Multivariate public-key cryptography (MPKC) is considered as one of the main candidates for post-quantum cryptography (PQC). In MPKC, the MinRank attacks, which try to solve the MinRank problem obtained from a public key, are important since a lot of multivariate schemes are broken by these attacks. Among them, the rectangular MinRank attack was recently proposed for the Rainbow scheme by Beullens, and it tries to solve a new kind of MinRank problem obtained by transforming the public key of Rainbow. Due to this attack, it is known that the security level of Rainbow was reduced. Rainbow is a multi-layered variant of the UOV scheme, and UOV is considered having a resistance to all MinRank attacks since its public key consists of full rank matrices. Recently, there have been submitted three new variants of the UOV scheme having a small public key, MAYO, QR-UOV and VOX in the NIST PQC standardization of additional digital signature schemes. In this paper, we show that the rectangular MinRank attack is applicable to MAYO, QR-UOV and VOX. Moreover, we estimate the complexity of the attack. In particular, we report that all the parameter sets of VOX submitted to NIST PQC standardization are broken in at most 2⁵⁵ gate operations.

In 2016, message franking was introduced by Facebook in end-to-end encrypted messaging. This feature enables recipients to report harmful content to their service provider in a verifiable manner. Grubbs et al. (CRYPTO 2017) formalized compactly committing authenticated encryption with associated data (ccAEAD) as a symmetric-key primitive that can be used for message franking and presented its generic constructions. Dodis et al. (CRYPTO 2018) proposed encryptment as a core component of ccAEAD and presented two transforms to build ccAEAD from encryptment. One transform builds randomized ccAEAD with one call to conventional AEAD, while the other builds nonce-based ccAEAD with two calls to a pseudorandom function (PRF). Hirose and Minematsu presented an improved transform that requires a tweakable block cipher instead of AEAD. This paper presents an even simplified transform to build randomized ccAEAD, which requires only one call to a PRF. The resulting ccAEAD is more efficient regarding bandwidth than Dodis et al. and has a smaller computation cost than Hirose and Minematsu. The presented transform can be extended to build nonce-based ccAEAD, which is also more efficient than the one presented by Dodis et al. regarding bandwidth, though it requires two calls to a PRF as well as their transform.

Physical attacks against cryptographic hardware have become a major threat. For example, side-channel attacks (SCAs) exploit information leakage from power consumption and electromagnetic radiation during encryption to recover secret keys. We recognize them as a powerful threat because the attackers can conduct them using relatively inexpensive equipment. Thus, embedded systems based on cryptographic hardware need to be secure against SCAs. Threshold Implementation (TI) is widely studied as an effective countermeasure against SCAs. Each sensitive intermediate value is divided into multiple values called shares using random bits, and each share is performed to realize the cryptographic algorithm securely. TI requires three important properties for secure computation: correctness, non-completeness, and uniformity. Note that non-linear operation, e.g., AES S-box, cannot preserve perfect uniformity. Compensating for the lack of uniformity, the intermediate values must be re-masked using a large amount of fresh random numbers, called refreshing. Therefore, it is necessary to use random numbers in random number generators (RNGs) to implement TI, but the security requirements for randomness in such RNGs are not yet well-discussed. In this paper, we investigate the impact of practical randomness on security against SCAs. More specifically, we implement AES hardware protected by second-order TI on an FPGA to evaluate the security in cases where the random number used for dividing the secret value into shares is fixed or random. Furthermore, we also explore information leakage in the case where randomized or fixed seed values are sent to the RNG used in refreshing or where the frequency of random number updates is reduced. Based on these results, we discuss practical randomness suitable for TI-based hardware countermeasures.

We propose a new covert-channel attack that exploits inaudible acoustic leakage from multilayer ceramic capacitors (MLCCs) using a laser Doppler vibrometer (LDV). Malware installed on a victim PC modulates the CPU load by transmitting data bits that induce acoustic noise from an MLCC on the victim PC’s motherboard. Unlike conventional attacks that use a microphone to capture such acoustic leakage, we use an LDV aimed at the MLCC to capture the acoustic leakage from the MLCC. Using LDV, instead of microphones, the attacker can exploit inaudible high-frequency signals and penetrate transparent obstacles such as a glass side panel on the victim’s PC by shining a laser on the target MLCC. The proposed method requires less privilege compared to conventional covert acoustic channel attacks that require privilege to use IO devices (e.g., loudspeaker, microphone). In addition, the proposed method exploits the acoustic leakage from MLCCs instead of a loudspeaker. Therefore, the proposed method is possible to attack PCs that do not have loudspeaker installed. Compared with conventional LDV-based eavesdropping attacks, the proposed method extends them to MLCC leakage in the covert-channel setting. We experimentally verify the proposed attack by measuring inaudible acoustic leakage from MLCC, induced by modulated CPU load, by using an LDV and evaluating the bitrate.

