Gajraj Kuldeep

Gajraj Kuldeep

This study compares sequential image classification methods based on recurrent neural networks. We describe methods based on recurrent neural networks such as Long-Short-Term memory(LSTM), bidirectional Long-Short-Term memory(BiLSTM) architectures, etc. We also review the state-of-the-art sequential image classification architectures. We mainly focus on LSTM, BiLSTM, temporal convolution network, and independent recurrent neural network architecture in the study. It is known that RNN lacks in learning long-term dependencies in the input sequence. We use a simple feature construction method using orthogonal Ramanujan periodic transform on the input sequence. Experiments demonstrate that if these features are given to LSTM or BiLSTM networks, the performance increases drastically. Our focus in this study is to increase the training accuracy simultaneously reducing the training time for the LSTM and BiLSTM architecture, but not on pushing the state-of-the-art results, so we use simple LSTM/BiLSTM architecture. We compare sequential input with the constructed feature as input to single layer LSTM and BiLSTM network for MNIST and CIFAR datasets. We observe that sequential input to the LSTM network with 128 hidden unit training for five epochs results in training accuracy of 33% whereas constructed features as input to the same LSTM network results in training accuracy of 90% with 1/3 lesser time.

Gajraj Kuldeep, Qi Zhang

In this paper, we design the multi-class privacy$\text{-}$preserving cloud computing scheme (MPCC) leveraging compressive sensing for compact sensor data representation and secrecy for data encryption. The proposed scheme achieves two-class secrecy, one for superuser who can retrieve the exact sensor data, and the other for semi-authorized user who is only able to obtain the statistical data such as mean, variance, etc. MPCC scheme allows computationally expensive sparse signal recovery to be performed at cloud without compromising the confidentiality of data to the cloud service providers. In this way, it mitigates the issues in data transmission, energy and storage caused by massive IoT sensor data as well as the increasing concerns about IoT data privacy in cloud computing. Compared with the state-of-the-art schemes, we show that MPCC scheme not only has lower computational complexity at the IoT sensor device and data consumer, but also is proved to be secure against ciphertext-only attack.

Gajraj Kuldeep, Qi Zhang

Recent study has shown that compressive sensing (CS) based computationally secure scheme using Gaussian or Binomial sensing matrix in resource-constrained IoT devices is vulnerable to ciphertext-only attack. Although the CS-based perfectly secure scheme has no such vulnerabilities, the practical realization of the perfectly secure scheme is challenging, because it requires an additional secure channel to transmit the measurement norm. In this paper, we devise a practical realization of a perfectly secure scheme by concealing energy in which the requirement of an additional secure channel is removed. Since the generation of Gaussian sensing matrices is not feasible in resource-constrained IoT devices, approximate Gaussian sensing matrices are generated using linear feedback shift registers. We also demonstrate the implementation feasibility of the proposed perfectly secure scheme in practice without additional complexity. Furthermore, the security analysis of the proposed scheme is performed and compared with the state-of-the-art compressive sensing based energy obfuscation scheme.