weights for more than 200 network models. We believe this toolbox is by far the most complete detection toolbox. In this paper, we introduce the various features of this toolbox. In addition, we also conduct a benchmarking study on different methods, components, and their hyper-parameters. We wish that the toolbox and benchmark could serve the growing research community by providing a flexible toolkit to reimplement existing methods and develop their own new detectors. Code and models are available at https://github.com/open-mmlab/mmdetection. The project is under active development and we will keep this document updated.

cs.CV↗