This study proposes a framework that integrates Transfer Learning (TL) into Multi-Agent Reinforcement Learning (MARL) for real-time, energy-efficient thermal control of the heating stage in thermoforming processes, while concurrently minimizing energy consumption and adapting to varying manufacturing conditions. To enable a scalable framework under these conditions, a Fully Connected Neural Network-based Multi-Agent Proximal Policy Optimization (FCNN-MAPPO) architecture is developed using a multi-objective reward function for concurrently minimizing control error, thermal energy consumption, and instability, while preserving temporal dynamics through state augmentation. The resulting Multi-Agent Transfer Reinforcement Learning (MATRL) framework combines direct parameter transfer for within-scenario learning with experience sharing for cross-scenario adaptation, enabling faster convergence and improved generalization. Results show that under high convective heat transfer conditions, MATRL can reduce training time by 29.3%, decrease energy consumption by 28.1%, and improve average error by 2.02 °C compared to baseline MARL (i.e., without TL). Under elevated ambient temperature conditions, energy usage and settling time were reduced by 42.6% and 41.7%, respectively. Under synchronous multi-parameter variations (extreme conditions of convective heat transfer, sheet conductivity, and ambient temperature), MATRL maintained energy efficiency within 0.5% of nominal conditions while reducing temperature dispersion by 47%, demonstrating robust multidimensional adaptability without retraining. Statistical validation across multiple runs with random seeds showed stable performance, with a coefficient of variation of 7.3% and no divergence.

We introduce Language2Gesture (L2G), a cross-modal generative model designed to predict head gesture animations directly from audio inputs. Unlike existing head gesture prediction models, L2G excels in modelling both macro and subtle micro head gesture motions. L2G further refines the generated motions by conditioning the output of the L2G decoder on the emotional dimensions of Valence, Arousal, and Dominance (VAD). L2G preserves gesture diversity and supports a wide range of expressive behaviors. Our evaluation shows that L2G can directly regress continuous 3D head gestures in multi-speaker settings, outperforming the current state-of-the-art method S3 [1].

We present SpecMAS, a novel multi-agent system that autonomously constructs and formally verifies executable system models from natural language specifications. Given a Standard Operating Procedure (SOP) describing a target system, SpecMAS parses the specification, identifies relevant operational modes, variables, transitions, and properties, and generates a formal model in NuSMV code syntax, an industry-standard symbolic model checker. A dedicated reasoning agent extracts both explicit and implicit properties from the SOP, and verification is performed via temporal logic model checking. If any properties fail to verify, an autonomous debugging agent analyzes counterexamples and iteratively corrects the model until all properties are satisfied. This closed-loop system design guarantees provable correctness by construction and advances the state of the art in automated, interpretable, and deployable verification pipelines. We demonstrate the generality, correctness, and practical feasibility of SpecMAS across a set of representative case studies and propose a new benchmark dataset for the evaluation and comparison of model checking performance.

Quantum Machine Learning (QML) has emerged as a promising field of research, aiming to leverage the capabilities of quantum computing to enhance existing machine learning methodologies. Recent studies have revealed that, like their classical counterparts, QML models based on Parametrized Quantum Circuits (PQCs) are also vulnerable to adversarial attacks. Moreover, the existence of Universal Adversarial Perturbations (UAPs) in the quantum domain has been demonstrated theoretically in the context of quantum classifiers. In this work, we introduce QuGAP: a novel framework for generating UAPs for quantum classifiers. We conceptualize the notion of additive UAPs for PQC-based classifiers and theoretically demonstrate their existence. We then utilize generative models (QuGAP-A) to craft additive UAPs and experimentally show that quantum classifiers are susceptible to such attacks. Moreover, we formulate a new method for generating unitary UAPs (QuGAP-U) using quantum generative models and a novel loss function based on fidelity constraints. We evaluate the performance of the proposed framework and show that our method achieves state-of-the-art misclassification rates, while maintaining high fidelity between legitimate and adversarial samples.

