When building a unified vision system or gradually adding new capabilities to a system, the usual assumption is that training data for all tasks is always available. However, as the number of tasks grows, storing and retraining on such data becomes infeasible. A new problem arises where we add new capabilities to a Convolutional Neural Network (CNN), but the training data for its existing capabilities are unavailable. We propose our Learning without Forgetting method, which uses only new task data to train the network while preserving the original capabilities. Our method performs favorably compared to commonly used feature extraction and fine-tuning adaption techniques and performs similarly to multitask learning that uses original task data we assume unavailable. A more surprising observation is that Learning without Forgetting may be able to replace fine-tuning with similar old and new task datasets for improved new task performance.

Learning without Forgetting

Software Defined Networking (SDN) has been proposed as a drastic shift in the networking paradigm, by decoupling network control from the data plane and making the switching infrastructure truly programmable. The key enabler of SDN, OpenFlow, has seen widespread deployment on production networks and its adoption is constantly increasing. Although openness and programmability are primary features of OpenFlow, security is of core importance for real-world deployment. In this work, we perform a security analysis of OpenFlow using STRIDE and attack tree modeling methods, and we evaluate our approach on an emulated network testbed. The evaluation assumes an attacker model with access to the network data plane. Finally, we propose appropriate counter-measures that can potentially mitigate the security issues associated with OpenFlow networks. Our analysis and evaluation approach are not exhaustive, but are intended to be adaptable and extensible to new versions and deployment contexts of OpenFlow.

OpenFlow: A security analysis

Traffic flow prediction using LSTM with feature enhancement

IEEE International Conference on Robotics and Automation

In recent years, industrial wireless sensor networks (IWSNs) have emerged as an important research theme with applications spanning a wide range of industries including automation, monitoring, process control, feedback systems, and automotive. Wide scope of IWSNs applications ranging from small production units, large oil and gas industries to nuclear fission control, enables a fast-paced research in this field. Though IWSNs offer advantages of low cost, flexibility, scalability, self-healing, easy deployment, and reformation, yet they pose certain limitations on available potential and introduce challenges on multiple fronts due to their susceptibility to highly complex and uncertain industrial environments. In this paper, a detailed discussion on design objectives, challenges, and solutions, for IWSNs, are presented. A careful evaluation of industrial systems, deadlines, and possible hazards in industrial atmosphere are discussed. This paper also presents a thorough review of the existing standards and industrial protocols and gives a critical evaluation of potential of these standards and protocols along with a detailed discussion on available hardware platforms, specific industrial energy harvesting techniques and their capabilities. This paper lists main service providers for IWSNs solutions and gives insight of future trends and research gaps in the field of IWSNs.

/pdf/a-critical-analysis-of-research-potential-challenges-and-14no6a3bs7.pdf

A Critical Analysis of Research Potential, Challenges, and Future Directives in Industrial Wireless Sensor Networks

Deep clustering is a fundamental yet challenging task for data analysis. Recently we witness a strong tendency of combining autoencoder and graph neural networks to exploit structure information for clustering performance enhancement. However, we observe that existing literature 1) lacks a dynamic fusion mechanism to selectively integrate and refine the information of graph structure and node attributes for consensus representation learning; 2) fails to extract information from both sides for robust target distribution (i.e., “groundtruth” soft labels) generation. To tackle the above issues, we propose a Deep Fusion Clustering Network (DFCN). Specifically, in our network, an interdependency learning-based Structure and Attribute Information Fusion (SAIF) module is proposed to explicitly merge the representations learned by an autoencoder and a graph autoencoder for consensus representation learning. Also, a reliable target distribution generation measure and a triplet self-supervision strategy, which facilitate cross-modality information exploitation, are designed for network training. Extensive experiments on six benchmark datasets have demonstrated that the proposed DFCN consistently outperforms the state-of-the-art deep clustering methods.

Deep Fusion Clustering Network

In the current medical system, the privacy and security of the medical information with respect to its transmission and storage is a key challenge. In pursuit of this, we proposed a contourlet transform and Discrete Cosine Transform (DCT) based multiple robust watermarking algorithms for medical imaging. The proposed scheme uses contourlet transform to extract multidirectional and multiscale texture information, and then the DCT was used to acquire the feature vector in the low frequency directional subbands. Then we used Logistic Map to encrypt the watermark to ensure the security of the original watermarking information under a chaotic system. In watermark embedding and extraction phases, we adopted zero-watermarking technique to ensure the integrity of the medical images. Our experimental results show that the proposed algorithm can embed much more data with less complexity and dose not change the pixel value of the original image. Moreover, the proposed algorithm can extract the watermark effectively with good invisibility and robustness to common and geometric attacks.