Side-channel attacks (SCAs) using deep learning techniques have been mainly reported as profiled attacks, but in TCHES 2019, differential deep learning analysis (DDLA) was proposed by Timon as a non-profiled attack. In this attack method, deep learning models for all candidate keys are trained, and the key corresponding to the most suitable learning metrics such as loss and accuracy is assumed to be the correct key. Timon focused on a single bit (least significant bit (LSB) or most significant bit (MSB)) of the intermediate value during the operation of the cryptographic circuit and successfully revealed the correct key against the software implementation of the advanced encryption standard (AES). However, when we applied this method to our hardware implementation, we could not reveal all partial keys due to the existence of registers whose Hamming distance (HD) leakage is difficult to observe. In this paper, we propose a multi-bit DDLA that focuses on all bits to solve this problem. When a DDLA was performed on the hardware implemented AES without SCA countermeasures, the HD-ID labels, which had been used as a conventional profiled type DL-SCA method focusing on 8 bits, cannot reveal the 0, 4, 8, and 12th byte keys, but the proposed multi-bit method succeeds in revealing all key bytes. On the other hand, compared to correlated power analysis (CPA) which is a typical non-profiled attack that does not use deep learning, the number of waveforms required to reveal all keys is 1.6 times higher, so the DDLA with our proposed method is not so useful to the target without SCA countermeasures. Thus, we also evaluated the proposed method against FPGA-implemented RSM-AES and WDDL-AES, which have some resistance to SCA, and successfully revealed all keys against RSM and WDDL with 100,000 and 50,000 waveforms, respectively. This is a significant improvement over conventional CPA, which reveals less than half of the key bytes despite using twice as many waveforms as the proposed method. These results suggest that multi-bit DDLA is effective on non-profiled attacks against hardware-implemented AES circuit with SCA countermeasures.

Differential deep learning analysis (DDLA) was proposed as a side-channel attack (SCA) with deep learning techniques in non-profiled scenarios at TCHES 2019. In the proposed DDLA, the adversary sets the LSB or MSB of the intermediate value in the encryption process assumed for the key candidates as the ground-truth label and trains a deep neural network (DNN) with power traces as an input. The adversary also observes metrics such as loss and accuracy during DNN training and estimates that the key corresponding to the best-fitting DNN is correct. One of the disadvantages of DDLA is the heavy computation time for the DNN models because the number of required models is the as same as the number of key candidates, which is 256 in the case of AES. Therefore 4096 DNNs are required for revealing keys of 16 bytes. Furthermore, the DNN models have to be trained again if the adversary changes a ground-truth label function from LSB to other labels such as MSB or HW. We propose a new deep-learning-based SCA in a non-profiled scenario to solve these problems. Our core idea is to extract feature of the leakage waveform using DNN. The adversary reveals the correct keys by conducting cluster analysis using the feature vectors extracted from power traces using DNN. We named this method as CA-SCA (cluster-analysis-based side-channel attacks), it is advantageous that only one DNN needs to be trained to reveal all key bytes. In addition, once the DNN is trained, multiple label functions can be tested without the additional cost of training DNNs. We provide four case studies of attacking against AES, including two software implementations and two hardware implementations. Our attacks against software implementations provide methods using a concatenated dataset that efficiently train the DNN. Also, our attack on the hardware implementation introduces multitask learning to exploit the Hamming distance leakage model. The results show that the proposed method requires fewer waveforms to reveal all key bytes than DDLA owing to the efficient learning performance on the above methods. Comparing the computation time to process the same number of waveforms, the proposed method requires only about 1/75 and 1/25 of the time when attacking software and hardware implementations, respectively, due to the significant reduction in the number of training models.

The Adversarial Tactics, Techniques, and Common Knowledge (ATT&CK) and Common Attack Pattern Enumeration and Classification (CAPEC) frameworks are essential knowledge bases that catalog traditional attack patterns and their interrelationships (e.g., abstract-concrete relationships). In addition, a knowledge base named Adversarial Threat Landscape for Artificial-Intelligence Systems (ATLAS) focuses on artificial intelligence (AI)/machine learning (ML)-related attack patterns. Newly discovered attack patterns are incorporated into these knowledge bases manually, potentially leading to missed relationships or delayed information updates. This study introduces a methodology that uses large language models (LLMs) to identify abstract-concrete relationships between attack patterns, aiding in rapid classification and in the rapid development of a defensive strategy. We trained BERT, GPT, and SVM models on ATT&CK, CAPEC, and their combined datasets for relation classification among attack patterns. The evaluation results show that the fine-tuned GPT-3.5 model outperformed the other investigated models, showing potential applicability even to AI/ML-related attack patterns and emphasizing the importance of using training data in the same format as test data. This study also finds that GPT-3.5 effectively focuses on critical descriptive terms, bolstering its performance. The proposed methodology is effective in discerning attack-pattern relationships, demonstrating its potential applicability in the AI security domain.

