Artificial Intelligence and Robotics

This paper proposes a new approach to detecting neural Trojans on Deep Neural Networks during inference. This approach is based on monitoring the inference of a machine learning model, computing the attribution of the model’s decision on different features of the input, and then statistically analyzing these attributions to detect whether an input sample contains the Trojan trigger. The anomalous attributions, aka misattributions, are then accompanied by reverse-engineering of the trigger to evaluate whether the input sample is truly poisoned with a Trojan trigger. We evaluate our approach on several benchmarks, including models trained on MNIST, Fashion MNIST, and German Traffic Sign Recognition Benchmark, and demonstrate the state of the art detection accuracy.

A “triaxial velocity sensor” consists of three uniaxial velocity sensors, which are nominally identical, orthogonally oriented among themselves, and co-centered at one point in space. A triaxial velocity sensor measures the acoustic particle velocity vector, by its three Cartesian components, individually component-by-component, thereby offering azimuth-elevation two-dimensional spatial directivity, despite the physical compactness that comes with the collocation of its three components. This sensing system's azimuth-elevation beam-pattern has been much analyzed in the open literature, but only for an idealized case of the three uniaxial velocity sensors being exactly identical in gain. If this nominal identity is violated among the three uniaxial velocity sensors, as may occur in practical hardware, what would happen to the corresponding “spatial matched filter” beam-pattern's peak direction? How would this effective peak direction deviate from the nominal “look direction”?...

Ant colony optimization algorithms have long been touted as providing an effective and efficient means of generating high quality solutions to NP-hard optimization problems. Unfortunately, while the structure of the algorithm is easy to parallelize, the nature and amount of communication required for parallel execution has meant that parallel implementations developed suffer from decreased solution quality, slower runtime performance, or both. This thesis explores a new strategy for ant colony parallelization that involves Area of Expertise (AOE) learning. The AOE concept is based on the idea that individual agents tend to gain knowledge of different areas of the search space when left to their own devices. After developing a sense of their own expertness on a portion of the problem domain, agents share information and incorporate knowledge from other agents without having to experience it first-hand. This thesis shows that when incorporated within parallel ACO and applied to multi-...

Task engagement is defined as loadings on energetic arousal (affect), task motivation, and concentration (cognition) [1]. It is usually challenging and expensive to label cognitive state data, and traditional computational models trained with limited label information for engagement assessment do not perform well because of overfitting. In this paper, we proposed two deep models (i.e., a deep classifier and a deep autoencoder) for engagement assessment with scarce label information. We recruited 15 pilots to conduct a 4-h flight simulation from Seattle to Chicago and recorded their electroencephalograph (EEG) signals during the simulation. Experts carefully examined the EEG signals and labeled 20 min of the EEG data for each pilot. The EEG signals were preprocessed and power spectral features were extracted. The deep models were pretrained by the unlabeled data and were fine-tuned by a different proportion of the labeled data (top 1%, 3%, 5%, 10%, 15%, and 20%) to learn new representations for engagement assessment. The models were then tested on the remaining labeled data. We compared performances of the new data representations with the original EEG features for engagement assessment. Experimental results show that the representations learned by the deep models yielded better accuracies for the six scenarios (77.09%, 80.45%, 83.32%, 85.74%, 85.78%, and 86.52%), based on different proportions of the labeled data for training, as compared with the corresponding accuracies (62.73%, 67.19%, 73.38%, 79.18%, 81.47%, and 84.92%) achieved by the original EEG features. Deep models are effective for engagement assessment especially when less label information was used for training.

Researchers have used distributed constraint optimization problems (DCOPs) to model various multi-agent coordination and resource allocation problems. Very recently, Ottens et al. proposed a promising new approach to solve DCOPs that is based on confidence bounds via their Distributed UCT (DUCT) sampling-based algorithm. Unfortunately, its memory requirement per agent is exponential in the number of agents in the problem, which prohibits it from scaling up to large problems. Thus, in this article, we introduce two new sampling-based DCOP algorithms called Sequential Distributed Gibbs (SD-Gibbs) and Parallel Distributed Gibbs (PD-Gibbs). Both algorithms have memory requirements per agent that is linear in the number of agents in the problem. Our empirical results show that our algorithms can find solutions that are better than DUCT, run faster than DUCT, and solve some large problems that DUCT failed to solve due to memory limitations.

Background. We previously developed a deep-learning (DL) network for image denoising in SPECT-myocardial perfusion imaging (MPI). Here we investigate whether this DL network can be utilized for improving detection of perfusion defects in standard-dose clinical acquisitions. Methods. To quantify perfusion-defect detection accuracy, we conducted a receiver-operating characteristic (ROC) analysis on reconstructed images with and without processing by the DL network using a set of clinical SPECT-MPI data from 190 subjects. For perfusion-defect detection hybrid studies were used as ground truth, which were created from clinically normal studies with simulated realistic lesions inserted. We considered ordered-subset expectationmaximization (OSEM) reconstruction with corrections for attenuation, resolution, and scatter and with 3D Gaussian post-filtering. Total perfusion deficit (TPD) scores, computed by Quantitative Perfusion SPECT (QPS) software, were used to evaluate the reconstructed images. Results. Compared to reconstruction with optimal Gaussian post-filtering (sigma = 1.2 voxels), further DL denoising increased the area under the ROC curve (AUC) from 0.80 to 0.88 (P-value < 10 24 ). For reconstruction with less Gaussian post-filtering (sigma = 0.8 voxels), thus better spatial resolution, DL denoising increased the AUC value from 0.78 to 0.86 (Pvalue < 10 24 ) and achieved better spatial resolution in reconstruction. Conclusions. DL denoising can effectively improve the detection of abnormal defects in standard-dose SPECT-MPI images over conventional reconstruction. (J Nucl Cardiol 2021

