Search

This paper proposes an automatic method for detection and mapping of Espeletia plants from aerial images acquired by UAV drone. The proposed approach integrated a computer vision for automatic extraction of training zones and tested on three well-established machine learning algorithms to detect regions belonging to Espeletia plants. The main components of the method are: (i) data capture and preprocessing; (ii) automatic extraction of training zones; and (iii) classification procedure using machine learning algorithms. Experimental results show that the method can achieve accurate detection and mapping of Espeletia plants, with up to 98.3% accuracy.

Quantifying the importance of each training point to a learning task is a fundamental problem in machine learning and the estimated importance scores have been leveraged to guide a range of data workflows such as data summarization and domain adaption. One simple idea is to use the leave-one-out error of each training point to indicate its importance. Recent work has also proposed to use the Shapley value, as it defines a unique value distribution scheme that satisfies a set of appealing properties. However, calculating Shapley values is often expensive, which limits its applicability in real-world applications at scale. Multiple heuristics to improve the scalability of calculating Shapley values have been proposed recently, with the potential risk of compromising their utility in real-world applications. \textit{How well do existing data quantification methods perform on existing workflows? How do these methods compare with each other, empirically and theoretically? Must we sacrifice scalability for the utility in these workflows when using these methods?} In this paper, we conduct a novel theoretical analysis comparing the utility of different importance quantification methods, and report extensive experimental studies on existing and proposed workflows such as noisy label detection, watermark removal, data summarization, data acquisition, and domain adaptation. We show that Shapley value approximation based on a $K$NN surrogate over pre-trained feature embeddings obtains comparable utility with existing algorithms while achieving significant scalability improvement, often by orders of magnitude. Our theoretical analysis also justifies its advantage over the leave-one-out error. The code is available at \url{https://github.com/AI-secure/Shapley-Study}.

Blockchain has emerged as a promising solution for secure data storage and retrieval for decentralized applications. To scale blockchain systems, a prevailing approach is to employ a hybrid storage model, where only small meta-data are stored on-chain while the raw data are outsourced to an off-chain storage service provider. The key issue for query processing in such a system is the design of gas-efficient authenticated data structure (ADS) to authenticate the query results. In this paper, we study novel ADS schemes for authenticated keyword search in hybrid-storage blockchains. We first propose the Suppressed Merkle inverted (Merkleinv) index, which maintains only a partial ADS structure on-chain that can be securely updated with a logarithm-sized cryptographic proof. Moreover, we propose a Chameleon inverted (Chameleoninv) index that leverages the chameleon vector commitment to achieve a constant maintenance cost. It is further optimized with Bloom filters to enhance the query and verification performance. We prove the security of the proposed ADS schemes and evaluate their performance using real datasets on the Ethereum platform. Experimental results show that, compared to a baseline solution, the proposed Merkleinv and Chameleoninv indexes reduce the average on-chain maintenance cost from US$10.39 down to US$2.50 and US$0.24, respectively, without sacrificing much the query performance.

Data quality affects machine learning (ML) model performances, and data scientists spend considerable amount of time on data cleaning before model training. However, to date, there does not exist a rigorous study on how exactly cleaning affects ML -- ML community usually focuses on developing ML algorithms that are robust to some particular noise types of certain distributions, while database (DB) community has been mostly studying the problem of data cleaning alone without considering how data is consumed by downstream ML analytics. We propose a CleanML study that systematically investigates the impact of data cleaning on ML classification tasks. The open-source and extensible CleanML study currently includes 14 real-world datasets with real errors, five common error types, seven different ML models, and multiple cleaning algorithms for each error type (including both commonly used algorithms in practice as well as state-of-the-art solutions in academic literature). We control the randomness in ML experiments using statistical hypothesis testing, and we also control false discovery rate in our experiments using the Benjamini-Yekutieli (BY) procedure. We analyze the results in a systematic way to derive many interesting and nontrivial observations. We also put forward multiple research directions for researchers., published in ICDE 2021

With the development of wearable wireless sensor network technology, the technology has been applied in more and more fields, such as military, medical rescue, transportation and so on. Especially in the field of medical rescue, because of the advantages of wearable wireless sensor network, such as portability, fast and flexible networking, it has a very high application prospect. However, due to the particularity of medical rescue scene, the communication quality of wearable wireless sensor network is facing a huge challenge. When there is a fault, it needs to repair the fault link timely and accurately. In this paper, a link fault repair algorithm based on polygon Fermat point (LFRA) is proposed. Firstly, the algorithm can repair the fault link by inserting the relay node after the fault occurs. Secondly, the algorithm considers the energy consumption of the nodes involved in the repair process, eliminates the nodes that do not meet the energy requirements, and avoids multiple repairs due to lack of energy. The simulation results show that the proposed algorithm has the advantages of fewer inserted relay nodes and longer maximum communication time between nodes.