Backdoor attacks on machine learning are a kind of attack whereby an adversary obtains the expected output for a particular input called a trigger, and the existing work, called latent backdoor attack (Yao et al., CCS 2019), can resist backdoor removal as countermeasures to the attacks, i.e., pruning and transfer learning. In this paper, we present a novel backdoor attack, TALPA, which outperforms the latent backdoor attack with respect to the attack success rate of backdoors as well as keeping the same-level accuracy. The key idea of TALPA is to directly overrides parameters of latent representations in competitive learning between a generative model for triggers and a victim model, and hence can more optimize model parameters and trigger generation than the latent backdoor attack. We experimentally demonstrate that TALPA outperforms the latent backdoor attack with respect to the attack success rate and also show that TALPA can resist both pruning and transfer learning through extensive experiments. We also show various discussions, such as the impact of hyperparameters and extensions to other layers from the latent representation, to shed light on the properties of TALPA. Our code is publicly available (https://github.com/fseclab-osaka/talpa).

Federated learning (FL) is a distributed deep learning technique involving multiple clients and a server. In FL, each client individually trains a model with its own training data and sends only the model to the server. The server then aggregates the received client models to build a server model. Because each client does not share its own training data with other clients or the server, FL is considered a distributed deep learning technique with privacy protection. However, several attacks that steal information about a specific client’s training data from the aggregated model on the server have been reported for FL. These include membership inference attacks (MIAs), which identify whether or not specific data was used to train a target model. MIAs have been shown to work mainly because of overfitting of the model to the training data, and mitigation techniques based on knowledge distillation have thus been proposed. Because these techniques assume a lot of training data and computational power, they are difficult to introduce simply to clients in FL. In this paper, we propose a knowledge-distillation-based defense against MIAs that is designed for application in FL. The proposed method is effective against various MIAs without requiring additional training data, in contrast to the conventional defenses.

To realize online biometric authentication systems with both of protection and utilization of biometric data, we propose a novel primitive called “Sample Recoverable Fuzzy Extractors (SRFEs).” Conventionally, Biometric Template Protection (BTP) is studied as an approach for preventing biometric data from leakage. An important requirement of BTP is that it is difficult to recover biometric data from the stored data, which is called irreversibility, and fuzzy extractors are known as one of promising primitives for realizing BTP. On the other hand, in some cases, it is desired that the system can utilize biometric samples such as images having captured during past enrollment and authentication processes. For example, when the authentication accuracy of a specific user is low, samples of past processes are helpful clue for investigation of a cause. Also, they can be used for multi-sample fusion to improve accuracy in a biometric template update, and for post verification of past processes. To enable utilization of past biometric samples for such various situations while protecting the biometric data, we define a SRFE as a primitive satisfying the following two properties: (i) It can recover the secret key along with samples of past enrollment and authentication processes from the stored data and a feature which can succeed in the authentication. (ii) It is computationally difficult to recover the secret key from the stored data. We give a construction based on a fuzzy extractor and a symmetric encryption scheme satisfying a kind of key dependent message security. By using a SRFE, we realize a protocol of an online biometric authentication system which satisfies irreversibility while the past biometric samples can be recovered from the stored data with a help of the genuine user.

Although social networking services (SNS) have enabled the free exchange of opinions and feelings, the posting of malicious content has increasingly become a problem. To solve this problem, malicious behavior detection methods based on posting behavior are being developed. Existing methods focus on semantic analysis of posts using natural language processing, and one existing approach uses graph neural networks to consider context from various elements, such as users, posts, hashtags, and entities. However, this approach does not adequately capture the complex patterns and interactions of SNS networks. In particular, it is insufficient to fully capture the complexity of heterogeneity between nodes and edges in an SNS network. In this paper, we propose a method for extended heterogeneous graph construction and an architecture for heterogeneous graph embedding learning. The proposed method focuses on and exploits the diverse heterogeneity of social networks, optimally integrates heterogeneous information from SNS posts, and analyzes the relationships in the data to improve the performance of malicious behavior detection. The effectiveness of the proposed method is demonstrated by evaluation on a newly collected large dataset.