This article describes the field application of small, low-cost robots for remote surface data collection and an automated workflow to support water balance computations and hydrologic understanding where water availability data is sparse. Current elevation measurement approaches, such as manual surveying and LiDAR, are costly and infrequent, leading to potential inefficiencies for quantifying the dynamic hydrologic storage capacity of the land surface over large areas. Experiments to evaluate a team of two different robots, including an unmanned aerial vehicle (UAV) and an unmanned surface vehicle (USV), to collect hydrologic surface data utilizing sonar and visual sensors were conducted at three different field sites within the Arkavathy Basin river network located near Bangalore in Karnataka, South India. Visual sensors were used on the UAV to capture high resolution imagery for topographic characterization, and sonar sensors were deployed on the USV to capture bathymetric readings; the data streams were fused in an automated workflow to determine the storage capacity of agricultural reservoirs (also known as "tanks") at the three field sites. This study suggests: (i) this robot-assisted methodology is low-cost and suitable for novice users, and (ii) storage capacity data collected at previously unmapped locations revealed strong power-type relationships between surface area, stage, and storage volume, which can be incorporated into modeling of landscape-scale hydrology. This methodology is of importance to water researchers and practitioners because it produces local, high-resolution representations of bathymetry and topography and enables water balance computations at small-watershed scales, which offer insight into the present-day dynamics of a strongly human impacted watershed.

This article describes the field application of small, low-cost robots for remote surface data collection and an automated workflow to support water balance computations and hydrologic understanding where water availability data is sparse. Current elevation measurement approaches, such as manual surveying and LiDAR, are costly and infrequent, leading to potential inefficiencies for quantifying the dynamic hydrologic storage capacity of the land surface over large areas. Experiments to evaluate a team of two different robots, including an unmanned aerial vehicle (UAV) and an unmanned surface vehicle (USV), to collect hydrologic surface data utilizing sonar and visual sensors were conducted at three different field sites within the Arkavathy Basin river network located near Bangalore in Karnataka, South India. Visual sensors were used on the UAV to capture high resolution imagery for topographic characterization, and sonar sensors were deployed on the USV to capture bathymetric readings; the data streams were fused in an automated workflow to determine the storage capacity of agricultural reservoirs (also known as "tanks") at the three field sites. This study suggests: (i) this robot-assisted methodology is low-cost and suitable for novice users, and (ii) storage capacity data collected at previously unmapped locations revealed strong power-type relationships between surface area, stage, and storage volume, which can be incorporated into modeling of landscape-scale hydrology. This methodology is of importance to water researchers and practitioners because it produces local, high-resolution representations of bathymetry and topography and enables water balance computations at small-watershed scales, which offer insight into the present-day dynamics of a strongly human impacted watershed.

In recent years softmax, together with its fast approximations, has become the defacto loss function for deep neural networks with multiclass predictions. However, softmax is used in many problems that do not fully fit the multiclass framework and where the softmax assumption of mutually exclusive outcomes can lead to biased results. This is often the case for applications such as language modeling, nextevent prediction, and matrix factorization, where many of the potential outcomes are not mutually exclusive, but are more likely to be conditionally independent given the state. To this end, for the set of problems with positive and unlabeled data, we propose Partially Mutual Exclusive Softmax (PMES), a relaxation of the original softmax formulation, where, given the observed state, each of the outcomes are conditionally independent but share a common set of negatives. Since we operate in a regime where explicit negatives are missing, we create a cooperatively-trained model of negatives, and derive a new negative sampling and weighting scheme which we call Cooperative Importance Sampling (CIS). We show empirically the advantages of our newly introduced negative sampling scheme by plugging it into the Word2Vec algorithm and evaluating it extensively against other negative sampling schemes on both language modeling and matrix factorization tasks, where we show large lifts in performance. Learning from positive and unlabeled data is a well-defined task in machine learning, known as positive-unlabeled (PU) learning ( & Noto (2008); Li & Liu (2003); Lee & Liu (2003); Ren et al. (2014); Paquet & Koenigstein ( )). The applications of PU learning are numerous, as in many fields negative data is either too expensive to obtain or too hard to define. This is the case in language modeling, where one only observes examples of valid sentences and documents, and tries to learn a generative process. Similarly, in computer vision, with the recent work on Generative Adversarial Networks (GANs), one tries to learn the underlying generative process that produces meaningful images. In both cases, the space of negatives (e.g., non-sentences or non-images) is not clear. Similarly, in matrix factorization, most of the matrices contain pairs of observed interactions, and there are no available explicit negatives. In all of the applications enumerated above, it is quite common to encounter solutions that are based on a softmax loss function. In language modeling, and more recently in matrix factorization, the Word2Vec algorithm (see Mikolov et al. (2013a)) models the conditional probability of observing all possible items in the vicinity of a context item as a categorical distribution using the softmax loss. However, modeling the probability of co-occurrence as a categorical distribution is a very biased assumption that is clearly refuted by the data, since for all context words there are multiple words that co-occur at the same time.