Convolutional neural networks (CNNs) are being increasingly adopted in medical imaging. However, in the race for developing accurate models, their robustness is often overlooked. This elicits a significant concern given the safety-critical nature of the healthcare system. Here, we highlight the vulnerability of CNNs against a sporadic and naturalistic adversarial patch attack (SNAP). We train SNAP to mislead the ResNet50 model predicting metastasis in histopathological scans of lymph node sections, lowering the accuracy by 27%. This work emphasizes the need for defense strategies before deploying CNNs in critical healthcare settings.

Many safety-critical applications have adopted machine learning models like autonomous driving, aviation control, and medical diagnosis. A large number of supervised learning techniques depend on the quality of data. Training data faults make it difficult for models to make correct predictions and may lead to complete failure. It is nearly impossible to scan large data manually to verify its correctness. In this paper, we develop a novel approach to mitigate training data faults by analyzing the data mislabeling using a clustering-based filtering process using features correlation. We evaluate the performance of the proposed approach on various percentages of data faults and observe the accuracy and resilience of the model. Since the performance of the proposed technique does not vary with the percentage of faults in the original data, it has shown a lower value of accuracy delta than state-of-the-art techniques. Thus, it limits effective training of data faults.

Nowadays, vision transformers (ViTs) are one of the most prominent state-of-the-art models for vision-based tasks. ViTs are being used widely in many safety-critical applications ranging from health care to automotive. However, these widespread deployments also bring the risk of adversarial attacks on ViTs to the forefront. Thus, understanding the vulnerability of vision transforms against possible adversarial attacks is necessary before deployment in real-time scenarios. Adversarial patch attacks represent a practical threat to the viability of ViT-based real-world applications. This study delves into the vulnerability of vision transformers to such attacks, exploring both single and multi-patch adversarial attacks to gauge the robustness of vision transformers across benchmark datasets, including CIFAR-10, CIFAR-100, Tiny ImageNet, and ImageNet-1k. Experimentally, our findings reveal that poly-multi patch attacks constitute formidable adversarial threats, with vision transformers exhibiting greater vulnerability to poly-multi attacks than single, mono, and split-multi attacks. Additionally, we qualitatively elucidate the impact of patch location on the efficacy of adversarial attacks, providing insights into the factors influencing their effectiveness. Through this study, we aim to enhance our understanding of vision transformers' susceptibility to adversarial patch attacks, contributing to developing strategies for strengthening their security and resilience in real-world applications.

Adversarial patches threaten visual AI models in the real world. The number of patches in a patch attack is variable and determines the attack’s potency in a specific environment. Most existing defenses assume a single patch in the scene, and the multiple patch scenarios are shown to overcome them. This paper presents a model-agnostic defense against patch attacks based on total variation for image resurfacing (TVR). The TVR is an image-cleansing method that processes images to remove probable adversarial regions. TVR can be utilized solely or augmented with a defended model, providing multi-level security for robust prediction. TVR nullifies the influence of patches in a single image scan with no prior assumption on the number of patches in the scene. We validate TVR on the ImageNet-Patch benchmark dataset and with real-world physical objects, demonstrating its ability to mitigate patch attack.

Few-shot image classification (FSIC) is a computer vision task from the few-shot learning (FSL) category in which the model learns to classify images using only a few training samples. It has been demonstrated that even neural networks trained on large scale datasets are vulnerable to adversarial samples. This vulnerability is magnified in FSIC due to the low volume of training data. This paper proposes the use of hypersphere embedding and supervised contrastive learning to improve the adversarial robustness of representation learning-based FSIC. Contrastive learning contributes through its ability to bring together similar samples while pushing away dissimilar ones. On the other hand, hypersphere embedding has been successful in the representation learning tasks by restricting the embeddings to a hypersphere manifold. The proposed approach was evaluated on both 5-shot and 1-shot learning using two standard FSL networks and the standard Mini-ImageNet benchmark dataset. The evaluation shows that supervised contrastive training provides inherent adversarial robustness to the FSIC model while hypersphere embedding with cosine distance metrics improves the accuracy of the FSIC model and, when used in conjunction with an adversarial defense mechanism, boosts the adversarial performance.

