26 May 2023 in IEEE Embedded Systems Letters
This paper proposes novel approximate 4:2 compressors developed using input reordering circuits and input combination probabilities. The input reordering circuit is used to reduce the hardware complexity of the proposed designs. This paper proposes two designs of approximate 4:2 compressors. This compressor is used in designing an approximate multiplier. The proposed multiplier designs utilize less energy than the already published ones due to acceptable inaccurate output/precision, which are best suitable for image processing applications. The proposed multiplier designs MUL1, MUL2, MUL3 and MUL4 saves 22.75%, 21.95%, 11.57%, and 8.95% energy than the best of the existing design 1.
18 March 2025 in IEEE Access
Adversarial attacks, specialized attacks, pose a severe threat to AI model performance in various applications, including the Internet of Things (IoT). Various defense mechanisms have been proposed to counter these attacks. However, their primary limitation lies in their inability to effectively handle broader datasets derived from diverse applications. In this study, we utilize multiple AI models with adaptive weights applied at different neural network layers to achieve enhanced performance and more robust results. This study introduced a novel AI-based deep learning model to detect adversarial threats within IoT systems, optimizing data preprocessing, feature extraction, and classification through a holistic approach. A three-stage filtering technique featuring Adaptive Weights was applied to enhance the data preprocessing efficiency. A two-level adaptive feature extraction strategy was utilized to maximize feature learning performance. This is refined using adaptive dilated enriched convolution operations, whereas statistical attributes are optimized through a Quantum-inspired Coati Optimization Algorithm (Q-COA). A dual system based on self-attention combines a Restricted Boltzmann Machine (RBM) with a Recurrent Convolutional Neural Network (RCNN). This configuration effectively identifies adversarial attacks by linking classifiers via a self-attention-driven weight-sharing mechanism. The proposed two-level weight-sharing approach surpasses conventional classifiers and achieves superior classification accuracy. This comprehensive Artificial Intelligence (AI) model significantly improves the preprocessing efficiency, feature learning performance, and classification accuracy, offering an innovative and robust solution for adversarial attack detection in IoT systems. The performance metric, Area Under the Curve (AUC), achieves values of 0.95 and 0.97 for two datasets using the proposed model, highlighting its effectiveness compared to the models in the comparison.
26 February 2024 in IEEE Access
The posit number system aims to be a drop-in replacement of the existing IEEE floating-point standard. Its properties-tapered precision and high dynamic range, allow a smaller size posit to almost match the performance of a much larger size floating-point in representing decimals. This becomes especially useful for performing error-tolerant tasks like deep learning inference computation where low latency and area are a priority. Recent research has found that the performance of deep neural network models saturates beyond a certain level of accuracy of multipliers used for convolutions. Therefore, the extra hardware cost of developing precise arithmetic circuits for such applications becomes an unnecessary overhead. This paper explores approximate posit multipliers in the convolutional layers of deep neural networks and attempts to find an ideal balance between hardware utilization and inference accuracy. Posit multiplication involves several steps, with the mantissa multiplication step utilizing maximum hardware resources. To mitigate this, a posit multiplier circuit using an approximate hybrid-radix Booth encoding for mantissa multiplication and techniques such as truncation and bit masking based on input regime size are proposed. In addition, a novel Booth encoding control scheme to prevent unnecessary bits from switching has been devised to reduce dynamic power dissipation. Compared to existing literature, these optimizations have contributed to a 23% decrease in power dissipation in the mantissa multiplication stage. Further, a novel area and energy-efficient decoder architecture have also been developed with an 11% reduction in dynamic power dissipation and area compared to existing decoders. Overall, the proposed < 16, 2 > posit multiplier offers a 14% reduction in the PDP over the existing approximate posit multiplier designs. The proposed < 16, 2 > multiplier also achieves over 90% accuracy in inferencing deep learning models such as ResNet20, VGG-19 and DenseNet.
30 October 2024 in IEEE Embedded Systems Letters
The posit number system represents a binary numerical format devised primarily to mitigate the shortcomings of the IEEE 754 floating-point standard commonly used in modern computer arithmetic. Unlike IEEE 754, which utilizes a fixed-width representation, posits offer a variable-size encoding, dynamically adjusting the number of bits allocated based on the magnitude of the represented number. This adaptability enables posits to provide enhanced precision across a broader range of values, addressing issues like gradual underflow and the multiple representations of zero and NaN values inherent in IEEE 754. However, the dynamic bit allocation also poses significant challenges in decoding the posit number into its constituent fields and packing back the fields of the resultant posit once the arithmetic operations have been performed. For posits to become a viable alternative to floating-points in practical computing systems, the decoding and encoding overheads of posits need to be minimized. Hence, the aim of this paper is to develop energy-efficient hardware for posit decoding and encoding. The proposed < 16, 2 > posit decoders, Decoder A and Decoder B show an improvement of over 65% and 33% compared to literature in terms of energy requirements. Similarly, the proposed < 16, 2 > encoder circuit is over 52% more energy-efficient than existing encoder circuits.