We demonstrate vCBIR, a verifiable search engine for Content-Based Image Retrieval. vCBIR allows a small or medium-sized enterprise to outsource its image database to a cloud-based service provider and ensures the integrity of query processing. Like other common data-as-a-service (DaaS) systems, vCBIR consists of three parties: (i) the image owner who outsources its database, (ii) the service provider who executes the authenticated query processing, and (iii) the client who issues search queries. By employing a novel query authentication scheme proposed in our prior work [4], the system not only supports cloud-based image retrieval, but also generates a cryptographic proof for each query, by which the client could verify the integrity of query results. During the demonstration, we will showcase the usage of vCBIR and also provide attendees interactive experience of verifying query results against an untrustworthy service provider through graphical user interface (GUI).

Distributed machine learning (ML) has triggered tremendous research interest in recent years. Stochastic gradient descent (SGD) is one of the most popular algorithms for training ML models, and has been implemented in almost all distributed ML systems, such as Spark MLlib, Petuum, MXNet, and TensorFlow. However, current implementations often incur huge communication and memory overheads when it comes to large models. One important reason for this inefficiency is the row-oriented scheme (RowSGD) that existing systems use to partition the training data, which forces them to adopt a centralized model management strategy that leads to vast amount of data exchange over the network.We propose a novel, column-oriented scheme (ColumnSGD) that partitions training data by columns rather than by rows. As a result, ML model can be partitioned by columns as well, leading to a distributed configuration where individual data and model partitions can be collocated on the same machine. Following this locality property, we develop a simple yet powerful computation framework that significantly reduces communication overheads and memory footprints compared to RowSGD, for large-scale ML models such as generalized linear models (GLMs) and factorization machines (FMs). We implement ColumnSGD on top of Apache Spark, and study its performance both analytically and experimentally. Experimental results on both public and real-world datasets show that ColumnSGD is up to 930 x faster than MLlib, 63 x faster than Petuum, and 14 x faster than MXNet.

In recent years, the Wearable Wireless Sensor Network (WWSN) has become one of the most popular disaster relief networks. On-Body network is a key part of WWSN, which is responsible for the communication between wearable wireless sensor nodes and the coordinator in WWSN. The routing algorithm for On-body network is related to the energy consumption of wearable wireless sensor nodes and even the lifetime of the network. Therefore, this paper proposes a topology variable routing algorithm (TVRA) in On-Body networks. Firstly, the algorithm designs multi-hop routing to prolong the network lifetime on the premise of energy balance under the normal state of the network. When the network is abnormal, a single hop routing is adopted to ensure that the abnormal data can be sent with the lowest delay. The simulation results show that the proposed TVRA algorithm has better performance in terms of lifetime and time delay.

With the explosive growth of online images and the popularity of search engines, a great demand has arisen for small and medium-sized enterprises to build and outsource large-scale image retrieval systems to cloud platforms. While reducing storage and retrieval burdens, enterprises are at risk of facing untrusted cloud service providers. In this paper, we take the first step in studying the problem of query authentication for large-scale image retrieval. Due to the large size of image files, the main challenges are to (i) design efficient authenticated data structures (ADSs) and (ii) balance search, communication, and verification complexities. To address these challenges, we propose two novel ADSs, the Merkle randomized k-d tree and the Merkle inverted index with cuckoo filters, to ensure the integrity of query results in each step of image retrieval. For each ADS, we develop corresponding search and verification algorithms on the basis of a series of systemic design strategies. Furthermore, we put together the ADSs and algorithms to design the final authentication scheme for image retrieval, which we name ImageProof. We also propose several optimization techniques to improve the performance of the proposed ImageProof scheme. Security analysis and extensive experiments are performed to show the robustness and efficiency of ImageProof.

Blockchain technology has attracted much attention due to the great success of the cryptocurrencies. Owing to its immutability property and consensus protocol, blockchain offers a new solution for trusted storage and computation services. To scale up the services, prior research has suggested a hybrid storage architecture, where only small meta-data are stored on-chain and the raw data are outsourced to off-chain storage. To protect data integrity, cryptographic proof can be constructed online for queries over the data stored in the system. However, the previous schemes only support simple key-value queries. In this paper, we take the first step toward studying authenticated range queries in the hybrid-storage blockchain. The key challenge lies in how to design an authenticated data structure (ADS) that can be efficiently maintained by the blockchain, in which a unique gas cost model is employed. By analyzing the performance of the existing techniques, we propose a novel ADS, called GEM^2-tree, which is not only gas-efficient but also effective in supporting authenticated queries. To further reduce the ADS maintenance cost without sacrificing much the query performance, we also propose an optimized structure, GEM^2*-tree, by designing a two-level index structure. Theoretical analysis and empirical evaluation validate the performance of the proposed ADSs.

