Abstract A review was conducted to identify possible applications of artificial intelligence and related technologies in the perpetration crime. The collected examples were used devise an approximate taxonomy criminal for purpose assessing their relative threat levels. exercise culminated a 2-day workshop on ‘AI & Future Crime’ with representatives from academia, police, defence, government private sector. remit (i) catalogue potential terror threats arising increasing adoption power...

The role of anomaly detection in X-ray security imaging, as a supplement to targeted threat detection, is described, and taxonomy types this domain presented. Algorithms are described for detecting appearance anomalies shape, texture, density, semantic object category presence. detected on the basis representations extracted from convolutional neural network pre-trained identify categories photographs, final pooling layer anomalies, logit anomalies. distribution normal data modeled using...

Human-centric computer vision (HCCV) data curation practices often neglect privacy and bias concerns, leading to dataset retractions unfair models. HCCV datasets constructed through nonconsensual web scraping lack crucial metadata for comprehensive fairness robustness evaluations. Current remedies are post hoc, persuasive justification adoption, or fail provide proper contextualization appropriate application. Our research focuses on proactive, domain-specific recommendations, covering...

Speech datasets are crucial for training Language Technologies (SLT); however, the lack of diversity underlying data can lead to serious limitations in building equitable and robust SLT products, especially along dimensions language, accent, dialect, variety, speech impairment - intersectionality features with socioeconomic demographic features. Furthermore, there is often a oversight on commonly built massive web-crawling and/or publicly available regard ethics such collection. To encourage...

Existing approaches to automated security image analysis focus on the detection of particular classes threat. However, this mode inspection is ineffectual when dealing with mature threat, for which adversaries have refined effective concealment techniques. Furthermore, these methods may be unable detect potential threats that never been seen before. Therefore, in paper, we investigate an anomaly framework, at X-ray patch-level, based on: (i) representations, and (ii) anomalies relative those...

Machine learning (ML) datasets, often perceived as neutral, inherently encapsulate abstract and disputed social constructs. Dataset curators frequently employ value-laden terms such diversity, bias, quality to characterize datasets. Despite their prevalence, these lack clear definitions validation. Our research explores the implications of this issue by analyzing "diversity" across 135 image text Drawing from sciences, we apply principles measurement theory identify considerations offer...

Facial verification systems are vulnerable to poisoning attacks that make use of multiple-identity images (MIIs)---face stored in a database resemble multiple persons, such novel any the constituent persons verified as matching identity MII. Research on this mode attack has focused defence by detection, with no explanation why vulnerability exists. New quantitative results presented support an terms geometry representations spaces used systems. In spherical those spaces, angular distance...

Cloud computing is a vital part of any little or huge organization. With cloud storage service clients can remotely store their information to the and realize sharing others. Information redistributing has danger delicate getting breached. Remote trustworthiness auditing proposed ensure integrity data stored in cloud. In some regular frameworks, for example, Electronic Health Records (EHR) document may contain which must not be modified. This paper proposes novel privacy preserving system...

As computer vision systems become more widely deployed, there is increasing concern from both the research community and public that these are not only reproducing but amplifying harmful social biases. The phenomenon of bias amplification, which focus this work, refers to models inherent training set biases at test time. Existing metrics measure amplification with respect single annotated attributes (e.g., $\texttt{computer}$). However, several visual datasets consist images multiple...

Designing models that are robust to small adversarial perturbations of their inputs has proven remarkably difficult. In this work we show the reverse problem---making more vulnerable---is surprisingly easy. After presenting some proofs concept on MNIST, introduce a generic tilting attack injects vulnerabilities into linear layers pre-trained networks by increasing sensitivity components low variance in training data without affecting performance test data. We illustrate multilayer perceptron...

Donations to charity-based crowdfunding environments have been on the rise in last few years. Unsurprisingly, deception and fraud such platforms also increased, but not thoroughly studied understand what characteristics can expose behavior allow its automatic detection blocking. Indeed, are only ones typically performing oversight for campaigns launched each service. However, they properly incentivized combat among users launch: one hand, a platform's revenue is directly proportional number...

Donations to charity-based crowdfunding environments have been on the rise in last few years. Unsurprisingly, deception and fraud such platforms also increased, but not thoroughly studied understand what characteristics can expose behavior allow its automatic detection blocking. Indeed, are only ones typically performing oversight for campaigns launched each service. However, they properly incentivized combat among users launch: one hand, a platform's revenue is directly proportional number...

Despite extensive efforts to create fairer machine learning (ML) datasets, there remains a limited understanding of the practical aspects dataset curation. Drawing from interviews with 30 ML curators, we present comprehensive taxonomy challenges and trade-offs encountered throughout curation lifecycle. Our findings underscore overarching issues within broader fairness landscape that impact data We conclude recommendations aimed at fostering systemic changes better facilitate fair practices.

Vision-language models (VLMs) pre-trained on extensive datasets can inadvertently learn biases by correlating gender information with specific objects or scenarios. Current methods, which focus modifying inputs and monitoring changes in the model's output probability scores, often struggle to comprehensively understand bias from perspective of model components. We propose a framework that incorporates causal mediation analysis measure map pathways generation propagation within VLMs. This...

Deep neural networks trained via empirical risk minimisation often exhibit significant performance disparities across groups, particularly when group and task labels are spuriously correlated (e.g., "grassy background" "cows"). Existing bias mitigation methods that aim to address this issue either rely on for training or validation, require an extensive hyperparameter search. Such data computational requirements hinder the practical deployment of these methods, especially datasets too large...

In this paper, we survey different authentication schemes for graphical passwords utilized in online services. Click-based password scheme is used to garner click-points or pixel-points from users and portend the hotspots. CAPTCHA gives protection against spyware attacks. case of Face DCAPTCHA scheme, must recognize visually-distorted human faces complex images an accurate manner. A Password Guessing Resistant Protocol (PGRP) can limit large number login attempts unknown remote hosts resist...

Technical paper presented at the 2016 Defence and Security Doctoral Symposium.Non-intrusive inspection systems are increasingly used to scan intermodal freight shipping containers, national ports, ensure cargo conformity with customs regulations. Initially, each container is risk assessed based on information such as origin, destination, manifest. If deemed sufficiently high imaged, typically by non-intrusive X-ray radiography. Finally, basis of image, a human operator must make shrewd...

Cloud computing is associate inclusive new approach on however services square measure made and utilized. accomplishment of assorted styles that has attracted several users in today’s state affairs. The foremost enticing service cloud information outsourcing, because this the homeowners will host any size server access from once needed. A dynamic outsourced auditing theme cannot solely defend against dishonest entity collision, conjointly support verifiable updates to information. epitome...

Summary Security is one of the important challenges faced by cloud computing. There are so many methods like encryption, firewall, and providing access control to achieve security. These alone cannot provide a flawless available resources. We in need paradigm that permits only authenticated users resources from cloud. The objective proposed strong authentication. Our method considers appropriate parameters plenty parameters, which used prove user be an person. selection required achieved...