This paper introduces novel privacy-preserving deep unrolling techniques for recovering sparse signals, integrating privacy-preserving methodologies grounded in random unitary transformation. This approach facilitates data analysis and signal processing while safeguarding privacy. Focusing on sparse signal recovery, we concentrate on LASSO solutions known as LISTA and TISTA. These LISTA and TISTA methods, based on deep unrolling, have been devised to achieve notably faster convergence compared to ISTA. Our contribution lies in proposing secure LISTA and secure TISTA algorithms that operate on encrypted observation signals. The efficacy of the proposed approach was validated through simulations using artificially generated data for sparse signal recovery. As an illustration of the proposed methodology’s utility, we applied secure LISTA and secure TISTA to image reconstruction, to evaluate their performance.

Gaussian Process (GP) has been acknowledged as a powerful kernel-based machine learning technique with broad application areas, such as time series prediction and system state estimation. However, in the era of big data, new challenges are raised for GP. For example, in the presence of huge amount of data distributed at different locations, how to perform GP without being faced with significant privacy concerns? In this paper, we are aiming at constructing a distributed and secured GP learning framework over networks. Specifically, we first propose the idea of secured GP by incorporating random unitary transform, such that locally, the processing of data is guaranteed to be secure. Then, noticing that gathering data to a central node for GP learning is neither efficient nor secure, we extend secured GP into distributed learning over networks through invoking Alternating Direction Method of Multipliers (ADMM) technique, such that global optimality can be asymptotically reached only with local computations and parameter exchange. Finally, we demonstrate the performance improvements through simulation.

In this paper, we propose the frequency-domain weighted FxLMS algorithm for feedback active noise control. This algorithm aims to resolve the slow convergence issue of the conventional FxLMS algorithms by integrating a frequency-weighted method. This method dynamically adjusts weights based on the amplitude-frequency characteristics of narrowband noise, thereby improving tracking performance for time-varying narrowband noise. Through simulation experiments, we reveal that the FD-WFxNLMS algorithm achieves fast convergence, outperforming the conventional algorithms in feedback ANC systems.

Action quality assessment (AQA) has gained prominence as it finds widespread applications in various scenarios. Most existing methods directly regress from single or pairwise videos, which leads to redundant temporal features and limited views affecting the scoring mechanism. Moreover, direct regression only applies supervision to the last layer, which leads to hardship in optimizing the intermediate layers such as gradient vanishing. To end this, we propose a Hierarchical Joint Training based Replay-Guided Contrastive Transformer, learned by a temporal concentration module. For network architecture, we design an extra contrastive module for the input and its replay, and the consistency of scores guides the model to learn the features of the same action under different views. A temporal concentration module is proposed to extract concentrated features such as errors or highlights, which are crucial factors affecting scoring. The proposed hierarchical joint training provides supervision on both shallow and deep layers, enhancing the performance of the scoring mechanism and speed of training convergence. Extensive experiments demonstrate that our method achieves Spearman’s Rank Correlation of 0.9642 on the RFSJ dataset, which is the new state-of-the-art result.

In this paper, we propose a method which converts an original image into the image that does not give a strange impression to trichromatic vision and is easy for dichromatic vision to discriminate colors by modifying only the lightness in the RGB color space, where the color gamut can be easily grasped. In the proposed method, the lightness modification is executed by adding the red-green component multiplied by a coefficient derived from an optimization problem into the lightness component for each pixel. The optimization problem is defined as the minimization of the lightness difference between pixels considering the difference in color. The effectiveness of the proposed method is illustrated through comparison with conventional methods.

Calcification regions, which may be observed on dental panoramic radiographs, are a sign of vascular disease. Therefore, automatic detection methods based on semantic segmentation (SS) have been proposed. However, because of the small amount of data in the available dataset, the segmentation accuracy was insufficient. This paper proposes a method that uses adversarial features (AFs) for this problem. We extend AFs, which are an adversarial training method for discriminative problems, to SS. The proposed method can improve performance, even with a small amount of data.

The aesthetic evaluation of Chinese calligraphy, an art form with deep cultural roots and subjective interpretations, poses significant challenges in artificial intelligence. In this paper, we extend the methodology introduced in previous work using TabNet, a deep learning approach, to enhance the accuracy and interpretability of assessing the aesthetic qualities of Chinese calligraphy. Our study incorporates an expanded feature set: we add 10 new characteristics to the previously established 22 global shape features. This comprehensive feature ensemble captures the subtleties of Chinese calligraphy in accordance with its traditional artistic standards. Using TabNet, well known for its interpretability within deep learning frameworks, we aim to predict aesthetic scores with increased precision. We performed a rigorous evaluation using the Chinese Handwriting Aesthetic Evaluation Database. Our approach improved accuracy and elucidated the underlying reasoning behind the model’s predictions, thereby enhancing transparency.

