In response to increasing demands for largescale machine learning training jobs, many organizations have deployed GPU clusters across geographically distributed regions.However, existing Integer Linear Programming (ILP)-or genetic-based cross-cluster training approaches largely overlook the topology of decentralized clusters, lacking both topology-aware task scheduling mechanisms and automated model parallelization strategies.As a result, naively applying these optimization-based methods in cross-cluster settings leads to prohibitive scheduling overhead, due to the drastically enlarged search space induced by complex inter-cluster topologies.To address these challenges, we propose SpiderFlow, a topologyaware scheduling system specifically designed for decentralized GPU clusters.We formulate cross-cluster task scheduling as a graph optimization problem and introduce SpinSearch, a low-overhead topology-aware scheduling algorithm.In addition, for automated model parallelization, we propose Topology-aware Parallelism Automation (TPA), a two-level scheduling framework that combines heuristic methods at the inter-cluster level with ILP-based optimization within clusters, effectively reducing the search space while maintaining high training throughput with substantially lower scheduling overhead.We evaluate SpiderFlow on a physical platform comprising 8 decentralized clusters, as well as on a simulation platform with up to 64 decentralized clusters.Experimental results demonstrate that SpiderFlow reduces job completion time (JCT) by 1.2-1.3×,improves throughput by 1.12-1.25×,and reduces scheduling overhead by 20-90× on average compared to state-of-the-art scheduling systems.

Backward-compatible training enables the deployment of advanced models without requiring updates to old gallery databases. However, existing methods, including old-prototype-based (i.e., those relying on prototypes from the old model) and instance-based approaches, often overlook the impact of the old model's quality. High-quality old models exhibit compact intra-class feature distributions, which facilitate effective alignment between old and new models across various methods. In contrast, low-quality old models produce dispersed features, making it difficult for old-prototype-based methods to extract sufficient information. Additionally, instance-based methods are overly restrictive, limiting the flexibility of new models. In this work, we propose SA-BCT, an extremely simple yet effective backward-compatible training method that offers a unified framework for accommodating old models of varying quality. SA-BCT employs a single loss function applied to both old and new features, self-adaptively adjusting the constraint space for new features based on the distribution of old features. Extensive experiments in diverse settings demonstrate the effectiveness of SA-BCT. Code is available athttps://github.com/yuleung/SA-BCT.

The Low-Rate Denial of Service (LDoS) attack poses a significant threat to Internet services. Exploiting vulnerabilities in adaptive mechanisms embedded within network protocols, LDoS attacks are covert and exhibit legal behavior, making defense challenging. Existing LDoS attack solutions cannot perform real-time LDoS attack defense at line speed. With the emergence of P4, users can program the per-packet processing logic of the P4 switch, which offers us the chance to propose PLUTO, the first data plane-aware LDoS attack defense system built upon the P4 switch, possessing line-speed execution capacity. To meet the resource constraints of the P4 switch, we propose the time window-based pre-inference strategy to detect LDoS attacks and the time-limited per-flow state management to filter the LDoS attack flows. For the practical deployment, we develop the P4 Function Tool to extend the P4 primitives for more function operations. We also adopt an encoding-based mapping method to deploy the pre-inference model. Furthermore, we develop the async-updated hash table for quickly filtering LDoS attack flows. Compared with the baseline, PLUTO reduces the equal error rate (EER) by 27.96% and the average mitigation response time by 12.749 s, increasing the AUC by 1.83%, the F1 Score by 7.27%, and the Recall by 9.58%.

In the burgeoning realm of data-driven medical research, the escalating scale and intricacy of contemporary medical datasets frequently surpass the processing capabilities of traditional computational environments. Specifically, I/O bottlenecks have emerged as pivotal constraints in several research areas. In this paper, a data service framework is introduced that harnesses supercomputers to parallelly access multi-modal datasets and supports multi-node processing, called ParaMET. To enhance user accessibility, a web-based user interface has been integrated, allowing permitted researchers to effortlessly interact via their laptops, complemented by an API suite available through SDK for those adept with supercomputing for deeper data manipulation. Through extensive empirical validation, our framework manifests a remarkable performance elevation in multi-node supercomputing settings, achieving acceleration of up to approximately 1000x compared to existing methods.