In Tencent Inc., more than 80% of the data are extracted and transformed using Spark. However, the commonly used machine learning systems are TensorFlow, XGBoost, and Angel, whereas Spark MLlib, an official Spark package for machine learning, is seldom used. One reason for this ignorance is that it is generally believed that Spark is slow when it comes to distributed machine learning. Users therefore have to undergo the painful procedure of moving data in and out of Spark. The question why Spark is slow, however, remains elusive. In this paper, we study the performance of MLlib with a focus on training generalized linear models using gradient descent. Based on a detailed examination, we identify two bottlenecks in MLlib, i.e., pattern of model update and pattern of communication. To address these two bottlenecks, we tweak the implementation of MLlib with two state-of-the-art and well-known techniques, model averaging and AllReduce. We show that, the new system that we call MLlib*, can significantly improve over MLlib and achieve similar or even better performance than other specialized distributed machine learning systems (such as Petuum and Angel), on both public and Tencent's workloads.

Power grid is developing towards the direction of intelligence. Nevertheless, the giant grid scale, complicated structure and external natural factors result in the potential of large-scaled power failure. Traditional black start power gives priority to hydropower. In areas with shortage of hydro power resources, suitable black start power is insufficient. The key of black start is able to realize the safe and stable operation of wind power and energy storage cogeneration system. Due to the combination of wind turbines and energy storage is a small system, energy storage, as a black start power source, starts the wind turbines together to supply power for the remote load. Therefore, it is necessary to run the balance and stable operation of a small system. In this paper, PSCAD simulation software is employed to conduct modeling of wind power and energy storage system, and the self-starting process of wind storage cogeneration system is analyzed by using resistance sensor as the load. The result verifies the wind power storage system's initiation process.

Voltage exceeding becomes the key factor influencing the safe and stable operation of power grid with access to the distributed PV in large quantity. In order to adapt to the depth of penetration and to the access of large amount of distributed PV in order to elevate system's ability of voltage regulation, in this paper, it puts forward a new method for the dynamic control of pivot point in rural power grid. First, based on PV output features and characteristics of loads in rural power grid, it analyzes the features of the scaled PV output and the typical status of loads in rural power grid. It adopts the calculation applicable to the power grid with the access to the large amount of PV and analyzes the overvoltage of Bus in the power distribution grid. Secondly, based on the overvoltage of Bus in the rural power grid, it puts forward a new reverse voltage regulation measure, namely, to invest graded reactors into the pivot point in the rural power grid to absorb the reactive power. Finally, it establishes the scaled PV operation simulation platform for one rural power grid in Liaoning Province through ETAP to have verified the validity of the aforesaid voltage regulation strategy.

With the prevalence of location-based services and geo-functioned devices, the trend of spatial data outsourcing is rising. In the data outsourcing scenario, result integrity must be ensured by means of a query authentication scheme. However, most of the existing studies are confined to a centralized environment. In this paper, we investigate the query authentication problem in distributed environments and focus on the k nearest neighbor (kNN) query, which is widely used in spatial data analytics. We design a new distributed spatial authenticated data structure (ADS), distributed MR-tree, to facilitate efficient kNN processing. Furthermore, we propose a basic algorithm to process authenticated kNN queries based on the new ADS. Apart from the results, some verification objects are generated to guarantee the results' integrity. We also design two optimized algorithms to reduce the size of verification objects as well as the verification cost. Our experiments validate the good performance of the proposed techniques in terms of query cost, communication overhead, and verification time.

Ridesharing has been becoming increasingly popular in urban areas worldwide for its low cost and environmental friendliness. Much research attention has been drawn to the optimization of travel costs in shared rides. However, other important factors in ridesharing, such as the social comfort and trust issues, have not been fully considered in the existing works. In this paper, we formulate a new problem, named Assignment of Requests to Offers (ARO), that aims to maximize the number of served riders while satisfying the social comfort constraints as well as spatial-temporal constraints. We prove that the ARO problem is NP-hard. We then propose an exact algorithm for a simplified ARO problem. We further propose three pruning strategies to efficiently narrow down the searching space and speed up the assignment processing. Based on these pruning strategies, we develop two novel heuristic algorithms, the request-oriented approach and offer-oriented approach, to tackle the ARO problem. We also study the dynamic ARO problem and present a novel algorithm to tackle this problem. Through extensive experiments, we demonstrate the efficiency and effectiveness of our proposed approaches on real-world datasets.