This paper serves as an introductory overview of quantum deletion error-correction codes, a burgeoning field within quantum coding theory. Covering foundational concepts, existing research, and open questions, it aims to be the first accessible resource on the subject. This paper contains basic definitions of terms so that readers can read it regardless of their background. This paper invites readers to explore this primer and take their initial steps into the realm of quantum deletion error-correcting codes research.

Racetrack memory is a new type of high-capacity memory that stores data in magnetic nanowires called racetracks. Data is transferred through the nanowires to the access port for reading and writing. However, the data transfer process is imperfect and can lead to errors. Inspired by racetrack memory array architecture, the authors propose a new channel model in which missing data is filled with erasures at the end of the racetrack. Channel capacity and symmetric information rate for the proposed channel, a double-stack erasure-filled (DSEF) channel, are derived. Since the DSEF channel is a quaternary-input septenary-output channel, constructing good error-correcting codes is not trivial. We decompose the DSEF channel into two binary-input ternary-output channels to overcome this difficulty. This decomposition allowed us to construct an adequate error-correction scheme using existing binary codes, which is a meaningful achievement in terms of implementation.

An alphabetic code is a source code that preserves the lexicographical order between sequences in the encoding process. This paper studies k-bit delay alphabetic code-tuples, which are alphabetic codes allowing multiple code tables and at most k-bit decoding delay. As the main results, we show theorems to limit the scope of codes to be considered when discussing k-bit delay alphabetic code-tuples with the optimal average codeword length in theoretical analysis and practical code construction. These theorems imply the existence of an optimal k-bit delay alphabetic code whose code tables are all injective. They also give an upper bound of the necessary number of code tables for an alphabetic code to be optimal. In addition to the results above for a general integer k ≥ 0, we prove further results for particular cases k = 1, 2.

H-mutual information (H-MI) is a wide class of information leakage measures, where H = (η, F) is a pair of monotonically increasing function η and a concave function F, which is a generalization of Shannon entropy. H-MI is defined as the difference between the generalized entropy H and its conditional version, including Shannon mutual information (MI), Arimoto MI of order α, g-leakage, and expected value of sample information. This study presents a variational characterization of H-MI via statistical decision theory. Based on the characterization, we propose an alternating optimization algorithm for computing H-capacity.

Realization of large quantum computers is believed to jeopardize the security of cryptosystems relying on computational complexity of some mathematical problems, such as prime factorization and discrete logarithm problem. In this light, post-quantum cryptography, which is secure even after large quantum computers are realized, has been getting a lot of attention. National Institute of Standards and Technology (NIST) recently started a standardization process for post-quantum cryptosystems. The McEliece public-key cryptosystem based on quasi-cyclic moderate-density parity-check (QC-MDPC) codes is a promising candidate in this NIST standardization. Recently, attacks on the QC-MDPC McEliece scheme have extensively been investigated. The one proposed by Guo et al. exploits statistical information of decoding errors to reconstruct the secret key. This attack is twofold: (1) obtaining the distance spectrum of the secret key from statistical information of decoding errors, and (2) reconstructing the secret key from the distance spectrum. The bit-flipping decoding, which is commonly used to decode the QC-MDPC scheme, is considered to be vulnerable to the first part of this attack. Meanwhile the second part of the attack in the original version by Guo et al. requires considerable time because they use recursive search in this part. In this paper, we propose another method to reconstruct the secret key from the obtained distance spectrum on the basis of a method proposed by Fabšič et al. They found that the key construction can be mapped to a clique problem in graph theory. Using their observation, we apply a breadth-first search algorithm to the key reconstruction. Numerical experiments show that our method reconstructs the secret key more efficiently than recursive search in the original key reconstruction proposed by Guo et al.

The widespread use of IoT devices is expected to enable the collection and utilization of a variety of data, including personal health information. For example, we could provide our personal information for machine learning operated by an external server, which in return detects signs of illness. However, it is necessary to protect privacy of personal information. Precisely, there are two issues in privacy preserving machine learning. One is data privacy which means to protect our privacy to external servers. The other is model privacy which means to protect our privacy from models. Local differential privacy (LDP) mechanisms have been proposed as a method to provide personal sensitive information to external servers with privacy protection. LDP mechanisms can ensure privacy by adding noise to data, but on the other hand, adding noise reduces their usefulness for analysis. In this paper, we propose a privacy-preserving machine learning framework, which can deal with both data privacy and model privacy. We also propose a LDP-mechanism framework which can deal with various attributes included in a single data. We also make sure feasibility of our mechanism in two cases of breast cancer screening data and ionosphere data set.