Credit card fraud is a common vice that affects not only the financial institutions issuing credit cards but the card holders themselves hence the need to address issues related to it by having proper detection measures in place. Thus, in this project , we seek to carry out the following analysis: We look critically at the machine learning techniques that can be used for credit card fraud detection in order to assess the effectiveness of using various machine learning techniques in enhancing precision and speed of the detection process. The technique of Data preprocessing have also used. This step is used to address problems related to missing values, outliers and class imbalance. We then go onto building up a machine learning pipeline that includes various classifiers such as Logistic Regression, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Decision Trees, Random Forests, and the Stochastic Gradient Descent Classifier. In our testing approach, which involves repeating the exploration phase several times with a different classifier, we can also measure its performance based on the provided set of generating metrics including precision, recall, F1-scores, ROC-AUC scores, and accuracy. To counter this we use techniques such as Synthetic Minority Over-sampling Technique (SMOTE) for creating synthetic samples of the distributions of the samples categorized as the minority class. Moreover, hyperparameters for each feature selection method are optimized using grid search with cross-validation since proper tuning improves the model's performance. The major goal of the study is to compare and analyze the comparative efficacy of different classifiers to arrive at an optimal approach for credit card fraud detection through the application of data science machine learning models. From this perspective, our paper is intended to offer insights into the advantages and disadvantages of deploying specific techniques which will help the stakeholders to make right decision on utilizing the fraud detection systems. Using the latest machine learning approaches and careful assessment strategies, we aim at improving the effectiveness of fraud detection tools, minimizing the adverse financial impacts of fraudulent activities, and maintaining the public's confidence i n automated security systems. Our work has significant implications for the state-of-the-art of fraud detection in the financial sector, and provides valuable information and insights into the fight against increasingly popular credit card.

The way communication platforms are used in military operations has changed a lot over the years. They're now essential for mission success, quick decision-making, and maintaining strategic advantages. This paper dives into the modern communication tools that are making waves in military settings, with a spotlight on networked communication, signal support, cybersecurity strategies, and how different forces work together in joint and multinational missions. It also tackles some of the major hurdles, like congestion in the electromagnetic spectrum, cyber threats, and the need for secure data transmission in challenging environments. By pulling insights from FM 6-02: Signal Support to Operations and the latest scientific research, this paper highlights recent advancements in military communication tech and how they boost operational effectiveness. Additionally, the research looks ahead at future possibilities, such as AI-driven communication platforms, quantum encryption, and cutting-edge satellite networks for defense purposes. This paper's primary contribution is the development of a structured, AI-enabled communication workflow that integrates quantum-safe encryption, blockchain authentication, and satellite-based coordination to improve decision-making and resilience in multi-domain operations.

Trademarked names, logos, and images may appear in this book. Rather than use a trademark symbol with every occurrence of a trademarked name, logo, or image we use the names, logos, and images only in an editorial fashion and to the benefit of the trademark owner, with no intention of infringement of the trademark. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. While the advice and information in this book are believed to be true and accurate at the date of publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect to the material contained herein.

This book has been informed by the many conversations I have had with colleagues and students about deep learning, in particular those with Robert Ross and Giancarlo Salton. This book is dedicated to my sister Elizabeth (Liz) Kelleher in recognition of her love and support, and her patience with a brother who can't stop explaining things.

In the era of digital commerce, customer reviews play a crucial role in shaping customer decisions and business strategies. They have an impact on price, target audience segmentation, marketing, product positioning, brand loyalty and purchase decisions. Through the utilization of sentiment analysis, e-commerce platforms can glean insights from the textual comments they receive from customers. This project presents a dual-purpose machine learning approach for sentiment analysis and fake review detection in customer feedback data. It focuses on sentiment analysis of customer reviews and fake reviews detection using machine learning techniques to classify real reviews as positive, negative or neutral. We utilized a dataset of customer reviews. By leverging Natural Language Processing (NLP) and machine learning algorithms such as Logistic Regression, Random Forest, K-Nearest Neighbors, Decision Tree, Gradient Boosting, XGBoost, CatBoost, Support Vector Machine, Naïve Bayes, the system efficiently detects real reviews and extracts meaningful insights from textual data. Performance comparison is made using barchart and confusion matrices. We evaluate their performance using various metrics including accuracy, precision, recall and F1 score and rank the models comparatively based on these results.. The results affirm that well-optimized traditional machine learning models can effectively detect fake content and extract sentiment without the need for complex deep learning frameworks. This approach holds strong potential for real-world deployment in e-commerce platforms, review-based recommendation engines, and feedback moderation systems.

