Biblio

Intent classification is an important and relevant area of research in artificial intelligence and machine learning, with applications ranging from marketing and product design to intelligent communication. This paper explores the performance of various models and techniques for short text intent classification in the context of chatbots. The problem was explored for use within the mental wellness and therapy chatbot application, Wysa, to give improved responses to free-text user input. The authors looked at classifying text samples in-to 4 categories - assertions, refutations, clarifiers and transitions. For this, the suitability of the following techniques was evaluated: count vectors, TF-IDF, sentence embeddings and n-grams, as well as modifications of the same. Each technique was used to train a number of state-of-the-art classifiers, and the results have been compiled and presented. This is the first documented implementation of Arora's modification to sentence embeddings for real world use. It also introduces a technique to generate custom stop words that gave a significant gain in performance (10 percentage points). The best pipeline, using these techniques together, gave an accuracy of 95 percent.

Neural response generation is to generate human-like response given human utterance by using a deep learning. In the previous studies, expressing emotion in response generation improve user performance, user engagement, and user satisfaction. Also, the conversational agents can communicate with users at the human level. However, the previous emotional response generation model cannot understand the subtle part of emotions, because this model use the desired emotion of response as a token form. Moreover, this model is difficult to generate natural responses related to input utterance at the content level, since the information of input utterance can be biased to the emotion token. To overcome these limitations, we propose an emotional response generation model which generates emotional and natural responses by using the emotion feature extraction. Our model consists of two parts: Extraction part and Generation part. The extraction part is to extract the emotion of input utterance as a vector form by using the pre-trained LSTM based classification model. The generation part is to generate an emotional and natural response to the input utterance by reflecting the emotion vector from the extraction part and the thought vector from the encoder. We evaluate our model on the emotion-labeled dialogue dataset: DailyDialog. We evaluate our model on quantitative analysis and qualitative analysis: emotion classification; response generation modeling; comparative study. In general, experiments show that the proposed model can generate emotional and natural responses.

The Controller Area Network (CAN) bus system works inside connected cars as a central system for communication between electronic control units (ECUs). Despite its central importance, the CAN does not support an authentication mechanism, i.e., CAN messages are broadcast without basic security features. As a result, it is easy for attackers to launch attacks at the CAN bus network system. Attackers can compromise the CAN bus system in several ways: denial of service, fuzzing, spoofing, etc. It is imperative to devise methodologies to protect modern cars against the aforementioned attacks. In this paper, we propose a Long Short-Term Memory (LSTM)-based Intrusion Detection System (IDS) to detect and mitigate the CAN bus network attacks. We first inject attacks at the CAN bus system in a car that we have at our disposal to generate the attack dataset, which we use to test and train our model. Our results demonstrate that our classifier is efficient in detecting the CAN attacks. We achieved a detection accuracy of 99.9949%.

The automotive Controller Area Network (CAN) allows Electronic Control Units (ECUs) to communicate with each other and control various vehicular functions such as engine and braking control. Consequently CAN and ECUs are high priority targets for hackers. As CAN implementation details are held as proprietary information by vehicle manufacturers, it can be challenging to decode and correlate CAN messages to specific vehicle operations. To understand the precise meanings of CAN messages, reverse engineering techniques that are time-consuming, manually intensive, and require a physical vehicle are typically used. This work aims to address the process of reverse engineering CAN messages for their functionality by creating a machine learning classifier that analyzes messages and determines their relationship to other messages and vehicular functions. Our work examines CAN traffic of different vehicles and standards to show that it can be applied to a wide arrangement of vehicles. The results show that the function of CAN messages can be determined without the need to manually reverse engineer a physical vehicle.

Parallel with the developing world, transportation technologies have started to expand and change significantly year by year. This change brings with it some inevitable problems. Increasing human population and growing transportation-needs result many accidents in urban and rural areas, and this recursively results extra traffic problems and fuel consumption. It is obvious that the issues brought by this spiral loop needed to be solved with the use of some new technological achievements. In this context, self-driving cars or automated vehicles concepts are seen as a good solution. However, this also brings some additional problems with it. Currently many cars are provided with some digital security systems, which are examined in two phases, internal and external. These systems are constructed in the car by using some type of embedded system (such as the Controller Area Network (CAN)) which are needed to be protected form outsider cyberattacks. These attack can be detected by several ways such as rule based system, anomaly based systems, list based systems, etc. The current literature showed that researchers focused on the use of some artificial intelligence techniques for the detection of this type of attack. In this study, an intrusion detection system based on machine learning is proposed for the CAN security, which is the in-vehicle communication structure. As a result of the study, it has been observed that the decision tree-based ensemble learning models results the best performance in the tested models. Additionally, all models have a very good accuracy levels.