Software-Defined Networking (SDN) is a new paradigm that facilitates network management by enabling programmability and disassociating the control plane from the data plane. SDN places the control plane into one or more controllers that take charge of the entire network. However, the logically centralized controller of SDN makes it subject to some security issues. Denial-of-Service (DoS) attacks are the main threat to SDN that can lead to impaired performance of the entire network. Low-rate Denial-of-Service (LDoS) attack is a variant of DoS attacks with a lower average attack rate and high concealability which is difficult to identify with traditional DDoS/DoS attack detection mechanisms. Additionally, existing LDoS attack detection and defense mechanisms often have weak real-time performance. To address this issue, we propose in this paper PeakSAX, a novel framework that can protect SDN against LDoS attacks in real-time by (1) Attack monitoring, (2) Traffic symbolization, (3) Malicious traffic identifying, (4) Attacker location, and (5) Mitigation strategy deployment. Simulation results show that PeakSAX can quickly identify and mitigate the impact of LDoS attacks about 4s, which improves over 70% compared to existing solutions.

Decision tree classification has become a prevailing technique for online diagnosis services. By outsourcing computation intensive tasks to a cloud server, cloud-assisted online diagnosis services are better ways for cases that the storage and computation requirements exceed the capability of medical institutions. With privacy concerns as well as intellectual property protection issues, the valuable diagnosis classifier and the sensitive user data should be protected against the cloud server. In this paper, we identify a work-flow for cloud-assisted online diagnosis services. We propose an efficient and secure decision tree classification scheme in the proposed work-flow. Specifically, the medical institution transforms a locally pre-trained decision tree classifier to a decision table, and later uses searchable symmetric encryption to encrypt the decision table. Then, the encrypted table is outsourced to the cloud server, and a user can submit encrypted physiological features to the cloud server and obtain an encrypted diagnosis prediction back. We provide formal security proofs to demonstrate that our scheme protects the confidentiality of the decision tree classifier and the user's data. The performance analysis shows that our scheme achieves faster-than-linear classification speed. Experimental evaluations show that our scheme requires several micro-seconds to process a diagnosis request in the tested datasets.

Rapidly evolving wireless technologies enable devices to directly exchange information without infrastructural support. In these device-to-device (D2D) communication scenarios, it is often difficult to setup cryptographic keys to initialize the secure communication, especially when the D2D connections are mobile and dynamic. This paper proposes an application layer solution scheme to bootstrap secure communications using the inherent randomness in the wireless transmissions. The proposed scheme is lightweight, easy to deploy and compatible with many physical layer wireless technologies. This paper contains security analysis to the scheme and conducts experiments to demonstrate its practicality.

Many chatbots have been developed that provide a multitude of services through a wide range of methods. A chatbot is a brand-new conversational agent in the highspeed changing technology world. With the advance of Artificial Intelligence and machine learning, chatbots are becoming more and more popular. A chatbot is the extension of human interface mediums such as the phone and social platforms. Similarly, Cryptocurrency is a new extension of digital or virtual currency designed to work as a medium of exchange. In the current digital exchanging world, investors and interested parties are eager to know more information about, and the capabilites of, this new type of currency. One of the potential paths to retrieve the info automatically and quickly is through a chatbot. We explored the open source python library, Chatterbot, to apply Itchat API (a WeChat interface) with the aim of building a robot chatting application, I&C Chat, on the topic of cryptocurrency. First, we collected question and answer pairs datasets from Quora websites. Furthermore, we also created API calls to query the real time quote for the top 25 cryptocurrencies. Then we used the collected data to train our chatbot and implemented a logic adapter to receive the price quote of cryptocurrencies based on the incoming question. The Itchat API method will return the best matched answer to the asking party automatically. The response time of different questions has been investigated. The results imply that this application is quite useful, feasible and beneficial to the digital currency world.