Cyber-Physical Systems (CPSs) are complex systems that integrate physical systems with their counterpart cyber components to form a close loop solution. Due to the ability of deep learning in providing sensor data-based models for analyzing physical systems, it has received increased interest in the CPS community in recent years. However, developing vision data-based deep learning models for CPSs remains critical since the models heavily rely on intensive, tedious efforts of humans to annotate training data. Besides, most of the models have a high tradeoff between quality and computational cost. This research studies deep learning algorithms to achieve affordable and upgradable network architecture which will provide better performance. Two important applications of CPS are studied in this work. In the first study, a Mask Region-based Convolutional Neural Network (Mask R-CNN) was adopted to segment regions of interest from surveillance videos of manufacturing plants. Then, the Mask R-CNN model was modified to have consistent detection results from videos using temporal coherence information of detected objects. This method was extended to the second study, a task of bridge inspection to detect and segment critical structural components. A cellular automata-based pattern recognition algorithm was integrated with the Mask R-CNN model to find the crack propagation rate in the structural components. Decision-makers can make a maintenance decision based on the rate. A discrete event simulation model was also developed to validate the proposed methodology. The work of this research demonstrates approaches to developing and implementing vision data-based deep neural networks to make the CPS more affordable, scalable, and efficient. v ACKNOWLEDGMENTS I would like to express the deepest appreciation to my advisor, Dr. Ruwen Qin, for her continuous support throughout my master's study and research. She was always patient and expertly guided me. Her motivation and enthusiasm always kept me engaged in my research. Without her, I couldn't have reached this stage. I would also like to extend my sincere gratitude to Dr. Dagli and Dr. Corns. It was a great experience to work with them during all the course works. Their comments and discussion during classes were very helpful. Their mentoring and encouragement was especially valuable. I am also grateful to Dr. Zhaozheng Yin and Dr. Genda Chen for their support and insightful comments about my research.

Interest in the field of Explainable Artificial Intelligence has been growing for decades and has accelerated recently. As Artificial Intelligence models have become more complex, and often more opaque, with the incorporation of complex machine learning techniques, explainability has become more critical. Recently, researchers have been investigating and tackling explainability with a user-centric focus, looking for explanations to consider trustworthiness, comprehensibility, explicit provenance, and context-awareness. In this chapter, we leverage our survey of explanation literature in Artificial Intelligence and closely related fields and use these past efforts to generate a set of explanation types that we feel reflect the expanded needs of explanation for today&#39;s artificial intelligence applications. We define each type and provide an example question that would motivate the need for this style of explanation. We believe this set of explanation types will help future system ...

We have witnessed an increased use of technology in every facet of our lives. These technologies come with great promises, such as enabling more independent living for older adults or people with physical disabilities, yet also fears, for instance, over privacy concerns or trust in automated systems. In this Topical Collection, we focus on Active and Assisted Living (AAL) technologies, which require trustworthiness and adherence to privacy regulations for successful adoption. The Collection contains six selected papers that address themes like privacy-by-design, trust in AI, and balancing privacy with technological innovation under regulations like GDPR and the AI Act. The presented articles emphasize the user-centered, privacy-friendly approaches to AAL designs, robust regulatory frameworks, and interdisciplinary methodologies to ensure ethical, trustworthy technologies.

Conceptual research on robots and privacy has increased but we lack empirical evidence about the prevalence, antecedents, and outcomes of different privacy concerns about social robots. To fill this gap, we present a survey, testing a variety of antecedents from trust, technology adoption, and robotics scholarship. Respondents are most concerned about data protection on the manufacturer side, followed by social privacy concerns and physical concerns. Using structural equation modeling, we find a privacy paradox, where the perceived benefits of social robots override privacy concerns.

The increase in electrical metering has created tremendous quantities of data and, as a result, possibilities for deep insights into energy usage, better energy management, and new ways of energy conservation. As buildings are responsible for a significant portion of overall energy consumption, conservation efforts targeting buildings can provide tremendous effect on energy savings. Building energy monitoring enables identification of anomalous or unexpected behaviors which, when corrected, can lead to energy savings. Although the available data is large, the limited availability of labels makes anomaly detection difficult. This research proposes a deep semi-supervised convolutional neural network with confidence sampling for electrical anomaly detection. To achieve semi-supervised learning, two sub-networks are used: the first performs reconstruction and uses unlabelled data, while the second performs classification with labelled data. The two sub-networks overlap: the encoder parameters are shared between the two. To quantify anomaly detection confidence, a valuable metric in anomaly detection, the network uses a dropout sampling method. The proposed approach has been evaluated with real-world electrical data from systems such as HVAC, lighting, and heat pumps. The results demonstrated the accuracy of the proposed anomaly detection solution.

Benchmarking makes it possible to identify low-performing buildings, establishes a baseline for measuring performance improvements, enables setting of energy conservation targets, and encourages energy savings by creating a competitive environment. Statistical approaches evaluate building energy efficiency by comparing measured energy consumption to other similar buildings typically using annual measurements. However, it is important to consider different time periods in benchmarking because of differences in their consumption patterns. For example, an office can be efficient during the night, but inefficient during operating hours due to occupants' wasteful behavior. Moreover, benchmarking studies often use a single regression model for different building categories. Selecting the regression model based on actual data would ensure that the model fits the data well. Consequently, this paper proposes Energy Slices, an energy benchmarking approach with time slicing for existing buildings. Time slicing enables separation of time periods with different consumption patterns. The regression model suited for the specific scenario is selected using cross validation, which ensures that the model performs well on previously unseen data. The evaluation is carried out on a case study involving two sports arenas; event energy efficiency is benchmarked to identify low-performing events. The case study demonstrates the Energy Slice procedure and shows the importance of model selection.

