The spreading of AI-generated images (AIGI), driven by advances in generative AI, poses a significant threat to in- formation security and public trust. Existing AIGI detectors, while effective against images in clean laboratory settings, fail to generalize to in-the-wild scenarios. These real-world images are noisy, varying from “obviously fake” images to realistic ones derived from multiple generative models and further edited for quality control. We address in-the-wild AIGI detection in this paper. We introduce MIRAGE, a challenging benchmark designed to emulate the complexity of in-the-wild AIGI. MIRAGE is constructed from two sources: (1) a large corpus of Internet-sourced AIGI verified by human experts, and (2) a synthesized dataset created through the collaboration between multiple expert generators, closely simulating the realistic AIGI in the wild. Building on this benchmark, we propose MIRAGE-R1, a vision- language model with heuristic-to-analytic reasoning, a reflective reasoning mechanism for AIGI detection. MIRAGE-R1 is trained in two stages: a supervised-fine-tuning cold start, followed by a reinforcement learning stage. By further adopting a inference-time adaptive thinking strategy, MIRAGE-R1 is able to provide either a quick judgment or a more robust and accurate conclusion, effectively balancing inference speed and performance. Extensive experiments show that our model leads state-of-the-art detectors by 5% and 10% on MIRAGE and public benchmark, respectively.

Existing software reliability models (SRMs) that describe software fault detection during the testing phase can be unified under the framework of self-exciting point processes. However, it remains unclear whether such generalized self-excitation models can outperform traditional non-homogeneous Poisson process (NHPP) and non-homogeneous Markov process (NHMP)-based SRMs in terms of goodness-of-fit and predictive capabilities. In this paper, we propose a family of Hawkes process (HKP)-based SRMs that incorporate time-varying base intensity functions, distinguishing them from conventional HKPs with constant base intensity. Specifically, we introduce a Pareto-type base intensity and explore ten different impact (kernel) functions in the stochastic intensity part, and then compare the performance of the proposed HKP-based models with that of representative NHPP and NHMP-based SRMs. Experimental results on eight software development project datasets demonstrate that the HKP-based models with self-excitation generally achieve superior goodness-of-fit and predictive performance.

Although a huge number of software reliability models (SRMs) have been proposed in the past literature, there is no unique SRM with satisfactory prediction accuracy, because it has been known that the best goodness-of-fit SRM to the underlying data is not always equivalent to the best prediction model for the future software fault-count process. It is common to select an appropriate SRM with highest goodness-of-fit to the underlying software fault-count data based on any information criteria such as Akaike information criterion (AIC). In this paper we focus on a prediction SRM consisting with linearly weighted combinational (LWC) non-homogeneous Poisson process (NHPP)-based SRMs, and overview an AdaBoosting-based approach by Li et al. (2012) to determine the optimal weights for the LWC NHPP-based SRMs. Next, we propose a refined AdaBoosting technique to keep the steps of original AdaBoosting, in order to present the original predictive performance of AdaBoosting in software reliability.

Temporal action localization (TAL), which involves recognizing and locating action instances, is a challenging task in video understanding. Most existing approaches directly predict action classes and regress offsets to boundaries, while overlooking the discrepant importance of each frame. In this paper, we propose an Action Sensitivity Learning framework (ASL) to tackle this task, which aims to assess the value of each frame and then leverage the generated action sensitivity to recalibrate the training procedure. We first introduce a lightweight Action Sensitivity Evaluator to learn the action sensitivity at the class level and instance level, respectively. The outputs of the two branches are combined to reweight the gradient of the two sub-tasks. Moreover, based on the action sensitivity of each frame, we design an Action Sensitive Contrastive Loss to enhance features, where the action-aware frames are sampled as positive pairs to push away the action-irrelevant frames. The extensive studies on various action localization benchmarks (i.e., MultiThumos, Charades, Ego4D-Moment Queries v1.0, Epic-Kitchens 100, Thumos14 and Activi-tyNet1.3) show that ASL surpasses the state-of-the-art in terms of average-mAP under multiple types of scenarios, e.g., single-labeled, densely-labeled and egocentric.

