Wang XU Yongliang MA Kehai CHEN Ming ZHOU Muyun YANG Tiejun ZHAO
Non-autoregressive generation has attracted more and more attention due to its fast decoding speed. Latent alignment objectives, such as CTC, are designed to capture the monotonic alignments between the predicted and output tokens, which have been used for machine translation and sentence summarization. However, our preliminary experiments revealed that CTC performs poorly on document abstractive summarization, where a high compression ratio between the input and output is involved. To address this issue, we conduct a theoretical analysis and propose Hierarchical Latent Alignment (HLA). The basic idea is a two-step alignment process: we first align the sentences in the input and output, and subsequently derive token-level alignment using CTC based on aligned sentences. We evaluate the effectiveness of our proposed approach on two widely used datasets XSUM and CNNDM. The results indicate that our proposed method exhibits remarkable scalability even when dealing with high compression ratios.
Xingyu WANG Ruilin ZHANG Hirofumi SHINOHARA
This paper introduces an inverter-based true random number generator (I-TRNG). It uses a single CMOS inverter to amplify thermal noise multiple times. An adaptive calibration mechanism based on clock tuning provides robust operation across a wide range of supply voltage 0.5∼1.1V and temperature -40∼140°C. An 8-bit Von-Neumann post-processing circuit (VN8W) is implemented for maximum raw entropy extraction. In a 130nm CMOS technology, the I-TRNG entropy source only occupies 635μm2 and consumes 0.016pJ/raw-bit at 0.6V. The I-TRNG occupies 13406μm2, including the entropy source, adaptive calibration circuit, and post-processing circuit. The minimum energy consumption of the I-TRNG is 1.38pJ/bit at 0.5V, while passing all NIST 800-22 and 800-90B tests. Moreover, an equivalent 15-year life at 0.7V, 25°C is confirmed by an accelerated NBTI aging test.
He TIAN Kaihong GUO Xueting GUAN Zheng WU
In order to improve the anomaly detection efficiency of network traffic, firstly, the model is established for network flows based on complex networks. Aiming at the uncertainty and fuzziness between network traffic characteristics and network states, the deviation extent is measured from the normal network state using deviation interval uniformly, and the intuitionistic fuzzy sets (IFSs) are established for the various characteristics on the network model that the membership degree, non-membership degree and hesitation margin of the IFSs are used to quantify the ownership of values to be tested and the corresponding network state. Then, the knowledge measure (KM) is introduced into the intuitionistic fuzzy weighted geometry (IFWGω) to weight the results of IFSs corresponding to the same network state with different characteristics together to detect network anomaly comprehensively. Finally, experiments are carried out on different network traffic datasets to analyze the evaluation indicators of network characteristics by our method, and compare with other existing anomaly detection methods. The experimental results demonstrate that the changes of various network characteristics are inconsistent under abnormal attack, and the accuracy of anomaly detection results obtained by our method is higher, verifying our method has a better detection performance.
Daiki HIRATA Norikazu TAKAHASHI
Convolutional Neural Networks (CNNs) have shown remarkable performance in image recognition tasks. In this letter, we propose a new CNN model called the EnsNet which is composed of one base CNN and multiple Fully Connected SubNetworks (FCSNs). In this model, the set of feature maps generated by the last convolutional layer in the base CNN is divided along channels into disjoint subsets, and these subsets are assigned to the FCSNs. Each of the FCSNs is trained independent of others so that it can predict the class label of each feature map in the subset assigned to it. The output of the overall model is determined by majority vote of the base CNN and the FCSNs. Experimental results using the MNIST, Fashion-MNIST and CIFAR-10 datasets show that the proposed approach further improves the performance of CNNs. In particular, an EnsNet achieves a state-of-the-art error rate of 0.16% on MNIST.