During building operation, a significant amount of energy is wasted due to equipment and humanrelated faults. To reduce waste, today's smart buildings monitor energy usage with the aim of identifying abnormal consumption behaviour and notifying the building manager to implement appropriate energysaving procedures. To this end, this research proposes a new pattern-based anomaly classifier, the collective contextual anomaly detection using sliding window (CCAD-SW) framework. The CCAD-SW framework identifies anomalous consumption patterns using overlapping sliding windows. To enhance the anomaly detection capacity of the CCAD-SW, this research also proposes the ensemble anomaly detection (EAD) framework. The EAD is a generic framework that combines several anomaly detection classifiers using majority voting. To ensure diversity of anomaly classifiers, the EAD is implemented by combining pattern-based (e.g., CCAD-SW) and prediction-based anomaly classifiers. The research was evaluated using real-world data provided by Powersmiths, located in Brampton, Ontario, Canada. Results show that the EAD framework improved the sensitivity of the CCAD-SW by 3.6% and reduced false alarm rate by 2.7%.

In recent years, advances in sensor technologies and expansion of smart meters have resulted in massive growth of energy data sets. These Big Data have created new opportunities for energy prediction, but at the same time, they impose new challenges for traditional technologies. On the other hand, new approaches for handling and processing these Big Data have emerged, such as MapReduce, Spark, Storm, and Oxdata H2O. This paper explores how findings from machine learning with Big Data can benefit energy consumption prediction. An approach based on local learning with support vector regression (SVR) is presented. Although local learning itself is not a novel concept, it has great potential in the Big Data domain because it reduces computational complexity. The local SVR approach presented here is compared to traditional SVR and to deep neural networks with an H2O machine learning platform for Big Data. Local SVR outperformed both SVR and H2O deep learning in terms of prediction accuracy and computation time. Especially significant was the reduction in training time; local SVR training was an order of magnitude faster than SVR or H2O deep learning.

The demand for knowledge extraction has been increasing. With the growing amount of data being generated by global data sources (e.g., social media and mobile apps) and the popularization of context-specific data (e.g., the Internet of Things), companies and researchers need to connect all these data and extract valuable information. Machine learning has been gaining much attention in data mining, leveraging the birth of new solutions. This paper proposes an architecture to create a flexible and scalable machine learning as a service. An open source solution was implemented and presented. As a case study, a forecast of electricity demand was generated using real-world sensor and weather data by running different algorithms at the same time.

Electricity price, consumption, and demand forecasting has been a topic of research interest for a long time. The proliferation of smart meters has created new opportunities in energy prediction. This paper investigates energy cost forecasting in the context of entertainment event-organizing venues, which poses significant difficulty due to fluctuations in energy demand and wholesale electricity prices. The objective is to predict the overall cost of energy consumed during an entertainment event. Predictions are carried out separately for each event category and feature selection is used to select the most effective combination of event attributes for each category. Three machine learning approaches are considered: k-nearest neighbor (KNN) regression, support vector regression (SVR) and neural networks (NN). These approaches are evaluated on a case study involving a large event venue in Southern Ontario. In terms of prediction accuracy, KNN regression achieved the lowest average error. Error rates varied greatly among different event categories.

Advances in sensor technologies and the proliferation of smart meters have resulted in an explosion of energy-related data sets. These Big Data have created opportunities for development of new energy services and a promise of better energy management and conservation. Sensor-based energy forecasting has been researched in the context of office buildings, schools, and residential buildings. This paper investigates sensor-based forecasting in the context of event-organizing venues, which present an especially difficult scenario due to large variations in consumption caused by the hosted events. Moreover, the significance of the data set size, specifically the impact of temporal granularity, on energy prediction accuracy is explored. Two machine-learning approaches, neural networks (NN) and support vector regression (SVR), were considered together with three data granularities: daily, hourly, and 15 minutes. The approach has been applied to a large entertainment venue located in Ontario, Canada. Daily data intervals resulted in higher consumption prediction accuracy than hourly or 15-min readings, which can be explained by the inability of the hourly and 15-min models to capture random variations. With daily data, the NN model achieved better accuracy than the SVR; however, with hourly and 15-min data, there was no definitive dominance of one approach over another. Accuracy of daily peak demand prediction was significantly higher than accuracy of consumption prediction.

We present a new algorithm for efficient learning of regular languages from examples and queries. A reliable teacher who knows the unknown regular grammar G (or is able to determine if certain strings are accepted by the grammar) will guide the learner in achieving the goal of inferring an equivalent grammar $\mbox{G}^{*}$. The teacher provides the learner with a structurally complete set of positive examples belonging to the unknown grammar G. Using this information the learner constructs a canonical automaton which accepts exactly those examples. The canonical automaton defines a set of grammars which are ordered on a lattice to form the hypothesis space. A bi-directional search algorithm is used to systematically search the lattice for the solution G*. While searching for the solution, the learner interacts with the teacher by posing queries. The teacher's responses enable the learner to eliminate one or more points on the lattice which do not correspond to the correct solution. After successive eliminations, when only one lattice element remains, the process is terminated and that element is accepted as the solution G*. The inferred grammar is proven to be equivalent to original grammar G.