In this paper, we discuss the hierarchical modeling in model-based dependability evaluation. The hierarchical modeling is a modeling approach that combines non-state-space models such as fault trees and state-space models such as Markov chains. It can mitigate the state explosion problem for state-space models. The fundamental idea is that state-space models are used to represent the dynamic behavior for basic events of non-state-space model. This enables us to represent the dynamic behavior of the system. On the computation of dependability indices such as system reliability, we utilize the non-state-space model. This structure reduces the number of states to be evaluated. The existing hierarchical approaches deal with binary-type reliability models, focusing only on the working and failure of the system/components. In this paper, we extend the existing hierarchical approach to handle the computation of dependability indices using the multi-state system instead of a fault tree, thereby accommodating both reliability and certain performance models, rather than being limited to binary-type models. Additionally, we introduce the computation method with MtMDD (multi-terminal multi-valued decision diagram). Numerical results demonstrated that the proposed hierarchical modeling can efficiently describe multi-state systems and effectively obtain desired dependability indices. It outperforms existing approaches such as Kronecker representation.

Intrusion tolerance is an ability to keep the correct service by masking the intrusion based on fault-tolerant techniques. With the rapid development of virtualization, the virtual machine (VM)-based intrusion tolerance scheme has been developed according to the concept of state machine replication with Byzantine fault tolerant technique. In this article, we present the quantitative security evaluation of the VM-based intrusion tolerant system with the time to security failure. We assume that the arrival stream follows a Markovian arrival process (MAP), which is one of the most general stochastic processes, and analytically derive the Laplace-Stieltjes transform of time to security failure based on the analysis of the MAP/G/1/∞ queue.

This paper presents the computation method for the dependability measure in fault trees when the uncertainty of model parameters is considered. The propagation of uncertainty of model parameters can be estimated by regarding the model parameters as random variables. However, the computation cost of expected values is now so low in such situation. In this paper, we focus on the moment-based approximation method for uncertainty propagation in fault trees. In particular, to obtain the moment-based approximation, we discuss the computation of first and second derivatives of dependability measures in fault trees with BDD (binary decision diagram) representation.

In this paper, we consider a stochastic model to evaluate the system availability of an intrusion tolerant system (ITS), where the system undergoes the patch management with a periodic vulnerability checking strategy, i.e., a pull-type patch management. Based on the model, this paper discusses the appropriate timing for patch applying. In particular, the paper models the attack behavior of adversary and the system behaviors under reactive defense strategies by a composite stochastic reward net (SRN). Furthermore, we formulate the interval availability by applying the phase-type (PH) approximation to solve the Markov regenerative process (MRGP) models derived from the SRNs. Numerical experiments are conducted to study the sensitivity of the system availability with respect to the number of checking.

This paper considers the statistical approach to model software degradation process from time series data of system attributes. We first develop the continuous-time Markov chain (CTMC) model to represent the degradation level of system. By combining the CTMC with system attributes distributions, a continuous-time hidden Markov model (CT-HMM) is proposed as the basic model to represent the degradation level of system. To estimate model parameters, we develop the EM algorithm for CT-HMM. The advantage of this modeling is that the estimated model is directly applied to existing CTMC-based software aging and rejuvenation models. In numerical experiments, we exhibit the performance of our method by simulated data and also demonstrate estimating the software degradation process with experimental data in MySQL database system.

Software rejuvenation is one of the proactive fault management techniques to prevent system performance degradation, which may lead to the system failure caused by software aging. In the design of software rejuvenation, it is important to determine the optimal timing of triggering the rejuvenation in terms of the system overhead. In this paper, we consider six software rejuvenation policies, which are categorized into time-based and workload-based policies, under the environment where the arrival stream of system follows a Markovian arrival process (MAP). After building the stochastic models with respective rejuvenation policies, we formulate the loss probability of transaction and the upper bound of mean response time as the system performance indices. In the numerical illustrations, we exhibit a comprehensive study to compare six software rejuvenation policies numerically and show that the proposed rejuvenation policies called wait-time policies are superior to the others under the MAP arrival stream.

Component importance analysis is to measure the effect on system reliability of component reliabilities, and it can be used to the design of system from the reliability point of view. In this paper, we consider the component importance analysis of real-time computing systems in the presence of common-cause failures (CCFs) (i.e., failure dependencies). Although the CCFs are known as a risk factor of degradation of system reliability, it is difficult to evaluate the component importance measures in the presence of CCFs analytically. This paper introduces a continuous-time Markov chain (CTMC) model for real-time computing system, and applies the CTMC-based component-wise sensitivity analysis which can evaluate the component importance measures without any structure function of system. Also, in numerical experiments, we evaluate the effect of CCFs by the comparison of system performance measures and component importance in the case of system with CCFs with those in the case that there is no CCF in the system.