Historical texts form the basis of the study of antiquities. In the case of Chinese historical texts different genres exist, e.g. chronological and biographical works etc. The contents of these texts normally consist of complex and interrelated information which covers long time period. Traditional history research relies heavily on information extraction and analysis by human researchers. With the recent development of the internet, data science and visualization technologies, digital history gradually attracts more and more attentions and in turn significantly impacts the field of historical study through altering the accessibility of the source materials, the narrative strategy and the analytical methodologies. This paper provides a system that enhances the Chinese historical research using word segmentation, texts analysis and visualization technologies. We can improve the workflow of traditional historical research via automatically detecting important keywords in Chinese historical texts and extracting, analyzing and visualizing the relations between a keyword and other words. This does not only accelerate the text based historical study but also to a great extent increase the scope of the search and analysis of the keywords in Chinese historical texts which used to be limited by the capacity of human researchers.

Tracking moving objects is always a critical challenge in cyber-physical systems. Researchers have proposed many tracking algorithms. However, most of the proposed algorithms cannot be used for on-demand deployment because of the unavailable preset fingerprints (prior landmark or context information) in their assumption. Another issue is that those algorithms with models built in an interference-free environment cannot work in interference-rich environments. To address those issues, we propose a localizing and tracking algorithm called Enhanced Field Division (EFD), which dynamically divides the field into areas with unique signatures and tracks the target without any fingerprints. We also implemented a proof-of-concept localization platform to demonstrate the tracking accuracy and the algorithm performance in practical, interference-rich environments.

Location-based services (LBSs) are increasingly popular in today’s society. People reveal their location information to LBS providers to obtain personalized services such as map directions, restaurant recommendations, and taxi reservations. Usually, LBS providers offer user privacy protection statement to assure users that their private location information would not be given away. However, many LBSs run on third-party cloud infrastructures. It is challenging to guarantee user location privacy against curious cloud operators while still permitting users to query their own location information data. In this paper, we propose an efficient privacy-preserving cloud-based LBS query scheme for the multiuser setting. We encrypt LBS data and LBS queries with a hybrid encryption mechanism, which can efficiently implement privacy-preserving search over encrypted LBS data and is very suitable for the multiuser setting with secure and effective user enrollment and user revocation. This paper contains security analysis and performance experiments to demonstrate the privacy-preserving properties and efficiency of our proposed scheme.

Privacy and integrity have been the main road block to the applications of two-tiered sensor networks. The storage nodes, which act as a middle tier between the sensors and the sink, could be compromised and allow attackers to learn sensitive data and manipulate query results. Prior schemes on secure query processing are weak, because they reveal non-negligible information, and therefore, attackers can statistically estimate the data values using domain knowledge and the history of query results. In this paper, we propose the first top-k query processing scheme that protects the privacy of sensor data and the integrity of query results. To preserve privacy, we build an index for each sensor collected data item using pseudo-random hash function and Bloom filters and transform top-k queries into top-range queries. To preserve integrity, we propose a data partition algorithm to partition each data item into an interval and attach the partition information with the data. The attached information ensures that the sink can verify the integrity of query results. We formally prove that our scheme is secure under IND-CKA security model. Our experimental results on real-life data show that our approach is accurate and practical for large network sizes.

Malware detection becomes mission critical as its threats spread from personal computers to industrial control systems. Modern malware generally equips with sophisticated anti-detection mechanisms such as code-morphism, which allows the malware to evolve into many variants and bypass traditional code feature based detection systems. In this paper, we propose to disassemble binary executables into opcodes sequences, and then convert the opcodes into images. By comparing the opcode images generated from binary targets with the opcode images generated from known malware sample codes, we can detect if the target binary executables contain variants of these known malwares. Theoretical analysis and real-life experiments results show that malware detection using visualized analysis is comparable in terms of accuracy, our approach can significantly improve 15\% of detection accuracy when the detection set contains a large quantity of binaries and the training set is small.