Jion HIROSE Junya NAKAMURA Fukuhito OOSHITA Michiko INOUE
We study the gathering problem requiring a team of mobile agents to gather at a single node in arbitrary networks. The team consists of k agents with unique identifiers (IDs), and f of them are weakly Byzantine agents, which behave arbitrarily except falsifying their identifiers. The agents move in synchronous rounds and cannot leave any information on nodes. If the number of nodes n is given to agents, the existing fastest algorithm tolerates any number of weakly Byzantine agents and achieves gathering with simultaneous termination in O(n4·|Λgood|·X(n)) rounds, where |Λgood| is the length of the maximum ID of non-Byzantine agents and X(n) is the number of rounds required to explore any network composed of n nodes. In this paper, we ask the question of whether we can reduce the time complexity if we have a strong team, i.e., a team with a few Byzantine agents, because not so many agents are subject to faults in practice. We give a positive answer to this question by proposing two algorithms in the case where at least 4f2+9f+4 agents exist. Both the algorithms assume that the upper bound N of n is given to agents. The first algorithm achieves gathering with non-simultaneous termination in O((f+|&Lambdagood|)·X(N)) rounds. The second algorithm achieves gathering with simultaneous termination in O((f+|&Lambdaall|)·X(N)) rounds, where |&Lambdaall| is the length of the maximum ID of all agents. The second algorithm significantly reduces the time complexity compared to the existing one if n is given to agents and |&Lambdaall|=O(|&Lambdagood|) holds.
Lige GE Shengming JIANG Xiaowei WANG Yanli XU Ruoyu FENG Zhichao ZHENG
Along with the fast development of blue economy, wireless communication in oceans has received extensive attention in recent years, and opportunistic networks without any aid from fixed infrastructure or centralized management are expected to play an important role in such highly dynamic environments. Here, link prediction can help nodes to select proper links for data forwarding to reduce transmission failure. The existing prediction schemes are mainly based on analytical models with no adaptability, and consider relatively simple and small terrestrial wireless networks. In this paper, we propose a new link prediction algorithm based on machine learning, which is composed of an extractor of convolutional layers and an estimator of long short-term memory to extract useful representations of time-series data and identify effective long-term dependencies. The experiments manifest that the proposed scheme is more effective and flexible compared with the other link prediction schemes.
Tsutomu SASAO Yuto HORIKAWA Yukihiro IGUCHI
A classification function maps a set of vectors into several classes. A machine learning problem is treated as a design problem for partially defined classification functions. To realize classification functions for MNIST hand written digits, three different architectures are considered: Single-unit realization, 45-unit realization, and 45-unit ×r realization. The 45-unit realization consists of 45 ternary classifiers, 10 counters, and a max selector. Test accuracy of these architectures are compared using MNIST data set.
Hiroaki YAMAMOTO Hiroshi FUJIWARA
This paper presents a new method to translate a regular expression into a nondeterministic finite automaton (an NFA for short). Let r be a regular expression and let M be a Thompson automaton for r. We first introduce a labeled Thompson automaton defined by assigning two types of expressions which denote prefixes and suffixes of words in L(r) to each state of M. Then we give new ϵ-free NFAs constructed from a labeled Thompson automaton. These NFAs are called a prefix equation automaton and a suffix equation automaton. We show that a suffix equation automaton is isomorphic to an equation automaton defined by Antimirov. Furthermore we give an NFA called a unified equation automaton by joining two NFAs. Thus the number of states of a unified equation automaton can be smaller than that of an equation automaton.
Kohei SHIMATANI Shigemasa TAKAI
We consider the bisimilarity control problem for partially observed nondeterministic discrete event systems with deterministic specifications. This problem requires us to synthesize a supervisor that achieves bisimulation equivalence of the supervised system and the deterministic specification under partial observation. We present necessary and sufficient conditions for the existence of such a deterministic supervisor and show that these conditions can be verified polynomially.
Tsunehiro YOSHINAGA Makoto SAKAMOTO
This paper investigates the closure properties of multi-inkdot nondeterministic Turing machines with sublogarithmic space. We show that the class of sets accepted by the Turing machines is not closed under concatenation with regular set, Kleene closure, length-preserving homomorphism, and intersection.
