Machine learning and Artificial Intelligence (AI) already support human decision-making and complement professional roles, and are expected in the future to be sufficiently trusted to make autonomous decisions. To trust AI systems with such tasks, a high degree of confidence in their behaviour is needed. However, such systems can make drastically different decisions if the input data is modified, in a way that would be imperceptible to humans. The field of Adversarial Machine Learning studies how this feature could be exploited by an attacker and the countermeasures to defend against them. This work examines the Fast Gradient Signed Method (FGSM) attack, a novel Single Value attack and the Label Flip attack on a trending architecture, namely a 1-Dimensional Convolutional Neural Network model used for time series classification. The results show that the architecture was susceptible to these attacks and that, in their face, the classifier accuracy was significantly impacted.

Threat information sharing is considered as one of the proactive defensive approaches for enhancing the over-all security of trusted partners. Trusted partner organizations can provide access to past and current cybersecurity threats for reducing the risk of a potential cyberattack—the requirements for threat information sharing range from simplistic sharing of documents to threat intelligence sharing. Therefore, the storage and sharing of highly sensitive threat information raises considerable concerns regarding constructing a secure, trusted threat information exchange infrastructure. Establishing a trusted ecosystem for threat sharing will promote the validity, security, anonymity, scalability, latency efficiency, and traceability of the stored information that protects it from unauthorized disclosure. This paper proposes a system that ensures the security principles mentioned above by utilizing a distributed ledger technology that provides secure decentralized operations through smart contracts and provides a privacy-preserving ecosystem for threat information storage and sharing regarding the MITRE ATT&CK framework.

Intrusion detection systems are integral to the security of networked systems for detecting malicious or anomalous network traffic. As traditional approaches are becoming less effective, machine learning and deep learning-based intrusion detection systems are vital research areas for improved detection systems. Past research into computer vision using deep learning revealed that the deep learning-based classifiers themselves are vulnerable to adversarial attacks, and these attacks have been investigated extensively. However, adversarial attacks are restricted not only to the domain of image recognition. As indicated by previous research, various domains employing machine learning/deep learning classifiers are vulnerable to attack. Our work evaluates the effectiveness of adversarial robustness training when applied to intrusion detection systems based on deep learning classification models. We propose a novel, simple adversarial retraining method to build models robust to adversarial evasion attacks.

Phishing is considered to be one of the most prevalent cyber-attacks because of its immense flexibility and alarmingly high success rate. Even with adequate training and high situational awareness, it can still be hard for users to continually be aware of the URL of the website they are visiting. Traditional detection methods rely on blacklists and content analysis, both of which require time-consuming human verification. Thus, there have been attempts focusing on the predictive filtering of such URLs. This study aims to develop a machine-learning model to detect fraudulent URLs and be used within the Splunk platform. Inspired from similar approaches in the literature, we trained the SVM and Random Forests algorithms using malicious and benign datasets found in the literature and one dataset that we created. We evaluated the algorithms' performance with precision and recall reaching up to 85% precision and 87% recall in the case of Random Forests while SVM achieved up to 90% precision and 88% recall using only descriptive features.

We present PyKaldi, a free and open-source Python wrapper for the widely-used Kaldi speech recognition toolkit. PyKaldi is more than a collection of Python bindings into Kaldi libraries. It is an extensible scripting layer that allows users to work with Kaldi and OpenFst types interactively in Python. It tightly integrates Kaldi vector and matrix types with NumPy arrays. We believe Py Kaldi will significantly improve the user experience and simplify the integration of Kaldi into Python workflows. PyKaldi comes with extensive documentation and tests. It is released under the Apache License v2.0 with support for both Python 2.7 and 3.5+.

We propose a semi-supervised learning method to improve classification performance in scenarios with limited labeled data. We employ adaptation strategies such as entropy-filtering and self-training, and show that our method achieves up to 17.2% relative improvement in UAR for a multi-class problem. We apply our method to two different tasks: speaker clustering for adult-child interactions during autism assessment sessions, and a variation of the language identification task (LID). We show that in both tasks our method improves classification accuracy while using lesser training data than the baseline and demonstrate the robustness of our setup to the degree of adaptation by controlling the threshold on uncertainty of classification.

Many speech processing algorithms and applications rely on the explicit knowledge of signal-to-noise ratio (SNR) in their design and implementation. Estimating the SNR of a signal can enhance the performance of such technologies. We propose a novel method for estimating the long-term SNR of speech signals based on features, from which we can approximately detect regions of speech presence in a noisy signal. By measuring the energy in these regions, we create sets of energy ratios, from which we train regression models for different types of noise. If the type of noise that corrupts a signal is known, we use the corresponding regression model to estimate the SNR. When the noise is unknown, we use a deep neural network to find the “closest” regression model to estimate the SNR. Evaluations were done based on the TIMIT speech corpus, using noises from the NOISEX-92 noise database. Furthermore, we performed cross-corpora experiments by training on TIMIT and NOISEX-92 and testing on the Wall Street Journal speech corpus and DEMAND noise database. Our results show that our system provides accurate SNR estimations across different noise types, corpora, and that it outperforms other SNR estimation methods.

Automatically evaluating pronunciation quality of non-native speech has seen tremendous success in both research and com-mercial settings, with applications in L2 learning. In this paper, submitted for the INTERSPEECH 2015 Degree of Nativeness Sub-Challenge, this problem is posed under a challenging cross-corpora setting using speech data drawn from multiple speakers from a variety of language backgrounds (L1) reading different English sentences. Since the perception of non-nativeness is re-alized at the segmental and suprasegmental linguistic levels, we explore a number of acoustic cues at multiple time scales. We experiment with both data-driven and knowledge-inspired fea-tures that capture degree of nativeness from pauses in speech, speaking rate, rhythm/stress, and goodness of phone pronunci-ation. One promising finding is that highly accurate automated assessment can be attained using a small diverse set of intuitive and interpretable features. Performance is further boosted by smoothing scores across utterances from the same speaker; our best system significantly outperforms the challenge baseline.

This paper introduces a supervised statistical framework for estimating the signal-to-noise (SNR) ratio of speech signals. Information on how noise corrupts a signal can help us compensate for its effects, especially in real life applications where the usual assumption of white Gaussian noise does not hold and speech boundaries in the signal are not known. We use features from which we can detect speech regions in a signal, without using Voice Activity Detection, and estimate the energies of those regions. Then we use these features to train ordinary least squares regression models for various noise types. We compare this supervised method with state-of-the-art SNR estimation algorithms and show its superior performance with respect to the tested noise types.

Least-squares estimation has always been the main approach when applying prediction error methods (PEM) in the identification of linear dynamical systems. Regardless of the estimation algorithm, if there are no restrictions on the form of the matrices we want to estimate, the matrices can be determined up to within a linear transformation and thus the result may be different than the true solution and the convergence of iterative algorithms may be affected. In this paper, we apply a new identification procedure based on the Expectation Maximization framework to a family of identifiable state-space models. To our knowledge, this is the first complete solution of Maximum-Likelihood estimation for general linear state-space models.