Globally, traffic accidents are the leading cause of death according to reports published by the World Health Organization (WHO). Advanced Driver Assistance Systems (ADAS) are effective in reducing the likelihood or severity of accidents. However, such systems are expensive and inaccessible to the majority of the population. Time-To-Accident (TTA) forecasting is a critical component of ADAS systems as it can allow for better decision-making in traffic, improve dynamic path planning, alert the driver of potential dangers, and more. Sensor or depth imaging-based TTA forecasting can be inaccurate and are unaffordable to the mass. Existing vision-based traffic safety research in the literature focuses on accident anticipation with limited work on TTA prediction. We propose forecasting TTA by utilizing spatio-temporal features from dashboard camera videos and introducing a novel algorithm to compute ground truth TTA values. Additionally, we present a CNN-Transformer Hybrid (CTH) model that overcomes the limitations of end-to-end CNN or Transformer-based models. Our best models can predict TTA values with an average prediction error of 0.24 and 0.80 seconds on the CCD and DoTA dataset respectively. Furthermore, We assess our approach's ongoing scene recognition ability using the TTA-ASiST@x (Time-to-Accident-based Accuracy at the$\mathrm{x}^{th}$frame Since Scene Transition) metric. In real-world scenarios, it is vital to identify when a normal driving scene transitions into a potential accident scene. With just 25% of the data, HyCT achieved over 85% accuracy in recognizing accident and non-accident scenes, further showcasing its performance on limited data.

Aptamers are oligonucleotides or peptides with unique binding properties for specific target molecules, and they have shown great potential in diagnostics, therapeutics, and bio-sensing. However, the current in vitro SELEX-based method for discovering new target-selective aptamers is challenging, time-consuming, and often unsuccessful in finding high-affinity aptamers. Recently, in silico methods have gained immense attention. However, since labeled interaction-pair data collection is expensive and needs highly trained specialists, available data is sparse. Further, since acquiring positive-class samples is even more challenging, available datasets showcase high-class imbalance. This makes designing deep learning models incredibly challenging, as they require a sufficiently large training set and are biased towards the dominant class. Additionally, current models cannot be updated in real-time, and end-to-end re-training is necessary for each new aptamer-target interaction pair discovery. The present work is the first to address both these challenges. We present cAPTured, a novel fuzzy continual learning method for predicting aptamer-target protein interaction pairs in a continual learning environment. cAPTured continually updates its learned feature space on a non-stationary interaction-pair data stream. We performed extensive evaluation studies and experiments to establish the effectiveness of the proposed approach. cAPTured outperforms existing methods on the benchmark dataset by a significant margin.

Visual AI systems are vulnerable to natural and synthetic physical corruption in the real-world. Such corruption often arises unexpectedly and alters the model's performance. In recent years, the primary focus has been on adversarial attacks. However, natural corruptions (e.g., snow, fog, dust) are an omnipresent threat to visual AI systems and should be considered equally important. Many existing works propose interesting solutions to train robust models against natural corruption. These works either leverage image augmentations, which come with the additional cost of model training, or place suspicious patches in the scene to design unadversarial examples. In this work, we propose the idea of naturalistic support artifacts (NSA) for robust prediction. The NSAs are shown to be beneficial in scenarios where model parameters are inaccessible and adding artifacts in the scene is feasible. The NSAs are natural looking objects generated through artifact training using DC-GAN to have high visual fidelity in the scene. We test against natural corruptions on the Imagenette dataset and observe the improvement in prediction confidence score by four times. We also demonstrate NSA's capability to increase adversarial accuracy by 8% on average. Lastly, we qualitatively analyze NSAs using saliency maps to understand how they help improve prediction confidence.

The current security mechanisms for embedded systems often rely on Intrusion Detection System (IDS) running on the system itself. This provides the detector with relevant internal resources but also exposes it to being bypassed by an attacker. If the host is compromised, its IDS can not be trusted anymore and becomes useless. Power consumption offers an accurate and trusted representation of the system’s state that can be leveraged to verify its integrity during the boot sequence. We present a novel IDS that uses the side-channel power consumption of a target device to protect it against various firmware and hardware attacks. The proposed Boot Process Verifier (BPV) uses a combination of rule-based and machine-learning-based side-channel analysis to monitor and evaluate the integrity of different networking equipment with an overall accuracy of 0,942. The BPV is part of a new layer of cybersecurity mechanisms that leverage the physical emissions of devices for protection.