A continuous-time average consensus system is a linear dynamical system defined over a graph, where each node has its own state value that evolves according to a simultaneous linear differential equation. A node is allowed to interact with neighboring nodes. Average consensus is a phenomenon that the all the state values converge to the average of the initial state values. In this paper, we assume that a node can communicate with neighboring nodes through an additive white Gaussian noise channel. We first formulate the noisy average consensus system by using a stochastic differential equation (SDE), which allows us to use the Euler-Maruyama method, a numerical technique for solving SDEs. By studying the stochastic behavior of the residual error of the Euler-Maruyama method, we arrive at the covariance evolution equation. The analysis of the residual error leads to a compact formula for mean squared error (MSE), which shows that the sum of the inverse eigenvalues of the Laplacian matrix is the most dominant factor influencing the MSE.

In the realm of web-based surveys, ensuring the quality of responses is a crucial yet challenging task. This study addresses the issue of detecting poor responses, particularly focusing on the phenomenon of ‘satisficing’ - a situation where respondents provide minimal effort responses. Traditional methods such as the Instructional Manipulation Check (IMC) and Directed Question Scale (DQS) have been commonly employed to tackle this issue. However, their effectiveness is often limited due to various constraints. This paper introduces a theoretical framework and a generalized model for the design of questionnaires. This framework aims to improve the detection of poor responses, thus enhancing the reliability and validity of survey data. Through numerical experiments, the paper demonstrates the applicability and effectiveness of the proposed model. The study’s approach is based on a thorough analysis of response patterns and the integration of quality control questions within the survey structure. The findings of this research have significant implications for the field of survey methodology, providing a more robust and systematic way of ensuring data integrity in web-based surveys.

Hadamard matrix is defined as a square matrix where any components are -1 or +1, and where any pairs of rows are mutually orthogonal. On the other hand, Hadamard-type matrix on finite fields has been proposed. This matrix is a similar one as a binary Hadamard matrix, but has multi-valued components on finite fields. To be more specific, we consider n×n matrices that have their elements on the given finite fields GF(p), and satisfy HH^T ≡ nI mod p, where I is an identity matrix. Any additions and multiplications should be executed under modulo p. In this paper, the authors introduce some new Hadamard-type matrices found in computer searches as well as their properties. Specifically, we define special types of Hadamard-type matrices called cyclic Hadamard-type matrices on finite fields, and propose the methods to generate them. In addition, it is shown that the order of an arbitrary Hadamard-type matrix of odd order is limited to quadratic residues of the given prime p. Some methods to extend the order of Hadamard-type matrices are also discussed.

A combinatorial optimization problem is the problem of minimizing the energy function among many combinations of variables, which is often difficult to solve with conventional classical computers. Recently, Ising machines, including quantum annealers, have gained attention as a promising architecture for efficiently solving combinatorial optimization problems. Among various methods for solving such problems using Ising machines, one prominent approach is the three-stage annealing method. The approach effectively solves a combinatorial optimization problem, utilizing an initial solution, but it performs the annealing process only once. Repeating the annealing process several times may enhance the solution more efficiently. In this paper, we propose a novel hybrid iterative annealing method that consists of an initial process using a classical computer, an annealing process using a quantum annealer, and a correction process/selection process using a classical computer. The proposed method repeats the annealing process and the correction process/selection process until the solution is sufficiently converged. In the experimental evaluations through the three types of typical combinatorial optimization problems, the proposed method shows improvements by up to 54.0% compared to the three-stage annealing method.

Fast, accurate, and energy-efficient object detection is increasingly important for edge applications, such as Internet of Things (IoT) devices. Among various convolutional neural network (CNN)-based methods, the You-Only-Look-Once (YOLO) algorithm series is regarded as one of the promising methods for real-time object detection due to its optimal balance between speed and accuracy. However, deploying YOLO on resource and power-constrained devices like field-programmable gate arrays (FPGAs) poses significant challenges due to the high demand for multiply-and-accumulate (MAC) operations and the corresponding significant off-chip memory accesses. This paper introduces an FPGA-based accelerator for the YOLOv6 algorithm, implemented on a VC707 FPGA board with a Virtex-7 VX485T chip, achieving satisfying throughput, accuracy, and energy efficiency. To our knowledge, this is the first FPGA implementation of YOLOv6. Unlike previous works that utilized early YOLO versions, our design deploys the hardware-friendly YOLOv6, achieving a mean average precision (mAP) of 84.9% on the PASCAL VOC2007 dataset at a 352*352 resolution - significantly outperforming most existing object detection implementations. Through model optimizations for FPGA deployment, such as changing from SiLU to ReLU activation, lowering input resolution, and applying quantization-aware training, we are able to greatly reduce computational cost with minimal accuracy loss. Furthermore, these optimizations allow for the entire YOLOv6 model to be stored in on-chip memory, eliminating the need for energy-intensive DRAM access. The proposed accelerator design and the convolution lowering technique also contribute to high processing speed and energy efficiency. Experimental results demonstrate that our accelerator can process 364.5 frames per second (fps) at 150 MHz on the Virtex-7 VX485T FPGA, achieving excellent power efficiency of 19.75 fps/W.