Taku YAMAZAKI Ryo YAMAMOTO Genki HOSOKAWA Tadahide KUNITACHI Yoshiaki TANAKA
In wireless multi-hop networks such as ad hoc networks and sensor networks, backoff-based opportunistic routing protocols, which make a forwarding decision based on backoff time, have been proposed. In the protocols, each potential forwarder calculates the backoff time based on the product of a weight and global scaling factor. The weight prioritizes potential forwarders and is calculated based on hop counts to the destination of a sender and receiver. The global scaling factor is a predetermined value to map the weight to the actual backoff time. However, there are three common issues derived from the global scaling factor. First, it is necessary to share the predetermined global scaling factor with a centralized manner among all terminals properly for the backoff time calculation. Second, it is almost impossible to change the global scaling factor during the networks are being used. Third, it is difficult to set the global scaling factor to an appropriate value since the value differs among each local surrounding of forwarders. To address the aforementioned issues, this paper proposes a novel decentralized local scaling factor control without relying on a predetermined global scaling factor. The proposed method consists of the following three mechanisms: (1) sender-centric local scaling factor setting mechanism in a decentralized manner instead of the global scaling factor, (2) adaptive scaling factor control mechanism which adapts the local scaling factor to each local surrounding of forwarders, and (3) mitigation mechanism for excessive local scaling factor increases for the local scaling factor convergence. Finally, this paper evaluates the backoff-based opportunistic routing protocol with and without the proposed method using computer simulations.
Tomoyuki SASAKI Hidehiro NAKANO
Particle swarm optimization (PSO) is a swarm intelligence algorithm and has good search performance and simplicity in implementation. Because of its properties, PSO has been applied to various optimization problems. However, the search performance of the classical PSO (CPSO) depends on reference frame of solution spaces for each objective function. CPSO is an invariant algorithm through translation and scale changes to reference frame of solution spaces but is a rotationally variant algorithm. As such, the search performance of CPSO is worse in solving rotated problems than in solving non-rotated problems. In the reference frame invariance, the search performance of an optimization algorithm is independent on rotation, translation, or scale changes to reference frame of solution spaces, which is a property of preferred optimization algorithms. In our previous study, piecewise-linear particle swarm optimizer (PPSO) has been proposed, which is effective in solving rotated problems. Because PPSO particles can move in solution spaces freely without depending on the coordinate systems, PPSO algorithm may have rotational invariance. However, theoretical analysis of reference frame invariance of PPSO has not been done. In addition, although behavior of each particle depends on PPSO parameters, good parameter conditions in solving various optimization problems have not been sufficiently clarified. In this paper, we analyze the reference frame invariance of PPSO theoretically, and investigated whether or not PPSO is invariant under reference frame alteration. We clarify that control parameters of PPSO which affect movement of each particle and performance of PPSO through numerical simulations.
Xijian ZHONG Yan GUO Ning LI Shanling LI Aihong LU
In the large-scale multi-UAV systems, the direct link may be invalid for two remote nodes on account of the constrained power or complex communication environment. Idle UAVs may work as relays between the sources and destinations to enhance communication quality. In this letter, we investigate the opportunistic relay selection for the UAVs dynamic network. On account of the time-varying channel states and the variable numbers of sources and relays, relay selection becomes much more difficult. In addition, information exchange among all nodes may bring much cost and it is difficult to implement in practice. Thus, we propose a decentralized relay selection approach based on mood-driven mechanism to combat the dynamic characteristics, aiming to maximize the total capacity of the network without information exchange. With the proposed approach, the sources can make decisions only according to their own current states and update states according to immediate rewards. Numerical results show that the proposed approach has attractive properties.
We consider a similarity control problem for discrete event systems modeled as nondeterministic automata. A nonblocking supervisor was synthesized in the previous work under the assumption that the event occurrence and the current state of the plant are observable. In this letter, we prove that the synthesized supervisor is a maximally permissive nonblocking one.