Globally, traffic accidents account for over 3,700 daily deaths, equating to 1.35 million deaths annually. Studies show that collision avoidance systems can significantly reduce the probability and intensity of accidents. Time-to-accident (TTA) is considered the principal parameter for collision avoidance systems allowing for decision-making in traffic, dynamic path planning, and accident mitigation. Despite the importance of TTA, the literature has insufficient research on TTA estimation for traffic scenarios. The majority of recent work focuses on accident anticipation by providing a probabilistic measure of an immediate or future collision. We propose to forecast TTA based on Spatio-temporal features extracted from accident videos obtained via dashboard cameras. Our model can also recognize accident and non-accident scenes with 100% accuracy. Additionally, the impact of spatial resolution and temporal depth on prediction error is analyzed in this work. We implement state-of-the-art video learning architectures and compare the results against static image architectures. Our comprehensive experiments suggest that leveraging Spatio-temporal features is an effective method to estimate TTA. Our best model can estimate the TTA with an average prediction error of 0.30 seconds with a mean prediction horizon of 3.4 seconds.

Adversarial training for neural networks has been in the limelight in recent years. The advancement in neural network architectures over the last decade has led to significant improvement in their performance. It sparked an interest in their deployment for real-time applications. This process initiated the need to understand the vulnerability of these models to adversarial attacks. It is instrumental in designing models that are robust against adversaries. Recent works have proposed novel techniques to counter the adversaries, most often sacrificing natural accuracy. Most suggest training with an adversarial version of the inputs, constantly moving away from the original distribution. The focus of our work is to use abstract certification to extract a subset of inputs for (hence we call it 'soft') adversarial training. We propose a training framework that can retain natural accuracy without sacrificing robustness in a constrained setting. Our framework specifically targets moderately critical applications which require a reasonable balance between robustness and accuracy. The results testify to the idea of soft adversarial training for the defense against adversarial attacks. At last, we propose the scope of future work for further improvement of this framework.

In an attempt to understand how deep learning models interpret inputs, it has been found that they change their prediction when a carefully optimized imperceptible noise termed adversarial perturbation is added to the input data. Many researchers are focusing on developing methods to counter such effects, but such methods do not generalize well to adversarial test data. Recently, Feature-Scatter adversarial training has come up to solve such a problem, but this method uses the traditional adversarial training framework as its basis that cannot generate diverse perturbations.In this paper, we propose an approach that combines both the Feature-Scatter adversarial training and the generator-based adversarial training framework to optimally explore the adversarial data manifold achieving better robust generalization. We perform extensive experimentation across a wide variety of datasets such as Cifar10, Cifar100, and SVHN. Our framework significantly outperforms the state-of-the-art methods against both strong white-box attacks and black-box attacks.

Spiking Neural Networks (SNNs) are an emerging domain of biologically inspired neural networks that have shown promise for low-power AI. A number of methods exist for building deep SNNs, with Artificial Neural Network (ANN)-to-SNN conversion being highly successful. MaxPooling layers in Convolutional Neural Networks (CNNs) are an integral component to downsample the intermediate feature maps and introduce translational invariance, but the absence of their hardware-friendly spiking equivalents limits such CNNs' conversion to deep SNNs. In this paper, we present two hardware-friendly methods to implement Max-Pooling in deep SNNs, thus facilitating easy conversion of CNNs with MaxPooling layers to SNNs. In a first, we also execute SNNs with spiking-MaxPooling layers on Intel's Loihi neuromorphic hardware (with MNIST, FMNIST, & CIFAR10 dataset); thus, showing the feasibility of our approach.

Deep learning models change their prediction through a carefully optimized imperceptible change in the input termed as an adversarial perturbation. Researchers have been focusing on developing methods to counter such effects. Recently, randomized smoothing a highly scalable technique to develop a certified classifier was introduced. However, this technique involves training a neural network from scratch. In this paper we present an empirical insight into the technique of randomized smoothing and propose a framework for a generalizable defence that works on a novel way of adding gaussian noise to the randomized smoothing procedure and is applicable to black-box pre-trained classifiers. We perform extensive experimentation across a variety of classification models and multiple datasets such as Cifar10 and ImageNet. Our framework is model-agnostic and out-performs the recent state-of-the-art certified defence methods by a large margin without the requirement of any intensive training, thus achieving high certified as well as unperturbed performance.