Contourlet-DCT based multiple robust watermarkings for medical images

Traffic flow forecasting is the key in intelligent transportation system, but the current traffic flow forecasting method has low accuracy and poor stability in the long-term period. For this reason, an improved LSTM Network is proposed. Firstly, the concept and calculation method of time singularity ratio of traffic data stream is proposed to predict long-term traffic flow. The singular point probability LSTM (SPP-LSTM) is presented. Namely, the algorithm discard the LSTM network unit form the network temporarily according to the singular point probability during the training process of the depth learning network, so as to get SPP-LSTM model. Finally, the paper amends the SPP-LSTM by ARIMA to realize the accurate prediction of 24-hour traffic flow data. Theoretical analysis and experimental results show that the SPP-LSTM has a high accuracy rate, stability and wide application prospect in the long-time traffic flow forecast with hourly period.

Singular Point Probability Improve LSTM Network Performance for Long-term Traffic Flow Prediction

Distributed denial-of-service (DDoS) attacks present serious threats to servers in the Internet. We argue that the difference of the goals, manners and results of the interaction behaviors of normal flows and attack flows, which show different characteristics on IP addresses and ports. IAI (IP Address Interaction Feature) algorithm is proposed based on the addresses interaction, abrupt traffic change, addresses many-to-one dissymmetry, distributed source IP addresses and concentrated target addresses. The IAI is designed to describe the essential characteristics of network flow states. Furthermore, a support vector machine (SVM) classifier, which is trained by IAI time series from normal flow and attack flow, is applied to classify the state of current network flows and identify the DDoS attacks. The experiment results show that, IAI can reflect the different characteristics of DDoS attack flows and normal flows; the IAI-based detection scheme can distinguish between normal flows and abnormal flows with DDoS attack flows effectively, and help to identify fast and accurate attack flows when the attacking traffic is hidden among a relatively large volume of normal flows or close to the attacking sources, and it has higher detection and lower false alarm rate compared with related works.

DDoS Attack Detection Using IP Address Feature Interaction

In order to solve the problem of patient information security protection in medical images, whilst also taking into consideration the unchangeable particularity of medical images to the lesion area and the need for medical images themselves to be protected, a novel robust watermarking algorithm for encrypted medical images based on dual-tree complex wavelet transform and discrete cosine transform (DTCWT-DCT) and chaotic map is proposed in this paper. First, DTCWT-DCT transformation was performed on medical images, and dot product was per-formed in relation to the transformation matrix and logistic map. Inverse transformation was undertaken to obtain encrypted medical images. Then, in the low-frequency part of the DTCWT-DCT transformation coefficient of the encrypted medical image, a set of 32 bits visual feature vectors that can effectively resist geometric attacks are found to be the feature vector of the encrypted medical image by using perceptual hashing. After that, different logistic initial values and growth parameters were set to encrypt the watermark, and zero-watermark technology was used to embed and extract the encrypted medical images by combining cryptography and third-party concepts. The proposed watermarking algorithm does not change the region of interest of medical images thus it does not affect the judgment of doctors. Additionally, the security of the algorithm is enhanced by using chaotic mapping, which is sensitive to the initial value in order to encrypt the medical image and the watermark. The simulation results show that the pro-posed algorithm has good homomorphism, which can not only protect the original medical image and the watermark information, but can also embed and extract the watermark directly in the encrypted image, eliminating the potential risk of decrypting the embedded watermark and extracting watermark. Compared with the recent related research, the proposed algorithm solves the contradiction between robustness and invisibility of the watermarking algorithm for encrypted medical images, and it has good results against both conventional attacks and geometric attacks. Under geometric attacks in particular, the proposed algorithm performs much better than existing algorithms.

Jieren Cheng

Papers

Deep Fusion Clustering Network

Contourlet-DCT based multiple robust watermarkings for medical images

Singular Point Probability Improve LSTM Network Performance for Long-term Traffic Flow Prediction

DDoS Attack Detection Using IP Address Feature Interaction

A novel robust watermarking algorithm for encrypted medical image based on DTCWT-DCT and chaotic map