We report a new physical mechanism, cycling excitation process (CEP), that produces intrinsic gain for photocurrent. The CEP process has shown much greater efficiency and lower noise than the impact ionization process, the mechanism for avalanche photodetectors. Using the CEP process, we have obtained detectors of low excess noise and a record high gain-bandwidth product of over 2 THz. Above all, such performance was obtained from a thin (30nm) layer of amorphous silicon as the gain medium.

Cycling excitation process (CEP) is a signal amplification mechanism showing high gain, low noise at low operation bias1,2. Depending on Auger excitation that involves localized states, CEP effect may be prominent in amorphous silicon, which has abundant localized states as a disordered material. Here we present a carbon-doped amorphous Si photodiode with ultra-high gain-bandwidth product with exceedingly low excess noise factor.

This paper introduces the three schemes of regenerative energy utilization in urban rail transit, i.e. the energy dissipation type, the energy storage type and the energy feedback type, and their application status. Comparative analysis of the three schemes is performed from multiple perspectives. Finally, this paper proposes that the line conditions, the vehicle type and other factors should be considered comprehensively when choosing which regenerative braking energy recovery devices should be used. In addition, the economic and technical comparison should also be made. Based on this, a reasonable allocation scheme can be got to achieve good economic benefit.

This paper defined a unified architecture of IoT system, on which IoT node model, virtual things, the basic service of things and overall hierarchical model of services had been described. IoT nodes must be connected directly to Internet and provide basic services of things in this architecture, and started from the base services which were compatible with Internet applications system of SOA, built the IoT application system with the middle layer of the Internet-based services and the base services of things, in order to implement a complete IoT system. this study also provides a practical application for monitoring water resources in the IoT way

Gradient boosting tree (GBT), a widely used machine learning algorithm, achieves state-of-the-art performance in academia, industry, and data analytics competitions. Although existing scalable systems which implement GBT, such as XGBoost and MLlib, perform well for datasets with medium-dimensional features, they can suffer performance degradation for many industrial applications where the trained datasets contain highdimensional features. The performance degradation derives from their inefficient mechanisms for model aggregation—either mapreduce or all-reduce. To address this high-dimensional problem, we propose a scalable execution plan using the parameter server architecture to facilitate the model aggregation. Further, we introduce a sparse-pull method and an efficient index structure to increase the processing speed. We implement a GBT system, namely TencentBoost, in the production cluster of Tencent Inc. The empirical results show that our system is 2-20× faster than existing platforms.

In the lossy wireless sensor network (WSN), packets loss can severely degrade the performance of data gathering algorithm based on compressive sensing (CS). Due to the limited retransmission cannot ensure the high reliability of the wireless link, the CS data gathering algorithm based on retransmission and temporal sequence correlation prediction (CS-RTSC) is proposed in this paper. When packets loss occur in the process of data gathering, the limited retransmissions is the primary choice to recover the packets. If the recovery is not successful, the temporal sequence correlation prediction is adopted. Simulation results show that CS-RTSC algorithm can guarantee the normal implementation of the CS data gathering under the unreliable links compared to existing CS data gathering algorithm and the relative error of the data reconstruction under 30% packet loss ratio is only 0.1% higher than that under the ideal links.

Android malware scanning services (e.g., VirusTotal) are websites that users submit suspicious Android programs and get an array of malware detection results. With the growing popularity of such websites, we suspect that, these services are not only used by innocent users, but also, malware writers for testing the evasion capability of their malware samples. May this hypothesis be true, it not only provides interesting insight on Android malware development (AMD), but also provides opportunities for important security applications such as zero-day sample detection. In this work, we first validate this hypothesis with massive data; then design a system AMDHunter to hunt for AMDs on VirusTotal that reveals new threats for Android that has never been revealed before. This is the first systematic study of the malware development phenomenon on VirusTotal, and the first system to automatically detect such malware development cases. AMDHunter has been used in a leading security company for months. Our study is driven by the large amount of data on VirusTotal— We analyzed 153 million submissions collected on VirusTotal during 102 days. Our system identifies 1,623 AMDs with 13,855 samples from 83 countries. We also performed case studies on 890 malware samples selected from the identified AMDs, which revealed lots of new threats, e.g., the development cases of fake system/banking phishing malware, new rooting exploits and etc.