Hypernymy plays a fundamental role in many AI tasks like taxonomy learning, ontology learning, etc. This has motivated the development of many automatic identification methods for extracting this relation, most of which rely on word distribution. We present a novel model HyperBox to learn box embeddings for hypernym discovery. Given an input term, HyperBox retrieves its suitable hypernym from a target corpus. For this task, we use the dataset published for SemEval 2018 Shared Task on Hypernym Discovery. We compare the performance of our model on two specific domains of knowledge: medical and music. Experimentally, we show that our model outperforms existing methods on the majority of the evaluation metrics. Moreover, our model generalize well over unseen hypernymy pairs using only a small set of training data.

Thermal Comfort Data is critical to generate machine learning models for efficient heating and cooling systems. However, thermal comfort datasets are often highly imbalanced due to subjective user feedback, thus making it challenging to accurately predict both majority and minority classes. This demands the use of data synthesis techniques prior to training classification models to balance the datasets. Commonly used techniques like Synthetic Minority Over-sampling Technique (SMOTE) or Adaptive Synthetic Sampling Method (ADASYN) often compromise testing accuracy and more sophisticated techniques like Conditional Wasserstein Generative Adversarial Network with gradient penalty (cWGAN-GP) are significantly expensive to train. In this paper we propose a novel data augmentation algorithm called Conditional Classifier-Generator (cCGen) to address these two issues. We evaluated the performance of cCGen with real thermal comfort data against SMOTE, ADASYN and cWGAN-GP at different imbalance ratios. Our experiments reveal that our approach can produce better F1 scores than other sampling methods while being more than 10 times faster than cWGAN-GP and not compromising test accuracy.

The identification of B-cell epitopes that elicit an antigen-specific immune response is essential for a variety of immunodetection and immunotherapeutic applications, including the development of safe and high efficacy vaccines. Identifying diagnostically or therapeutically useful epitopes is a difficult, time-consuming, and resource-intensive procedure. In silico prediction of B-cell epitope has gained immense attention in recent years due to its low cost, fast results, and less labor-intensive method compared to NMR spectroscopy and 3D X-ray structural analysis of antibody-antigen complexes. However, one of the major problems that most established models confront is gathering huge volumes of data. Moreover, most models do not achieve high levels of accuracy. The current work is the first to propose the ‘Fuzzy’ approach to in silico B-cell epitope prediction. The effectiveness of the proposed approach is demonstrated on severely imbalanced and limited datasets through several experiments. The results show that using the proposed method enhances both accuracy and precision when compared to existing approaches. Further, the model is tested on the SARS-CoV-1 antigen-antibody PDB complex. The proposed approach outperforms state-of-the-art machine learning (ML) models trained on the same dataset. Results obtained indicate that applying the proposed method improves the prediction compared to the other approaches.

In discrete sequence based Generative Adversarial Networks (GANs), it is important to both land the samples in the initial distribution and drive the generation towards desirable properties. However, in the case of longer molecules, the existing models seem to under-perform in producing new molecules. In this work, we propose the use of Self-Attention mechanism for Generative Adversarial Networks to allow long range dependencies. Self-Attention mechanism has produced improved rewards in novelty and promising results in generating molecules.

Automatic rumor detection is the task of finding rumors on social networks. Previous techniques leveraged the propagation structure of tweets to detect the rumors, which makes the propagation of tweets necessary to detect rumors. However, current text-based works provide sub-optimal results as compared to propagation-based techniques. This work presents a retrieval-based framework that leverages the similar tweets from the given train set and chooses the best model from an ensemble of models to predict the test tweet label. Our proposed framework is based on transformers-based pre-trained models (PTM's). Experiments on two public data sets used in previous works, show that our framework can detect the tweets with equivalent accuracy as propagation-based techniques. The primary advantage of this work is in early rumor detection. The proposed framework can detect rumors in few minutes compared to propagation-based works, which requires a significant amount of propagation of tweets that can take hours before they can be detected.