With the advent of cloud computing, it has become increasingly popular for data owners to outsource their data to public cloud servers while allowing data users to retrieve this data. For privacy concerns, secure searches over encrypted cloud data has motivated several research works under the single owner model. However, most cloud servers in practice do not just serve one owner; instead, they support multiple owners to share the benefits brought by cloud computing. In this paper, we propose schemes to deal with privacy preserving ranked multi-keyword search in a multi-owner model (PRMSM). To enable cloud servers to perform secure search without knowing the actual data of both keywords and trapdoors, we systematically construct a novel secure search protocol. To rank the search results and preserve the privacy of relevance scores between keywords and files, we propose a novel additive order and privacy preserving function family. To prevent the attackers from eavesdropping secret keys and pretending to be legal data users submitting searches, we propose a novel dynamic secret key generation protocol and a new data user authentication protocol. Furthermore, PRMSM supports efficient data user revocation. Extensive experiments on real-world datasets confirm the efficacy and efficiency of PRMSM.

Localization proves to be a promising application of wireless sensor networks. Although a considerable number of algorithms have been designed for low-overhead and high-accuracy localization, problems remain to be tackled such as the way to use anchor-deploying. In this paper, we present a mechanism for range-free localization called Enhanced Map Segmentation (EMS) to deploy and segment the map where precise indoor localization is required. Despite the limits of environmental noise, sensing irregularity, received signal strength (RSS) variation and other unavoidable factors, EMS can be reliable by improving the quality of map segmentation. This paper will present and analyze the enhancing method by a series of simulations. In addition, to deal with ambiguous context positions that confounds the localization, this paper ameliorates the segmentation with context conception mentioned in [1] by statistical methods. In fact, a well-organized deployment and a context-based decision mechanism can make such a layer of abstraction more reliable and compatible.

Wireless sensor networks are often deployed for tracking moving objects. Many tracking algorithms have been proposed with two general assumptions: the preset fingerprints(prior landmark or context information) and an interference-free environment. These algorithms, however, cannot be used for on-demand deployment where finger-prints are unavailable and would perform poorly in interference-rich environments. In this paper, we present a fingerprint-free localizing and tracking algorithm, called Enhanced Field Division (EFD). The EFD algorithm is used to dynamically divide the field into areas with unique signatures and tracks the target, without any finger-prints. We also implemented a proof-of-concept localization platform to demonstrate the tracking accuracy and the algorithm performance in practical, interference rich environment.

In wireless sensor networks, owing to the limited energy of the sensor node, it is very meaningful to propose a dynamic scheduling scheme with data management that reduces energy as soon as possible. However, traditional techniques treat data management as an isolated process on only selected individual nodes. In this article, we propose an aggressive data reduction architecture, which is based on error control within sensor segments and integrates three parallel dynamic control mechanisms. We demonstrate that this architecture not only achieves energy savings but also guarantees the data accuracy specified by the application. Furthermore, based on this architecture, we propose two implementations. The experimental results show that both implementations can raise the energy savings while keeping the error at an predefined and acceptable level. We observed that, compared with the basic implementation, the enhancement implementation achieves a relatively higher data accuracy. Moreover, the enhancement implementation is more suitable for the harsh environmental monitoring applications. Further, when both implementations achieve the same accuracy, the enhancement implementation saves more energy. Extensive experiments on realistic historical soil temperature data confirm the efficacy and efficiency of two implementations.

With the advent of cloud computing, it becomes increasingly popular for data owners to outsource their data to public cloud servers while allowing data users to retrieve these data. For privacy concerns, secure searches over encrypted cloud data motivated several researches under the single owner model. However, most cloud servers in practice do not just serve one owner, instead, they support multiple owners to share the benefits brought by cloud servers. In this paper, we propose schemes to deal with secure ranked multi-keyword search in a multi-owner model. To enable cloud servers to perform secure search without knowing the actual data of both keywords and trapdoors, we systematically construct a novel secure search protocol. To rank the search results and preserve the privacy of relevance scores between keywords and files, we propose a novel Additive Order and Privacy Preserving Function family. Extensive experiments on real-world datasets confirm the efficacy and efficiency of our proposed schemes.