Statistical wave theories have been developed over many years and have been applied to imaging, remote sensing, and communications for geophysical and biological media. There have been increasing interest in integrating statistical wave theories with other disciplines. For example, communication and signal processing through random media have largely studied without using the results of extensive studies on statistical wave theories even though physical problems that are considered are closely related. It is therefore important to integrate the results of statistical wave theories with signal processing. If we include stochastic Green's function and mutual coherence function (MCF) in the formulations of communication, the formulation of channel capacity can show clearly physical characteristics of random media, and the eigenvalues that include Statistical Green's Function and MCF.

Microwave power transmission has its unique advantages for its ability of supplying power to not-accessible and mobile receivers and is also called long-range wireless power transmission (WPT). The implementation of long-range WPT requires the phased array transmitter technology, in order to achieve the satisfactory transmission efficiency with high array gain and beam steering capability to mobile power receivers. The passive electronically scannable array (PESA) transmitter is a popular low cost solution for its shared RF signal and independent phase-shifting in analog domain. However, when it comes to the large-scale array beamforming, the cost of high resolution phase shifter and high efficiency power amplifier makes PESA prohibitively expensive if it is applied to WPT system in the market of civilian electronics.

Optical imaging and detection systems use photodetectors to turn optical signals into electric signals and use electronic amplifiers to increase the signal intensity. If the detector can produces sufficient gain by itself, the sensitivity can no longer be limited by the thermal noise of electronics. Instead, the system sensitivity will be limited by the shot noise of the detector characterized by the excess noise factor. The most prominent example is avalanche photodetector (APD) where signal amplification is achieved by the intrinsic process of impact ionization. However, although APD based photoreceiver shows the best sensitivity for fiber optic communications, it has high excess noise and requires high operation voltage (50-200V for Si APDs). This presents the need for an entirely new intrinsic signal amplification mechanism in semiconductors, which leads to the discovery and invention of the cycling excitation process (CEP) mechanism. CEP for signal amplification can be realized in a heavily doped and highly compensated Si p-n junction as illustrated in Fig. 1(a), so that an energetic electron (hole) can excite an e-h pair from a localized impurity state via the Auger process as shown in Fig. 1(b). Note that the carrier in the localized impurity gives rise to a high uncertainty in momentum, thus relaxing the momentum conservation or k-selection rule. This new amplification process cannot be completed without the specific electron-phonon interaction shown in Fig. 1(c). Hence the gain is regulated by phonons, which provide the “intrinsic negative feedback” to stabilize the gain and reduce the excess noise.

It has been noted that angular and frequency correlations of the scattered wave from multiple scattering or rough surfaces exhibit the characteristics indicating that the scattered wave remembers the angular and frequency characteristics of incident wave under certain conditions. This is called the “Memory Effects” and was discussed by Feng in 1988. In July 2014 at URSI meeting in Memphis, we presented a preliminary study of the use of Memory Effects to reduce the clutter from rough surface and to improve the resolution of images of objects located near rough surface.

“Through-Wall Radar Imaging” (TWRI) has been studied extensively to reconstruct the image from the backscattered waves through the opaque dielectric wall. When the objects are located in the shadow region of the metallic wall (hard wall) where the waves can only bypass the wall via diffraction, a new problem arises as “Hard-Wall Radar Imaging” (HWRI), which has been proposed and discussed in our previous papers.

Statistical Electromagnetic Theories have been developed over many years and applied to a wide range of practical problems in remote sensing of geophysical media, biological media, medical optics, ultrasound imaging and object detection and communication in clutter. This paper gives a review of recent developments in applications of statistical wave theories. The super resolution of images occurs in random media due to the multiple scattering and the increase of apparent aperture size of the transmitter. Another interesting effect is that the scattered wave from multiple scattering remembers the direction of the incident wave and strong correlations can be observed under certain conditions. This is called the “Memory Effect”. Recent study shows that under certain conditions, the angular and frequency correlations of the scattered wave can be enhanced or reduced and this effect can be used to reduce the clutter from the rough surface. It is also noted that the coherence in multiple scattering causes the increase of radar cross sections in turbulence due to the interference between the forward and backward waves, called the “double passage effect”. This is related to the Anderson localization and the coherent backscattering.

In view of the appearance of error bits in the process of data transmission, the individual erroneous bit of real-time services do not affect the use of services packets. The fault-tolerant type recognition of real-time packet is the first problem in the process of restructuring. In this paper, a non-linear recognition method which combined “center distance” and BP net together is proposed and applied to factual data for erroneous packets, The results show that the method proposed performs much better than the regular method.