Specifications for software systems are quite often missing or are obsolete given the evolutionary nature of these systems. Lack of precise software specifications makes the task of debugging and detecting a malfunction of system behavior challenging. Prior works have primarily focused on extracting system specifications in the form of template-based mining frameworks or interactive simulation models. In safety-critical systems where the time of occurrence of events is of prime importance extracting specifications with a quantitative notion of time seems a daunting task. This work presents an unsupervised approach to mine timed temporal properties in the form of deterministic finite state machines with a custom-designed trie data structure. Our frame-work, MINTS learns dominant system specifications from their system traces that are represented as a timed deterministic finite state machine. MINTS is shown to be sound and complete. MINTS scalability and correctness is validated using real-world industry strength traces.

Dynamic analysis of real-world software systems is challenging due to imperfections, noise and data loss. Moreover, these systems evolve with time and their requirements are usually either not clearly specified or unknown, which makes it hard to analyze them. Therefore, it is important to create models that can learn to behave similarly to these systems to enable us to predict their actions, recover missing data, or detect potential failures ahead of time.Several models have been proposed to model sequential data, but the vast majority of them only have a qualitative notion of time or no notion of it at all. In this paper, we extend the work on incorporating a quantitative notion of time to RNN and introduce Time GRU. This modified GRU can learn the behaviour of complex software systems to a very high degree of accuracy. Our approach is scalable and has shown state-of-the-art performance on industry-strength software with real operating logs from Blackberry’s QNX real-time operating system. The proposed model can predict upcoming sequences of events more than 100 timesteps ahead in time with more than 90% accuracy. This allows for significant improvement in trace reconstruction and failure explainability.

Time series data is available from a diverse set of sensors in real life. It is of prime importance in the domain of machine learning and artificial intelligence to analyze such data and identify outliers or anomalies, characteristic of the underlying activities and predict the future. Traditionally, time-series analysis involves identifying features using exploratory data analysis and using statistical approaches for classification and prediction. However, with the advent of convolutional neural networks (CNN), our ability to extract features automatically has substantially improved. In this paper, we propose a novel lightweight deep learning architecture of dilated CNN for classification and predicting time series data sets. We evaluate our model on a real-world human activity recognition time series data set and a synthetically crafted pseudo-realistic dataset for human intent recognition. Our model outperforms the state-of-the-art models and is light-weight.

System traces are a collection of time-stamped messages recorded by the operating system while the system is running. Analysis of these traces is crucial for tasks such as system fault finding. Moreover, detecting anomalies in system behavior becomes crucial in safety-critical and time-sensitive systems where delayed detections can lead to catastrophic outcomes. Therefore, we focus on developing a lightweight and explainable approach for safety-critical time-sensitive systems.Given a set of system traces under normal conditions and anomalous conditions, trace-based anomaly detection aims at classifying the trace as anomalous or not. In this work, we introduce GWAD, a greedy workflow graph framework for anomaly detection, a novel greedy graph construction approach for both offline and online anomaly detection in system traces. Our approach utilizes both sequence of occurrence of events and the time interval between their occurrences in learning the normal system behavior. We propose two approaches, first for offline classification of the trace as anomalous or normal using the event occurrence workflow graphs and secondly an online streaming algorithm that monitors the events as they occur in real-time for detecting anomalies increasing system resilience. Our approach also provides reasoning for the cause of anomalous behavior. We show that GWAD is better than traditional state-of-the-art models. The paper shows the technical feasibility and viability of GWAD through multiple case studies using traces from a field-tested hexacopter.

Time-sensitive safety-critical systems store traces as a collection of time-stamped messages that are generated while a system is operating. Analysis of these traces becomes a key task as it allows one to find faults or errors within a system that is otherwise difficult to discern, especially in complex systems. Furthermore, finding any form of anomalous behaviour becomes critical in time-sensitive and safety-critical systems where a late detection will often lead to dire consequences. Most available approaches are generally used in networking or business process analysis. We focus on creating a lightweight and explainable approach for time-sensitive safety-critical systems. By using a set of system traces under both normal and anomalous conditions, our approach attempts to classify whether or not a trace is anomalous. In this work, we introduce MA2DF, Multi-Agent Anomaly Detection Framework, a novel multi-agent based graph design approach for online and offline anomaly detection in system traces. Our approach takes advantage of the timing information between a sequence of events and also the event sequences to learn and discern between normal and anomalous traces. We present two approaches, an offline approach to discern anomalous behaviour by utilizing the event occurrence workflow graph. The second approach is an online streaming algorithm that monitors the sequence of events as they arrive in real-time. This can be used to detect anomalies, find the cause, and improve system resilience. We show how our approach, MA2DF, is superior to other state-of-the-art models. The paper will explore the technical feasibility and viability of MA2DF by utilizing industry strength case study using traces from a field-tested hexacopter.