To overcome the excessive memory capacity of non-volatile CiM (NV-CiM) for multi-modal AI, this paper proposes Mapping-oriented enhanced-FPN (Feature Pyramid Network) fusion (More-FPN) as an RGB-event fusion object detection model. More-FPN includes three proposals. First proposal, Mapping-aware Group Convolution (MAGC), reduces the required NV-CiM capacity by suppressing the number of subarrays in NV-CiM at a fixed subarray size. In MAGC, the number of groups is optimized with no inference accuracy degradation. By adopting MAGC to FPN fusion of an RGB-event fusion object detection model, 54.7% subarrays are reduced. The second proposal, Separable Bridge (SepBridge), further reduces the number of subarrays by 26.1% from MAGC-adopted FPN fusion. Third proposal, Top-down path trainable BiFPN (TDT-BiFPN), achieves accuracy improvement with a slight subarray increase by adding bottom-up path and making top-down path trainable. By combining three proposals, More-FPN achieves both the reduction in subarrays by 61% and the accuracy improvement by 4.6%, compared with conventional FPN fusion CiM.

A soft error in LSI is a temporary malfunction in which signals in combinational circuits or data stored in registers are flipped. Double modular redundancy performs computation execution and data storing in duplicate, detects soft errors through comparison, and corrects errors by re-executing the computation. It is preferable in terms of LSI area and power consumption compared to triple modular redundancy. In this paper, a double modular redundancy design for LSI controllers is proposed. The register for the control step and the combinational circuit to compute the control step value are doubled to check an error in the controller. Re-execution of operations necessary to correct an error in datapath and controller is controlled using one bit signal which is also doubled for error detection and correction. The area of the proposed controller is reduced to about half of that of the conventional triple modular redundant controller.

In this paper, SDG channel routing algorithm for generalized channel is proposed. In generalized channel, it is assumed that horizontal routing capacity is tight and that all pins of a net are needed to be connected by a Single Trunk Steiner Tree. Also, it is requested to reduce the total length of nets to accomplish the routing, and our goal is to reduce the total vertical length as much as possible. Our proposed algorithm determines the track assignment of nets iteratively according to the net priority that is defined to reduce the vertical length. In experiments, it is confirmed that the vertical length is reduced by around 30% to 50% compared with the track assignment by Left-Edge, and that it is very close to a lower bound.

An algorithm for three-layer bottleneck channel routing problem that uses ILP is proposed. The proposed algorithm determines the track and layer assignment of nets for problems with layout constraints in which pins of each net are placed on the upper boundary of the adjacent regions on both sides of the bottleneck channel. The proposed algorithm restricts the routing pattern of each net to one of three patterns by taking feasibility into account, and outputs a solution in a few seconds when the number of nets is 300.

In this paper, a routing problem for advanced chip designs is modeled as Gridless Gap Channel Routing (GGCR). GGCR is a routing problem to allocate variable-width trunks of nets to as fewer gaps as possible where a gap is defined horizontally between obstacles arranged regularly in the routing region. We propose Ceiling-and-Packing algorithm (CAP) for GGCR. CAP allocates the trunk of a net repeatedly so that each gap is filled as much as possible by adopting an appropriate order of allocation, and uses fewer gaps to complete the routing compared with conventional algorithms.

This paper reports our open source cell library characterizer which can generate timing models and power models of standard cell library and its multitread implementation. Implementation results show 14.5 times speed-up from single-thread operation.

To improve multiple input multiple output (MIMO) radar performance and reduce computation complexity, we propose a thinned waveform design algorithm. We form an optimization problem, which the objective function is combined with signal-to-interference-and-noise-ratio (SINR) and Cramer-Rao lower bound (CRLB). Waveform code, waveform selection and antenna selection can be as optimization variables. We define the waveform code-antenna selection vector, which can denote waveform code and antenna simultaneous selection. Due to the problem is multivariate non-convex optimization, we propose a sequential iterative optimization algorithm. The problem is decomposed into two subproblems about waveform code and code-antenna selection vector. To optimize the code-antenna selection vector, we transform the optimization problem into a conventional convex problem by logarithms and first-order Taylor expansions. For waveform code optimization, we introduced an auxiliary variable and solved it by Alternating Direction Method of Multipliers (ADMM). Convex (CVX) can solve the subproblems. The simulation result shows that the better performance and lower computation complexity by the proposed approach than other methods.