Science DMZs are specialized networks that enable large-scale distributed scientific research, providing efficient and guaranteed performance while transferring large amounts of data at high rates. The high-speed performance of a Science DMZ is made viable via data transfer nodes (DTNs), therefore they are a critical point of failure. DTNs are usually monitored with network intrusion detection systems (NIDS). However, NIDS do not consider system performance data, such as network I/O interrupts and context switches, which can also be useful in revealing anomalous system performance potentially arising due to external network based attacks or insider attacks. In this paper, we demonstrate how system performance metrics can be applied towards securing a DTN in a Science DMZ network. Specifically, we evaluate the effectiveness of system performance data in detecting TCP-SYN flood attacks on a DTN using DBSCAN (a density-based clustering algorithm) for anomaly detection. Our results demonstrate that system interrupts and context switches can be used to successfully detect TCP-SYN floods, suggesting that system performance data could be effective in detecting a variety of attacks not easily detected through network monitoring alone.

Cloud computing is applied in various domains, among which social recommender systems are well-received because of their effectivity to provide suggestions for users. Social recommender systems perform well in alleviating cold start problem, but it suffers from shilling attack due to its natural openness. Shilling attack is an injection attack mainly acting on the training process of machine learning, which aims to advance or suppress the recommendation ranking of target items. Some researchers have studied the influence of shilling attacks in two perspectives simultaneously, which are user-item's rating and user-user's relation. However, they take more consideration into user-item's rating, and up to now, the construction of user-user's relation has not been explored in depth. To explore shilling attacks with complex relations, in this paper, we propose two novel attack models based on triangle relations in social networks. Furthermore, we explore the influence of these models on five social recommendation algorithms. The experimental results on three datasets show that the recommendation can be affected by the triangle relation attacks. The attack model combined with triangle relation has a better attack effect than the model only based on rating injection and the model combined with random relation. Besides, we compare the functions of triangle relations in friend recommendation and product recommendation.

In this paper, we are going to verify the possibility to create a ransomware simulation that will use an arbitrary combination of known tactics and techniques to bypass an anti-malware defense. To verify this hypothesis, we conducted an experiment in which an agent was trained with the help of reinforcement learning to run the ransomware simulator in a way that can bypass anti-ransomware solution and encrypt the target files. The novelty of the proposed method lies in applying reinforcement learning to anti-ransomware testing that may help to identify weaknesses in the anti-ransomware defense and fix them before a real attack happens.

The solution of using existing WiFi devices for measurement and maintenance, and establishing a WiFi fingerprint database for precise localization has become a popular method for indoor localization. The traditional WiFi fingerprint privacy protection scheme increases the calculation amount of the client, but cannot completely protect the security of the client and the fingerprint database. In this paper, we make use of WiFi devices to present a Practical Privacy Protection Scheme In WiFi Fingerprint-based Localization PPWFL. In PPWFL, the localization server establishes a pre-partition in the fingerprint database through the E-M clustering algorithm, we divide the entire fingerprint database into several partitions. The server uses WiFi fingerprint entries with partitions as training data and trains a machine learning model. This model can accurately predict the client's partition based on fingerprint entries. The client uses the trained machine learning model to obtain its partition location accurately, picks up WiFi fingerprint entries in its partition, and calculates its geographic location with the localization server through secure multi-party computing. Compared with the traditional solution, our solution only uses the WiFi fingerprint entries in the client's partition rather than the entire fingerprint database. PPWFL can reduce not only unnecessary calculations but also avoid accidental errors (Unexpected errors in fingerprint similarity between non-adjacent locations due to multipath effects of electromagnetic waves during the propagation of complex indoor environments) in fingerprint distance calculation. In particular, due to the use of Secure Multi-Party Computation, most of the calculations are performed in the local offline phase, the client only exchanges data with the localization server during the distance calculation phase. No additional equipment is needed; our solution uses only existing WiFi devices in the building to achieve fast localization based on privacy protection. We prove that PPWFL is secure under the honest but curious attacker. Experiments show that PPWFL achieves efficiency and accuracy than the traditional WiFi fingerprint localization scheme.