Most real-world datasets, and particularly those collected from physical systems, are full of noise, packet loss, and other imperfections. However, most specification mining, anomaly detection and other such algorithms assume, or even require, perfect data quality to function properly. Such algorithms may work in lab conditions when given clean, controlled data, but will fail in the field when given imperfect data. We propose a method for accurately reconstructing discrete temporal or sequential system traces affected by data loss, using Long Short-Term Memory Networks (LSTMs). The model works by learning to predict the next event in a sequence of events, and uses its own output as an input to continue predicting future events. As a result, this method can be used for data restoration even with streamed data. Such a method can reconstruct even long sequence of missing events, and can also help validate and improve data quality for noisy data. The output of the model will be a close reconstruction of the true data, and can be fed to algorithms that rely on clean data. We demonstrate our method by reconstructing automotive CAN traces consisting of long sequences of discrete events. We show that given even small parts of a CAN trace, our LSTM model can predict future events with an accuracy of almost 90%, and can successfully reconstruct large portions of the original trace, greatly outperforming a Markov Model benchmark. We separately feed the original, lossy, and reconstructed traces into a specification mining framework to perform downstream analysis of the effect of our method on state-of-the-art models that use these traces for understanding the behavior of complex systems.

Data generated by real-world systems specifically cyber-physical systems is full of noise, packet loss, and other imperfections. However, most intrusion detection, anomaly de-tection, monitoring, and mining algorithms and frameworks assume that data provided is of perfect quality. Therefore, these algorithms tend to perform extremely well in controlled lab environments but fail in the real-world. We propose a method for accurately restoring discrete tem-poral or sequential system traces affected by data loss, using Word 2vec's Continuous Bag of Words (CBOW) model. The model works by learning to predict the next event in a sequence of events, the model feeds its output back into it for subsequent future predictions. Such a method can reconstruct even long sequence of missing events, and help validate and improve data quality for noisy data. The restored traces are very close to the real-data and can be used by algorithms depending on real-data for system analysis. We demonstrate our method by reconstructing traces from QNX real-time operating system consisting of long sequences of discrete events. We show that given even small parts of a QNX trace, our CBOW model can predict future events with an accuracy of almost 90% outperforming the Markov Model benchmark.

Dynamic evaluation of a computer program is done by analyzing the data generated by it during the execution. The program is considered correct and efficient if it meets a set of pre-specified temporal properties. Temporal properties define the order of occurrence and timing constraints on event occurrence. These properties become all the more important in the case of safety-critical real-time systems where a delayed response to a request may lead to a fault in the system. In the context of real-time systems, it is thus desirable to mine timing information and a set of dominant properties from system execution traces for testing, verification, anomaly detection, and debugging purposes.We propose a framework to automatically mine timing characteristics for properties that are in the form of timed regular expressions (TREs) from system traces. The framework estimates a set of dominant and most frequently occurring properties and their timing characteristics. The framework is evaluated on traces from industrial safety-critical real-time applications (a deployed autonomous hexacopter system) using traces with more than 1 Million entries.

Temporal properties define the order of occurrence and timing constraints on event occurrence. Such specifications are important for safety-critical real-time systems. We propose a framework for automatically mining temporal properties that are in the form of timed regular expressions (TREs) from system traces. Using an abstract structure of the property, the framework constructs a finite state machine to serve as an acceptor. We analytically derive speedup for the fragment and confirm the speedup using empirical validation with synthetic traces. The framework is evaluated on industrial-strength safety-critical real-time applications using traces with more than 1 million entries.