Rapid advancement in the domain of quantum technologies have opened up researchers to the real possibility of experimenting with quantum circuits, and simulating smallscale quantum programs. Nevertheless, the quality of currently available qubits and environmental noise pose a challenge in smooth execution of the quantum circuits. Therefore, efficient design automation flows for mapping a given algorithm to the Noisy Intermediate Scale Quantum (NISQ) computer becomes of utmost importance. State-of-the-art quantum design automation tools are primarily focused on reducing logical depth, gate count and qubit counts with recent emphasis on topology-aware (nearest-neighbour compliance) mapping. In this work, we extend the technology mapping flows to simultaneously consider the topology and gate fidelity constraints while keeping logical depth and gate count as optimization objectives. We provide a comprehensive problem formulation and multi-tier approach towards solving it. The proposed automation flow is compatible with commercial quantum computers, such as IBM QX and Rigetti. Our simulation results over 10 quantum circuit benchmarks, show that the fidelity of the circuit can be improved up to 3.37× with an average improvement of 1.87×.

Needle steering is a technology for guiding needles around sensitive internal obstacles in minimally invasive surgery. Traditional techniques apply rotation at the base of a needle with an asymmetric tip, enabling steering through the redirection of radial forces. Magnetic steering of catheters and continuum manipulators is another technology that allows steering of a shaft in the body. Both of these techniques rely on mechanical or manual shaft advancement methods. Needle steering has not achieved widespread clinical use due to several limitations: 1- buckling and compression effects in the shaft and needle rotation cause excessive tissue damage; 2- torsion effects on the shaft and needle deflection at tissue boundaries lead to difficulty in control; and 3- restricted radius of curvature results in limited workspace. Magnetically steered catheters and continuum manipulators also suffer from limited curvature and the possibility of buckling. This paper proposes a novel needle steeri...

Planning under uncertainty for multiple agents has grown rapidly with the development of formal models such as multi-agent MDPs and decentralized MDPs. But despite their richness, the applicability of these models remains limited due to their computational complexity. We present the class of event-detecting multi-agent MDPs (eMMDPs), designed to detect multiple mobile targets by a team of sensor agents. We show that eMMDPs are NP-Hard and present a scalable 2-approximation algorithm for solving them using matroid theory and constraint optimization. The complexity of the algorithm is linear in the state-space and number of agents, quadratic in the horizon, and exponential only in a small parameter that depends on the interaction among the agents. Despite the worst-case approximation ratio of 2, experimental results show that the algorithm produces near-optimal policies for a range of test problems.

In this paper, we propose a defence strategy to improves adversarial robustness incorporating hidden layer representation. The key of this defence strategy aims to compress or filter input's information including adversarial perturbation. And this defence strategy can be regarded as an activation function which can be applied any kind of neural networks. We also prove theoretically the effectiveness of this defense strategy under certain conditions. Besides, incorporating hidden layer representation we propose three types of adversarial attacks to generate three types of adversarial examples, respectively. The experiments show that our defence method can significantly improve the adversarial robustness of deep neural networks which achieves the state-of-the-art performance even though we do not adopt adversarial training.

Our interactions with an application frequently leave a heterogeneous and contemporaneous trail of actions and adoptions (e.g., clicks, bookmarks, purchases). Given a sequence of a particular type (e.g., purchases)-- referred to as the target sequence, we seek to predict the next item expected to appear beyond this sequence. This task is known as next-item recommendation. We hypothesize two means for improvement. First, within each time step, a user may interact with multiple items (a basket), with potential latent associations among them. Second, predicting the next item in the target sequence may be helped by also learning from another supporting sequence (e.g., clicks). We develop three twin network structures modeling the generation of both target and support basket sequences. One based on &quot;Siamese networks&quot; facilitates full sharing of parameters between the two sequence types. The other two based on &quot;fraternal networks&quot; facilitate partial sharing of paramete...

A 28 days experiment was conducted to evaluate the effect of garlic and turmeric as additives on broiler chicken growth performance and gastrointestinal microbes at finisher phase. One hundred and eighty broilers (180) day old broiler chickens were utilized in four treatments diets namely T1 serves as the control diet, T2, T3, and T4 representing 0, 2.5g turmeric and 2.5g garlic mixture, 2g turmeric and 3g garlic mixture and 3g turmeric and 2g garlic mixture incorporated into 100kg of feed in a completely randomized design at finisher phase. Feed and water was provided ad libitum, and

This paper addresses quantum circuit mapping for Noisy Intermediate-Scale Quantum (NISQ) computers. Since NISQ computers constraint two-qubit operations on limited couplings, an input circuit must be transformed into an equivalent output circuit obeying the constraints. The transformation often requires additional gates that can affect the accuracy of running the circuit. Based upon a previous work of quantum circuit mapping that leverages gate commutation rules, this paper shows algorithms that utilize both transformation and commutation rules. Experiments on a standard benchmark dataset confirm the algorithms with more rules can find even better circuit mappings compared with the previously-known best algorithms.