The main duty of key management is to keep cryptographic keys in secret. However, it is difficulty to quantitatively assess that how well does a key management scheme protect the keys. In this paper, we propose to use reliability theory, which was mainly used to evaluate performance persistence for engineering systems, to estimate the performance of key management schemes. The reliability analysis leads to counter-intuitive results such as the widely deployed periodic key update scheme is ineffective when key thefts are possible. The analysis also shows that using password with an electronic security token for authentication is a strong security measure in the beginning but is unreliable in the long run. In general, the reliability analysis demonstrates that current key management schemes focus too much on postponing the first key theft from occurring but lack of considerations on quickly recovering stolen keys. In the later part of this paper, we discuss possible directions that may improve the reliability of key management schemes.

Cloud computing provides abundant benefits including easy access, decreased costs and flexible resource management. For privacy concerns, sensitive data have to be encrypted before outsourcing, which obsoletes traditional data utilization based on plaintext keyword search. Therefore, developing a secure search service over encrypted cloud data is of paramount importance. There are several researches concerned about this problem. However, all these schemes are based on a single cloud model which has the threat of single point of failure, loss and corruption of data, loss of availability and loss of privacy. In this paper, we explore the problem of secure distributed keyword search in a multi-cloud paradigm. We first define a distributed search model. Based on this model, we propose two schemes. In scheme_I, we propose to cross-store all encrypted file slices, keywords and keys. In scheme_II, we systematically construct a keyword distributing strategy and a file distributing strategy. Further, we extend both schemes with Shamir's secret schemes to achieve better availability and robustness. Extensive experiments on real-world datasets confirm the efficacy and efficiency of our schemes.

This paper introduces a set of low-complexity algorithms that when coupled with link layer retransmission mechanisms, strengthen wireless communication security. Our basic idea is to generate a series of secrets from inevitable transmission errors and other random factors in wireless communications. Because these secrets are constantly extracted from the communication process in realtime, we call them dynamic secrets. Dynamic secrets have interesting security properties. They offer a complementary mechanism to existing security protocols. Even if the adversary exploits a vulnerability and steals the underlying system secret, security can be automatically replenished. In many scenarios, it is also possible to bootstrap a secure communication with the dynamic secrets.

Security becomes a prominent issue for mobile cloud computing since valuable information moves to the cloud. An important security problem is how the users identify themselves to the cloud. If an attacker is capable of faking or stealing user credentials, such as passwords and digital certificates, current security measures are not sufficient to safe-guard user's valuable information in the cloud. This paper proposes a set of light weight algorithms to generate dynamic credential to defend against such powerful attackers. Many communication randomness, like the user mobility, were commonly believed harmful for security. In this paper, they are converted to helpful elements to generate dynamic credentials. Dynamic credential introduces interesting security properties.

Recently there has been a great deal of interest in load-balancing switches due to their simple architecture and high bandwidth. In this paper we propose a three-stage load- balancing switch along with output load-balancing to address the mis-sequencing problem. We show that our proposed scheme provides a delay guarantee bounded by the delay of an OQ switch with the same input traffic plus a constant while achieving 100% throughput for admissible traffic with (sigma, rho) -upper constraint.

It is generally believed harmful to have transmission errors in the wireless communications. The high decoding complexity of dense parity check codes is unfavorable. This paper proposes to apply these two "negative" facts to enable the secrecy sharing with the information theoretical security. We claim that the secrecy sharing is always possible if the wiretap channel is not error-free, regardless of the main channel performance. Particularly, the proposed secrecy sharing protocol can provide provable and testable security using the existing wireless technologies.