Peerasak INTARAPAIBOON Thanaruk THEERAMUNKONG
Multi-slot information extraction, also known as frame extraction, is a task that identify several related entities simultaneously. Most researches on this task are concerned with applying IE patterns (rules) to extract related entities from unstructured documents. An important obstacle for the success in this task is unknowing where text portions containing interested information are. This problem is more complicated when involving languages with sentence boundary ambiguity, e.g. the Thai language. Applying IE rules to all reasonable text portions can degrade the effect of this obstacle, but it raises another problem that is incorrect (unwanted) extractions. This paper aims to present a method for removing these incorrect extractions. In the method, extractions are represented as intuitionistic fuzzy sets, and a similarity measure for IFSs is used to calculate distance between IFS of an unclassified extraction and that of each already-classified extraction. The concept of k nearest neighbor is adopted to design whether the unclassified extraction is correct or not. From the experiment on various domains, the proposed technique improves extraction precision while satisfactorily preserving recall.
The Discrete Fourier Transform Test (DFTT) is a randomness test in NIST SP800-22. However, to date, the theoretical reference distribution of the DFTT statistic has not been derived, which is problematic. We propose a new test using power spectrum variance as the test statistic whose reference distribution can be derived theoretically. Note that the purpose of both the DFTT and the proposed test is to detect periodic features. Experimental results demonstrate that the proposed test has stronger detection power than the DFTT and that it test can be used even for short sequences.
This letter presents two outage-optimal relaying schemes to improve the performance of a wireless energy harvesting system in cognitive radio networks. The performance of both schemes is then evaluated and compared by carrying out numerical simulations, and we also derive the analytic expression for the outage probability of the secondary system.
In this paper, we consider a similarity control problem for nondeterministic discrete event systems, which requires us to synthesize a nonblocking supervisor such that the supervised plant is simulated by a given specification. We assume that a supervisor can observe not only the event occurrence but also the current state of the plant. We present a necessary and sufficient condition for the existence of a nonblocking supervisor that solves the similarity control problem and show how to verify it in polynomial time. Moreover, when the existence condition of a nonblocking supervisor is satisfied, we synthesize such a supervisor as a solution to the similarity control problem.
Deterministic ID-based signatures are digital signatures where secret keys are probabilistically generated by a key generation center while the signatures are generated deterministically. Although the deterministic ID-based signatures are useful for both systematic and cryptographic applications, to the best of our knowledge, there is no scheme with a tight reduction proof. Loosely speaking, since the security is downgraded through dependence on the number of queries by an adversary, a tighter reduction for the security of a scheme is desirable, and this reduction must be as close to the difficulty of its underlying hard problem as possible. In this work, we discuss mathematical features for a tight reduction of deterministic ID-based signatures, and show that the scheme by Selvi et al. (IWSEC 2011) is tightly secure by our new proof framework under a selective security model where a target identity is designated in advance. Our proof technique is versatile, and hence a reduction cost becomes tighter than the original proof even under an adaptive security model. We furthermore improve the scheme by Herranz (The Comp. Jour., 2006) to prove tight security in the same manner as described above. We furthermore construct an aggregate signature scheme with partial aggregation, which is a key application of deterministic ID-based signatures, from the improved scheme.
Osama SALAMEH Koen DE TURCK Dieter FIEMS Herwig BRUNEEL Sabine WITTEVRONGEL
In Cognitive Radio Networks (CRNs), spectrum sensing is performed by secondary (unlicensed) users to utilize transmission opportunities, so-called white spaces or spectrum holes, in the primary (licensed) frequency bands. Secondary users (SUs) perform sensing upon arrival to find an idle channel for transmission as well as during transmission to avoid interfering with primary users (PUs). In practice, spectrum sensing is not perfect and sensing errors including false alarms and misdetections are inevitable. In this paper, we develop a continuous-time Markov chain model to study the effect of false alarms and misdetections of SUs on several performance measures including the collision rate between PUs and SUs, the throughput of SUs and the SU delay in a CRN. Numerical results indicate that sensing errors can have a high impact on the performance measures.