The following topics are dealt with: authorisation; data privacy; mobile computing; security of data; cryptography; Internet of Things; message authentication; invasive software; Android (operating system); vectors.

Keystroke Inference has been a hot topic since it poses a severe threat to our privacy from typing. Existing learning-based Keystroke Inference suffers the domain adaptation problem because the training data (from attacker) and the test data (from victim) are generally collected in different environments. Recently, Optimal Transport (OT) is applied to address this problem, but suffers the “ground metric” limitation. In this work, we propose a novel method, OTDA, by incorporating Discriminant Analysis into OT through an iterative learning process to address the ground metric limitation. By embedding OTDA into a vibration-based Keystroke Inference platform, we conduct extensive studies about domain adaptation with different factors, such as people, keyboard position, etc.. Our experiment results show that OTDA can achieve significant performance improvement on classification accuracy, i.e., outperforming baseline by 15% to 30%, state-of-the-art OT and other domain adaptation methods by 10% to 20%.

In the present scenario, security is the biggest concern in any domain of applications. The latest and widely used system for user authentication is a biometric system. This includes fingerprint recognition, retina recognition, and voice recognition. But these systems can be bypassed by masqueraders. To avoid this, a combination of these systems is used which becomes very costly. To overcome these two drawbacks keystroke dynamics were introduced in this field. Keystroke dynamics is a biometric authentication-based system on behavior, which is an automated method in which the identity of an individual is identified and confirmed based on the way and the rhythm of passwords typed on a keyboard by the individual. The work in this paper focuses on identifying the best algorithm for implementing an authentication system with the help of machine learning for user identification based on keystroke dynamics. Our proposed model which uses XGBoost gives a comparatively higher accuracy of 93.59% than the other algorithms for the dataset used.

The rapid growth of IPv6 traffic makes security issues become more important. This paper proposes an IPv6 network security system that integrates signature-based Intrusion Detection Systems (IDS) and machine learning classification technologies to improve the accuracy of IPv6 denial-of-service (DoS) attacks detection. In addition, this paper has also enhanced IPv6 network security defense capabilities through software-defined networking (SDN) and network function virtualization (NFV) technologies. The experimental results prove that the detection and defense mechanisms proposed in this paper can effectively strengthen IPv6 network security.

With the development of technology, application of the Internet in daily life is increasing, making our connection with the Internet closer. However, with the improvement of convenience, information security has become more and more important. How to ensure information security in a convenient living environment is a question worth discussing. For instance, the widespread deployment of IP-cameras has made great progress in terms of convenience. On the contrary, it increases the risk of privacy exposure. Poorly designed surveillance devices may be implanted with suspicious software, which might be a thorny issue to human life. To effectively identify vulnerable devices, we design an SDN-based identification system that uses machine learning technology to identify brands and probable model types by identifying packet features. The identifying results make it possible for further vulnerability analysis.

In this paper, we propose a novel deep learning and blockchain-based energy framework for smart grids, entitled DeepCoin. The DeepCoin framework uses two schemes, a blockchain-based scheme and a deep learning-based scheme. The blockchain-based scheme consists of five phases: setup phase, agreement phase, creating a block phase and consensus-making phase, and view change phase. It incorporates a novel reliable peer-to-peer energy system that is based on the practical Byzantine fault tolerance algorithm and it achieves high throughput. In order to prevent smart grid attacks, the proposed framework makes the generation of blocks using short signatures and hash functions. The proposed deep learning-based scheme is an intrusion detection system (IDS), which employs recurrent neural networks for detecting network attacks and fraudulent transactions in the blockchain-based energy network. We study the performance of the proposed IDS on three different sources the CICIDS2017 dataset, a power system dataset, and a web robot (Bot)-Internet of Things (IoT) dataset.