Software specifications are useful for software validation, model checking, runtime verification, debugging, monitoring, etc. In context of safety-critical real-time systems, temporal properties play an important role. However, temporal properties are rarely present due to the complexity and evolutionary nature of software systems. We propose Timed Regular Expression Mining (TREM) a hosted tool for specification mining using timed regular expressions (TREs). It is designed for easy and robust mining of dominant temporal properties. TREM uses an abstract structure of the property; the framework constructs a finite state machine to serve as an acceptor. TREM is scalable, easy to access/use, and platform independent specification mining framework. The tool is tested on industrial strength software system traces such as the QNX real-time operating system using traces with more than 1.5 Million entries. The tool demonstration video can be accessed here: youtu.be/cSd_aj3_LH8.

Short-term electricity demand forecasting is critical to utility companies. It plays a key role in the operation of power industry. It becomes all the more important and critical with increasing penetration of renewable energy sources. Short-term load forecasting enables power companies to make informed business decisions in real-time. Demand patterns are extremely complex due to market deregulation and other environmental factors. Although there has been extensive research in the area of short-term electrical load forecasting, difficulties in implementation and lack of transparency in results has been cited as a main challenge. Deep neural architectures have recently shown their ability to mine complex underlying patterns in various domains. In our work, we present a deep recurrent neural architecture to unearth the complex patterns underlying the regional demand profiles without specific insights from the utilities. The model learns from historical data patterns. We show that deep recurrent neural network with long-short term memory architecture presents a robust methodology for accurate short term load forecasting with the ability to adapt and learn the underlying complex features over time. In most cases it matches the performance of the latest state-of-the-art techniques and even supercedes it in a few cases.

This paper presents an approach to time series prediction based on Asymmetric Subsethood-Product Fuzzy Neural Inference System (ASuPFuNIS). The standard time series techniques have standard averaging where a fixed weight is added to the past values. In this paper we present a novel neuro-fuzzy inference system based on asymmetric subsethood with intervention based transfer function based time series model for accurate prediction of time series. The design of the model is described, and the scheme is evaluated by application to real-world problem of cost of electricity prices over a period of seven year in Ontario, Canada. We also study the various statistical properties of the data.

The Quadratic Knapsack Problem (QKP) deals with maximizing a quadratic objective function subject to given constraints on the capacity of the Knapsack. We assume all coefficients to be non-negative and all variables to be binary. Solution to QKP generalizes the problem of finding whether a graph contains a clique of given size. We propose in this paper a Novel Quantum Evolutionary Algorithm (NQEA) for QKPs. These algorithms are general enough and can be used for similar subsection of problems. We report in this paper solutions which lie in less than 1% of the optimal solutions. We also show that our algorithm is scalable to much larger problem sizes and is capable of exploiting the search space to its maximum.

Recently, the mobile phones are evolving into mobile multifunctional terminals with the incorporation of various useful functions; the display devices also have advanced in terms of size, design, and resolution as a user interface of the mobile phone. Many low power display techniques have been proposed for example dynamic luminance scaling, liquid crystal orientation shift, variable frame refresh, organic LEDs (light emitting diodes) etc. It has been observed that optimizing software alone or hardware alone the solution might not be most optimal. The paper describes the design of a dynamic fuzzy controller based multilayered architecture which is a combination of software and hardware levels to interface between software application layer and physical layer to eventually control the current through LEDs of cell phone and thereby improving battery life depending on user choices and applications running on mobile devices.

In the context of parallel master-slave implementations of evolutionary learning in fuzzy-neural network models, a major issue that arises during runtime is how to balance the load-the number of strings assigned to a slave for evaluation during a generation-in order to achieve maximum speed up. Slave evaluation times can fluctuate drastically depending upon the local computational load on the slave (given fixed node specifications). Communication delays compound the problem of proper load assignment. In this paper we propose the design of a novel dynamic fuzzy load estimator for application to load balancing on heterogeneous LAM/MPI clusters. Using average evaluation time and communication delay feedback estimates from slaves, string assignments for evaluation to slaves are dynamically changed during runtime. Extensive tests on heterogenous clusters shows that considerable speedups can be achieved using the proposed fuzzy controller.