It is often desirable to be able recognize when inputs a recognition function learned in supervised manner correspond classes unseen at training time. With this ability, new class labels could assigned these by human operator, allowing them incorporated into the --- ideally under an efficient incremental update mechanism. While good algorithms that assume from fixed set of exist, e.g., artificial neural networks and kernel machines, it not immediately obvious how extend perform learning...

State-of-the-art deep neural networks suffer from a fundamental problem - they misclassify adversarial examples formed by applying small perturbations to inputs. In this paper, we present new psychometric perceptual similarity score (PASS) measure for quantifying images, introduce the notion of hard positive generation, and use diverse set not just closest ones data augmentation. We novel hot/cold approach example which provides multiple possible every single image. The generated our often...

As our professional, social, and financial existences become increasingly digitized as government, healthcare, military infrastructures rely more on computer technologies, they present larger lucrative targets for malware. Stealth malware in particular poses an increased threat because it is specifically designed to evade detection mechanisms, spreading dormant, the wild extended periods of time, gathering sensitive information or positioning itself a high-impact zero-day attack. Policing...

Much research has been conducted on both face identification and verification, with greater focus the latter. Research mostly focused using closed-set protocols, which assume that all probe images used in evaluation contain identities of subjects are enrolled gallery. Real systems, however, where only a fraction sample gallery, cannot make this assumption. Instead, they must an open set samples be able to reject/ignore those correspond unknown identities. In paper, we address widespread...

Typically, most network intrusion detection systems use supervised learning techniques to identify anomalies. A problem exists when identifying the unknowns and automatically updating a classifier with new query classes. This is defined as an open set incremental we propose extend recently introduced method, Extreme Value Machine (EVM) address issue of classes during time. The EVM derived from statistical extreme value theory first that can perform kernel-free, nonlinear, variable bandwidth...

Facial attributes are emerging soft biometrics that have the potential to reject non-matches, for example, based on mismatching gender. To be usable in stand-alone systems, facial must extracted from images automatically and reliably. In this paper, we propose a simple yet effective solution automatic attribute extraction by training deep convolutional neural network (DCNN) each separately, without using any pre-training or dataset augmentation, obtain new state-of-the-art classification...

Email attachments are a growing delivery vector for malware. While machine learning (ML) has been successfully applied to portable executable (PE) malware detection, we ask, can extend static ML approaches detect across common email attachment file types, e.g., office documents and Zip archives? To this end, collected dataset of over 5 million malicious/benign Microsoft Office along with smaller data set, which use provide more realistic estimates thresholds false positive rates on...

Confidently distinguishing a malicious intrusion over network is an important challenge. Most detection system evaluations have been performed in closed set protocol which only classes seen during training are considered classification. Thus far, there has no realistic application novel types of behaviors unseen at - unknown as it were must be recognized for manual categorization. This paper comparatively evaluates malware classification using both and open protocols recognition on the...

When implementing real-world computer vision systems, researchers can use mid-level representations as a tool to adjust the trade-off between accuracy and efficiency. Unfortunately, existing that improve tend decrease efficiency, or are specifically tailored work well within one pipeline problem at exclusion of others. We introduce novel, efficient representation improves classification efficiency without sacrificing accuracy. Our Exemplar Codes based on linear classifiers probability...

In recent years, traditional cybersecurity safeguards have proven ineffective against insider threats. Famous cases of sensitive information leaks caused by insiders, including the WikiLeaks release diplomatic cables and Edward Snowden incident, greatly harmed U.S. government's relationship with other governments its own citizens. Data Leak Prevention (DLP) is a solution for detecting preventing from within an organization's network. However, state-of-art DLP detection models are only able...

For applications such as airport border control, biometric technologies that can process many capture subjects quickly, efficiently, with weak supervision, and minimal discomfort are desirable. Facial recognition is particularly appealing because it minimally invasive yet offers relatively good performance. Unfortunately, the combination of supervision invasiveness makes even highly accurate facial systems susceptible to spoofing via presentation attacks. Thus, there great demand for an...

Many security related big data problems, including document, traffic, and system log analysis require of unstructured text. Consider the task analyzing company documents for secure storage. Some might be too sensitive to put on a public cloud private storage with associated backup overhead, some may safe in encrypted form, sufficiently non-sensitive stored plain-text without encryption decryption overhead. Being able make such categorizations autonomously can significantly strengthen...

The U.S Government has been the target for cyberattacks from all over world. Just recently, former President Obama accused Russian government of leaking emails to Wikileaks and declared that U.S. might be forced respond. While Russia denied involvement, it is clear take some defensive measures protect its data infrastructure. Insider threats have cause other sensitive information leaks too, including infamous Edward Snowden incident. Most recent were in form text. Due nature text data,...

When the cost of misclassifying a sample is high, it useful to have an accurate estimate uncertainty in prediction for that sample. There are also multiple types which best estimated different ways, example, intrinsic training set may be well-handled by Bayesian approach, while introduced shifts between and query distributions better-addressed density/support estimation. In this paper, we examine three uncertainty: model capacity uncertainty, data open review techniques been derived address...

Architectures relying on continuous authentication require a secure way to challenge the user's identity without trusting that Continuous Authentication Subsystem (CAS) has not been compromised, i.e., response layer which manages service/application access is fake. In this paper, we introduce CALIPER protocol, in separate Access Verification Entity (CAVE) directly challenges regime. Instead of simply returning probabilities or confidence scores, CALIPER's CAS uses live hard and soft...