This paper introduces a technique that enhances the capabilities of an intrusion detection system (IDS) in a wireless body area network (WBAN). This technique involves adopting two known machine-learning algorithms: artificial neural network (ANN) and the J48 form of decision trees. The enhanced technique reduces the security threats to a WBAN, such as denial-of-service (DoS) attacks. It is essential to manage noise, which might affect the data gathered by the sensors. In this paper, noise in data is measured because it can affect the accuracy of the machine learning algorithms and demonstrate the level of noise at which the machine-learning model can be trusted. The results show that J48 is the best model when there is no noise, with an accuracy reaching 99.66%, as compared to the ANN algorithm. However, with noisy datasets, ANN shows more tolerance to noise.

MIL-STD-1553 is a military standard that defines the specification of a serial communication bus that has been implemented in military and aerospace avionic platforms for over 40 years. MIL-STD-1553 was designed for a high level of fault tolerance while less attention was paid to cyber security issues. Thus, as indicated in recent studies, it is exposed to various threats. In this article, we suggest enhancing the security of MIL-STD-1553 communication buses by integrating a machine learning-based intrusion detection system (IDS); such anIDS will be capable of detecting cyber attacks in real time. The IDS consists of two modules: 1) a remote terminal (RT) authentication module that detects illegitimately connected components and data transfers and 2) a sequence-based anomaly detection module that detects anomalies in the operation of the system. The IDS showed high detection rates for both normal and abnormal behavior when evaluated in a testbed using real 1553 hardware, as well as a very fast and accurate training process using logs from a real system. The RT authentication module managed to authenticate RTs with +0.99 precision and +0.98 recall; and detect illegitimate component (or a legitimate component that impersonates other components) with +0.98 precision and +0.99 recall. The sequence-based anomaly detection module managed to perfectly detect both normal and abnormal behavior. Moreover, the sequencebased anomaly detection module managed to accurately (i.e., zero false positives) model the normal behavior of a real system in a short period of time ( 22 s).

Operating system (OS) classification is of growing importance to network administrators and cybersecurity analysts alike. The composition of OSs on a network allows for a better quality of device management to be achieved. Additionally, it can be used to identify devices that pose a security risk to the network. However, the sheer number and diversity of OSs that comprise modern networks have vastly increased this management complexity. We leverage insights from social networking theory to provide an encryption-invariant OS classification technique that is quick to train and widely deployable on various network configurations. In particular, we show how an affiliation graph can be used as an input to a machine learning classifier to predict the OS of a device using only the IP addresses for which the device communicates with.We examine the effectiveness of our approach through an empirical analysis of 498 devices on a university campus’ wireless network. In particular, we show our methodology can classify different OS families (i.e., Apple, Windows, and Android OSs) with an accuracy of 99.3%. Furthermore, we extend this study by: 1) examining distinct OSs (e.g., iOS, OS X, and Windows 10); 2) investigating the interval of time required to make an accurate prediction; and, 3) determining the effectiveness of our approach after six months.

Earth science metadata keyword assignment is a challenging problem. Dataset curators select appropriate keywords from the Global Change Master Directory (GCMD) set of keywords. The keywords are integral part of search and discovery of these datasets. Hence, selection of keywords are crucial in increasing the discoverability of datasets. Utilizing machine learning techniques, we provide users with automated keyword suggestions as an improved approach to complement manual selection. We trained a machine learning model that leverages the semantic embedding ability of Word2Vec models to process abstracts and suggest relevant keywords. A user interface tool we built to assist data curators in assignment of such keywords is also described.

Smart grid communication system deeply rely on information technologies which makes it vulnerable to variable cyber-attacks. Among possible attacks, False Data Injection (FDI) Attack has created a severe threat to smart grid control system. Attackers can manipulate smart grid measurements such as collected data of phasor measurement units (PMU) by implementing FDI attacks. Detection of FDI attacks with a simple and effective approach, makes the system more reliable and prevents network outages. In this paper we propose a Decomposed Nearest Neighbor algorithm to detect FDI attacks. This algorithm improves traditional k-Nearest Neighbor by using metric learning. Also it learns the local-optima free distance metric by solving a convex optimization problem which makes it more accurate in decision making. We test the proposed method on PMU dataset and compare the results with other beneficial machine learning algorithms for FDI attack detection. Results demonstrate the effectiveness of the proposed approach.