Abstrak
Latar Belakang: Adopsi Python yang meluas di berbagai bidang seperti pengembangan web, data science, dan otomasi telah menjadikannya target utama serangan siber. Meskipun memiliki filosofi "batteries included" dan kemudahan penggunaan, aplikasi Python tetap rentan terhadap berbagai ancaman keamanan ketika pengembang kurang memahami praktik coding yang aman.
Tujuan: Panduan komprehensif ini bertujuan membekali pengembang Python dengan pengetahuan praktis dan solusi yang dapat diimplementasikan untuk membangun aplikasi yang aman, menjembatani kesenjangan antara konsep keamanan teoretis dengan implementasi di dunia nyata.
Metode: Artikel ini menyajikan pendekatan sistematis terhadap keamanan Python melalui sembilan topik yang saling terkait: identifikasi ancaman, praktik coding aman, autentikasi dan enkripsi, manajemen dependensi, logging aman, teknik lanjutan, dan integrasi pengujian keamanan. Setiap konsep diilustrasikan dengan contoh kode yang rentan diikuti implementasi yang aman, dengan total lebih dari 50 cuplikan kode praktis.
Temuan Utama: Kerentanan umum dalam aplikasi Python berasal dari validasi input yang tidak tepat, deserialisasi yang tidak aman, kredensial yang di-hardcode, dan manajemen dependensi yang tidak memadai. Artikel ini mendemonstrasikan bahwa kerentanan-kerentanan tersebut dapat dimitigasi secara efektif melalui pola-pola yang telah teruji seperti sanitasi input, penggunaan library kriptografi yang tepat, konfigurasi berbasis environment, dan pengujian keamanan otomatis.
Implikasi Praktis: Dengan mengimplementasikan praktik keamanan yang diuraikan dalam panduan ini, pengembang dapat secara signifikan mengurangi attack surface aplikasi mereka. Artikel menyediakan kode siap pakai untuk hashing password dengan Argon2/bcrypt, operasi file yang aman, environment eksekusi sandbox, dan integrasi keamanan CI/CD, memungkinkan penerapan langsung dalam proyek nyata.
Kesimpulan: Keamanan dalam pengembangan Python adalah proses berkelanjutan yang memerlukan pembelajaran dan kewaspadaan terus-menerus. Panduan ini berfungsi sebagai sumber edukasi sekaligus referensi praktis, memberdayakan pengembang untuk membangun aplikasi yang aman sejak awal (secure by design) bukan sebagai tambahan di akhir. Checklist komprehensif dan sumber daya terkurasi memastikan nilai jangka panjang sebagai referensi keamanan.
Kata Kunci: Keamanan Python, secure coding, validasi input, kriptografi, manajemen dependensi, pengujian keamanan, OWASP, DevSecOps

Embodied cognition posits that the development of thinking skills is distributed among mind, senses, and the environment. Research in this field has resulted into the development of applications in different areas including mathematics. This paper reports one part of a larger series of studies on the design and implementation of embodied cognition-based mathematics educational systems. We describe the evaluation of a game called Estimate It!, a wearable-based game for teaching measurement estimation and geometry. Experts were invited to evaluate the game, resulting in a generally positive rating. The game's collaborative nature, its hands-on way of teaching estimation, and the incorporation of technology were seen as promising points. Infrastructure readiness, classroom control, and adjustment to the new technology were areas of concern.

An increasing need of running Convolutional Neural Network (CNN) models on mobile devices with limited computing power and memory resource encourages studies on efficient model design. A number of efficient architectures have been proposed in recent years, for example, MobileNet, ShuffleNet, and MobileNetV2. However, all these models are heavily dependent on depthwise separable convolution which lacks efficient implementation in most deep learning frameworks. In this study, we propose an efficient architecture named PeleeNet, which is built with conventional convolution instead. On ImageNet ILSVRC 2012 dataset, our proposed PeleeNet achieves a higher accuracy and over 1.8 times faster speed than MobileNet and MobileNetV2 on NVIDIA TX2. Meanwhile, PeleeNet is only 66% of the model size of MobileNet. We then propose a real-time object detection system by combining PeleeNet with Single Shot MultiBox Detector (SSD) method and optimizing the architecture for fast speed. Our proposed dete...

A reliable and searchable knowledge database of adverse drug reactions (ADRs) is highly important and valuable for improving patient safety at the point of care. In this paper, we proposed a neural multi-task learning system, NeuroADR, to extract ADRs as well as relevant modifiers from free-text drug labels. Specifically, the NeuroADR system exploited a hierarchical multi-task learning (HMTL) framework to perform named entity recognition (NER) and relation extraction (RE) jointly, where interactions among the learned deep encoder representations from different subtasks are explored. Different from the conventional HMTL approach, NeuroADR adopted a novel task decomposition strategy to generate auxiliary subtasks for more inter-task interactions and integrated a new label encoding schema for better handling discontinuous entities. Experimental results demonstrate the effectiveness of the proposed system.