This paper analyzes the superstable periodic orbits (SSPOs) generated by a piecewise-constant chaotic spiking oscillator. For this oscillator, the oscillation and reset of states are repeated, which generates chaos and SSPOs. A one-dimensional return map described by the dynamics is theoretically derived. This map has discontinuous points. A theoretical analysis of border-collision bifurcations is performed to clarify the regions of the SSPOs in the parameter space.

With the fast development of the mobile internet, various data has been generated explosively in recent years. To solve the data redundancy problem caused by IoE, knowledge graph construction is considered one of the indispensable techniques for performing systematic and accurate representation of data, especially in some specific domains. In this paper, we propose a construction method of panoramic domain-specific knowledge graphs in various domains, which treat sensitive (secret) data, e.g., medical and industrial domains. This method mainly uses the web crawler to obtain data from relevant web pages by category and saves the obtained data as a structured JSON file in the form of dictionaries. This paper takes the military field as an example to construct the domain-specific knowledge graph based on our proposed method. Specifically, the specific domain knowledge graph is stored in Neo4j and MongoDB to provide an intuitive and applied representation of knowledge, respectively. Based on the knowledge graph stored in MongoDB, we develop an intelligent question-answering system in a specific domain, which can better satisfy the information retrieval and knowledge learning of related personnel. Moreover, the template-based question-answering system is designed to effectively solve the problem of semantic repetition of questions. Finally, the constructed knowledge graph and question-answering system are evaluated and analyzed.

The type-II Z-complementary pairs (ZCPs) are used to suppress asynchronous interference in multi-carrier communication systems. There are many methods for the construction of type-II ZCPs, including horizontal concatenation of GCP, generalized Boolean functions, interleaving techniques, and so on. In this paper, two types of type-II binary ZCPs are proposed based on the insertion method, with parameters of (4N + 2, 2N + 1) and zero correlation zone (ZCZ) ratio of 1/2. Also, we obtained a new type of type-II binary ZCPs with parameters of (4N + 2, 3N + 1) and ZCZ_ratio of 3/4. The proposed type-II binary ZCPs have aperiodic auto-correlation sums (AACS) magnitude of 4 and 8 outside the ZCZ zone (except for the last time-shift taking AACS value of zero). In particular, the AACS magnitude of the type-II binary ZCP with parameters of (4N + 2, 3N + 1) is only 4.

Interpolation-based frequency estimation methods can be used to improve the frequency estimation accuracy of discrete Fourier transform (DFT) methods for complex exponential or real sinusoidal signals. However, traditional interpolation methods first need to search for the maximum spectral line and its adjacent spectral lines in order to interpolate the frequency estimate. This type of method has a low degree of flexibility and does not make full use of the effective information in the frequency domain. In order to solve this problem, this paper proposes a scalable frequency estimation method based on the multiple point interpolation of trigonometry, eliminating the need to find the peak of the spectrum and using multiple point spectral information to improve the frequency estimation accuracy. This paper first derives the formula for frequency estimation of complex sinusoidal signals using multiple spectral lines by using the trigonometric constant equation, then analyses the effects of frequency interval and number of selected frequency points on the frequency estimation error, and finally verifies the estimation performance of the proposed estimator against competing estimators by simulation. Simulation results show that the root mean square error (RMSE) of the estimator is closer to Cramér-Rao Lower Bound than those of the competing estimators over the whole effective signal-to-noise ratio range.

Clock synchronization represents an essential necessity for EtherCAT networks. This letter proposes a practical and efficient clock drift compensation scheme for the master-slave clock synchronization in EtherCAT networks, which decouples and then minimizes the adverse effect of master jitters for synchronization without degrading the clock synchronization performance between slave devices. Moreover, our method requires no modifications to the original EtherCAT protocol, nor does it introduce any additional bandwidth overhead. Comparative experimental results demonstrate the performance enhancement of the proposed approach over existing methods.

Terahertz (THz) Integrated sensing and communication (ISAC) is viewed as one of the most promising technologies for future 6G wireless communications, which can simultaneously enable data transmission at terabits-per-second (Tbps) rates and millimeter-level accurate sensing. To meet the communication requirements of high-mobility users, an efficient beam tracking method is crucial. However, existing beam tracking methods have a high beam training overhead and cannot cope with the severe performance loss caused by the beam split effect under the wideband system. In this letter, we propose a wideband THz beam tracking method based on the ISAC system, which introduces the time-delay network in the precoding to cope with the challenge of the beam split effect and reduce the beam training overhead with the aid of radar sensing. Simulation results show that the method can achieve near-optimal sum-rate performance with only little beam overhead compared to conventional methods.