Classification of Misbehaving Nodes in wireless mobile adhoc networks (MANET) by applying machine learning techniques is an attempt to enhance security by detecting the presence of malicious nodes. MANETs are prone to many security vulnerabilities due to its significant features. The paper compares two machine learning techniques namely Support Vector Machine (SVM) and Back Propagation Neural Network (BPNN) and finds out the best technique to detect the misbehaving nodes. This paper is simulated with an on-demand routing protocol in NS2.35 and the results can be compared using parameters like packet Delivery Ratio (PDR), End-To-End delay, Average Throughput.

Deep learning plays an increasingly important role in various fields due to its superior performance, and it also achieves advanced recognition performance in the field of image classification. However, the vulnerability of deep learning in the adversarial environment cannot be ignored, and the prediction result of the model is likely to be affected by the small perturbations added to the samples by the adversary. In this paper, we propose a two-layer dynamic defense method based on defensive techniques pool and retrained branch model pool. First, we randomly select defense methods from the defense pool to process the input. The perturbation ability of the adversarial samples preprocessed by different defense methods changed, which would produce different classification results. In addition, we conduct adversarial training based on the original model and dynamically generate multiple branch models. The classification results of these branch models for the same adversarial sample is inconsistent. We can detect the adversarial samples by using the inconsistencies in the output results of the two layers. The experimental results show that the two-layer dynamic defense method we designed achieves a good defense effect.

Massive images are being shared via a variety of ways, such as social networking. The rich content of images raise a serious concern for privacy. A great number of efforts have been devoted to designing mechanisms for privacy protection based on the assumption that the privacy is well defined. However, in practice, given a collection of images it is usually nontrivial to decide which parts of images should be protected, since the sensitivity of objects is context-dependent and user-dependent. To meet personalized privacy requirements of different users, we propose a system IEye to automatically detect private parts of images based on both common knowledge and personal knowledge. Specifically, for each user's images, multi-layered semantic graphs are constructed as feature representations of his/her images and a rule set is learned from those graphs, which describes his/her personalized privacy. In addition, an optimization algorithm is proposed to protect the user's privacy as well as minimize the loss of utility. We conduct experiments on two datasets, the results verify the effectiveness of our design to detect and protect personalized image privacy.

With the wild applications of machine learning (ML) technology, automatic speech recognition (ASR) has made great progress in recent years. Despite its great potential, there are various evasion attacks of ML-based ASR, which could affect the security of applications built upon ASR. Up to now, most studies focus on white-box attacks in ASR, and there is almost no attention paid to black-box attacks where attackers can only query the target model to get output labels rather than probability vectors in audio domain. In this paper, we propose an evasion attack against ASR in the above-mentioned situation, which is more feasible in realistic scenarios. Specifically, we first train a substitute model by using data augmentation, which ensures that we have enough samples to train with a small number of times to query the target model. Then, based on the substitute model, we apply Differential Evolution (DE) algorithm to craft adversarial examples and implement black-box attack against ASR models from the Speech Commands dataset. Extensive experiments are conducted, and the results illustrate that our approach achieves untargeted attacks with over 70% success rate while still maintaining the authenticity of the original data well.

The design of codes for communicating reliably over a statistically well defined channel is an important endeavor involving deep mathematical research and wide-ranging practical applications. In this work, we present the first family of codes obtained via deep learning, which significantly outperforms state-of-the-art codes designed over several decades of research. The communication channel under consideration is the Gaussian noise channel with feedback, whose study was initiated by Shannon; feedback is known theoretically to improve reliability of communication, but no practical codes that do so have ever been successfully constructed. We break this logjam by integrating information theoretic insights harmoniously with recurrent-neural-network based encoders and decoders to create novel codes that outperform known codes by 3 orders of magnitude in reliability and achieve a 3dB gain in terms of SNR. We also demonstrate several desirable properties of the codes: (a) generalization to larger block lengths, (b) composability with known codes, and (c) adaptation to practical constraints. This result also has broader ramifications for coding theory: even when the channel has a clear mathematical model, deep learning methodologies, when combined with channel-specific information-theoretic insights, can potentially beat state-of-the-art codes constructed over decades of mathematical research.