A reliable and searchable knowledge database of adverse drug reactions (ADRs) is highly important and valuable for improving patient safety at the point of care. In this paper, we proposed a neural multi-task learning system, NeuroADR, to extract ADRs as well as relevant modifiers from free-text drug labels. Specifically, the NeuroADR system exploited a hierarchical multi-task learning (HMTL) framework to perform named entity recognition (NER) and relation extraction (RE) jointly, where interactions among the learned deep encoder representations from different subtasks are explored. Different from the conventional HMTL approach, NeuroADR adopted a novel task decomposition strategy to generate auxiliary subtasks for more inter-task interactions and integrated a new label encoding schema for better handling discontinuous entities. Experimental results demonstrate the effectiveness of the proposed system.

Clustering still leaves problems in selecting optimal clusters in order to obtain a right and correct classification analysis. Right in the sense of the number of clusters, while correct in terms of the information generated by a group of cluster members that is optimally grouped. Determining the optimal number of clusters is a difficult problem in non-polynomials. A number of existing approaches generally still rely on the number of K tests tested. This study aims to produce a new approach that can determine and place data in clusters optimally in a combinatorial form. This can be done by considering that the problem of selecting cluster placement has a combinatorial optimization structure pattern. However, the resulting combinatorial optimization model is quadratic. Therefore, in order to make the combinatorial clustering problem easier to solve, linearization of the cluster data was carried out so that a combinatorial optimization approach was produced with the algorithm. Several...

Artificial intelligence (AI) driven by machine learning has revolutionized the identification of financial fraud in Internet of Things (IoT) environments. This technique quickly and accurately identifies suspicious patterns in the vast and diverse data streams from IoT devices through the application of advanced algorithms, potentially identifying fraudulent activity. Using methods like anomaly detection and clustering, along with supervised and unsupervised learning that are trained on historical transaction data, artificial intelligence systems are able to distinguish between legitimate and fraudulent transactions with great accuracy in real time. In order to create trustworthy fraud detection models in Internet of Things environments, this study looks at the methodology, datasets, and assessment metrics that are necessary for adaptive learning through frequent retraining and automatic reaction mechanisms.

Debit card fraud is one of the major financial crimes globally, causing a very great financial losses for financial institutions and individuals. The traditional mode of fraud detection systems often struggles to keep with the latest change in fraud patterns, due to the dynamism of the criminals resulting in high rates of false positives. This project proposes an improved system based on machine learning models to accurately and effectively identify fraudulent transactions. With machine learning models, fraudulent activities can be monitored and identified in real time. It is able to adapt to the changing nature or approach of fraudsters due to advancement in technology unlike the traditional model that is static in nature. Machine learning approach to fraud detection will mitigate the instances of false positive. This project focuses on utilizing machine learning algorithms, namely Random Forest (RF) and Logistic Regression for detecting debit card fraud. A

An increasing number of studies apply machine learning techniques to corporate fraud detection, and most claim that machine learning is more efficient than traditional methods at this task. The primary purpose of this research is to assess the efficiency of machine learning in detecting accounting fraud. Using a sample of Chinese firms sanctioned for accounting fraud between 2007 and 2022 and by combining powerful machine learning and SAS visualisation techniques, we find mixed results. On the one hand, unlike many recent studies, we fail to generate robust and conclusive evidence that existing machine learning techniques are more efficient than traditional logistic regression in detecting accounting fraud. On the other hand, we find that deep learning offers great potential for fraud detection. Our study contributes to a better understanding of the application of machine learning in accounting research.

The increasing lifespan and large individual differences in cognitive capability highlight the importance of comprehending the aging process of the brain. Contrary to visible signs of bodily ageing, like greying of hair and loss of muscle mass, the internal changes that occur within our brains remain less apparent until they impair function. Brain age, distinct from chronological age, reflects our brain's health status and may deviate from our actual chronological age. Notably, brain age has been associated with mortality and depression. The brain is plastic and can compensate even for severe structural damage by rewiring. Functional characterization offers insights that structural cannot provide. Contrary to the multitude of studies relying on structural magnetic resonance imaging (MRI), we utilize resting-state functional MRI (rsfMRI). We also address the issue of inclusion of subjects with abnormal brain ageing through outlier removal. In this study, we employ the Least Absolute Shrinkage and Selection Operator (LASSO) to identify the 39 most predictive correlations derived from the rsfMRI data. The data is from a cohort of 176 healthy right-handed volunteers, aged 18-78 years (95/81 male/female, mean age 48, SD 17) collected at the Mind Research Imaging Center at the National Cheng Kung University. We establish a normal reference model by excluding 68 outliers, which achieves a leave-one-out mean absolute error of 2.48 years. By asking which additional features that are needed to predict the chronological age of the outliers with a smaller error, we identify correlations predictive of abnormal aging. These are associated with the Default Mode Network (DMN). Our normal reference model has the lowest prediction error among published models evaluated on adult subjects of almost all ages and is thus a candidate for screening for abnormal brain aging that has not yet manifested in cognitive decline. This study advances our ability to predict brain aging and provides insights into potential biomarkers for assessing brain age, suggesting that the role of DMN in brain aging should be studied further.