-
L2R-CIPU: Efficient CNN Computation with Left-to-Right Composite Inner Product Units
Authors:
Malik Zohaib Nisar,
Mohammad Sohail Ibrahim,
Muhammad Usman,
Jeong-A Lee
Abstract:
This paper proposes a composite inner-product computation unit based on left-to-right (LR) arithmetic for the acceleration of convolution neural networks (CNN) on hardware. The efficacy of the proposed L2R-CIPU method has been shown on the VGG-16 network, and assessment is done on various performance metrics. The L2R-CIPU design achieves 1.06x to 6.22x greater performance, 4.8x to 15x more TOPS/W,…
▽ More
This paper proposes a composite inner-product computation unit based on left-to-right (LR) arithmetic for the acceleration of convolution neural networks (CNN) on hardware. The efficacy of the proposed L2R-CIPU method has been shown on the VGG-16 network, and assessment is done on various performance metrics. The L2R-CIPU design achieves 1.06x to 6.22x greater performance, 4.8x to 15x more TOPS/W, and 4.51x to 53.45x higher TOPS/mm2 than prior architectures.
△ Less
Submitted 8 July, 2024; v1 submitted 1 June, 2024;
originally announced June 2024.
-
DSLOT-NN: Digit-Serial Left-to-Right Neural Network Accelerator
Authors:
Muhammad Sohail Ibrahim,
Muhammad Usman,
Malik Zohaib Nisar,
Jeong-A Lee
Abstract:
We propose a Digit-Serial Left-tO-righT (DSLOT) arithmetic based processing technique called DSLOT-NN with aim to accelerate inference of the convolution operation in the deep neural networks (DNNs). The proposed work has the ability to assess and terminate the ineffective convolutions which results in massive power and energy savings. The processing engine is comprised of low-latency most-signifi…
▽ More
We propose a Digit-Serial Left-tO-righT (DSLOT) arithmetic based processing technique called DSLOT-NN with aim to accelerate inference of the convolution operation in the deep neural networks (DNNs). The proposed work has the ability to assess and terminate the ineffective convolutions which results in massive power and energy savings. The processing engine is comprised of low-latency most-significant-digit-first (MSDF) (also called online) multipliers and adders that processes data from left-to-right, allowing the execution of subsequent operations in digit-pipelined manner. Use of online operators eliminates the need for the development of complex mechanism of identifying the negative activation, as the output with highest weight value is generated first, and the sign of the result can be identified as soon as first non-zero digit is generated. The precision of the online operators can be tuned at run-time, making them extremely useful in situations where accuracy can be compromised for power and energy savings. The proposed design has been implemented on Xilinx Virtex-7 FPGA and is compared with state-of-the-art Stripes on various performance metrics. The results show the proposed design presents power savings, has shorter cycle time, and approximately 50% higher OPS per watt.
△ Less
Submitted 21 September, 2023; v1 submitted 12 September, 2023;
originally announced September 2023.
-
Counterfactual Explanation and Instance-Generation using Cycle-Consistent Generative Adversarial Networks
Authors:
Tehseen Zia,
Zeeshan Nisar,
Shakeeb Murtaza
Abstract:
The image-based diagnosis is now a vital aspect of modern automation assisted diagnosis. To enable models to produce pixel-level diagnosis, pixel-level ground-truth labels are essentially required. However, since it is often not straight forward to obtain the labels in many application domains such as in medical image, classification-based approaches have become the de facto standard to perform th…
▽ More
The image-based diagnosis is now a vital aspect of modern automation assisted diagnosis. To enable models to produce pixel-level diagnosis, pixel-level ground-truth labels are essentially required. However, since it is often not straight forward to obtain the labels in many application domains such as in medical image, classification-based approaches have become the de facto standard to perform the diagnosis. Though they can identify class-salient regions, they may not be useful for diagnosis where capturing all of the evidences is important requirement. Alternatively, a counterfactual explanation (CX) aims at providing explanations using a casual reasoning process of form "If X has not happend, Y would not heppend". Existing CX approaches, however, use classifier to explain features that can change its predictions. Thus, they can only explain class-salient features, rather than entire object of interest. This hence motivates us to propose a novel CX strategy that is not reliant on image classification. This work is inspired from the recent developments in generative adversarial networks (GANs) based image-to-image domain translation, and leverages to translate an abnormal image to counterpart normal image (i.e. counterfactual instance CI) to find discrepancy maps between the two. Since it is generally not possible to obtain abnormal and normal image pairs, we leverage Cycle-Consistency principle (a.k.a CycleGAN) to perform the translation in unsupervised way. We formulate CX in terms of a discrepancy map that, when added from the abnormal image, will make it indistinguishable from the CI. We evaluate our method on three datasets including a synthetic, tuberculosis and BraTS dataset. All these experiments confirm the supremacy of propose method in generating accurate CX and CI.
△ Less
Submitted 21 January, 2023;
originally announced January 2023.
-
Towards Measuring Domain Shift in Histopathological Stain Translation in an Unsupervised Manner
Authors:
Zeeshan Nisar,
Jelica Vasiljević,
Pierre Gançarski,
Thomas Lampert
Abstract:
Domain shift in digital histopathology can occur when different stains or scanners are used, during stain translation, etc. A deep neural network trained on source data may not generalise well to data that has undergone some domain shift. An important step towards being robust to domain shift is the ability to detect and measure it. This article demonstrates that the PixelCNN and domain shift metr…
▽ More
Domain shift in digital histopathology can occur when different stains or scanners are used, during stain translation, etc. A deep neural network trained on source data may not generalise well to data that has undergone some domain shift. An important step towards being robust to domain shift is the ability to detect and measure it. This article demonstrates that the PixelCNN and domain shift metric can be used to detect and quantify domain shift in digital histopathology, and they demonstrate a strong correlation with generalisation performance. These findings pave the way for a mechanism to infer the average performance of a model (trained on source data) on unseen and unlabelled target data.
△ Less
Submitted 9 May, 2022;
originally announced May 2022.
-
System Identification and Controller Design for Hydraulic Actuator
Authors:
Zainab Nisar,
Hammad Munawar
Abstract:
System Identification of Hydraulic Actuators is critical for analyzing their performance and designing a suitable Control System. Hydraulic actuators are extensively used in many applications, ranging from flight simulators, robotics, orthopaedic surgery, material testing, construction and many other industrial types of machinery. In the aviation industry, hydraulic actuators are currently being u…
▽ More
System Identification of Hydraulic Actuators is critical for analyzing their performance and designing a suitable Control System. Hydraulic actuators are extensively used in many applications, ranging from flight simulators, robotics, orthopaedic surgery, material testing, construction and many other industrial types of machinery. In the aviation industry, hydraulic actuators are currently being used in full flight simulators used for controlling the position and orientation of the motion platform. Every actuator has its own characteristics, therefore, the choice of excitation signals for System Identification must take into account the dynamics of the actuator under consideration. This work proposes the selection of excitation signals based on bandwidth of the hydraulic actuator. Validation of the proposed selection is done by performing system identification, obtaining a mathematical model and comparing it with a nonlinear hydraulic actuator model designed in Simscape. After validation, a nonlinear PID control has been tuned on the identified model and tested on the nonlinear model. Extensive simulations have been run and results show accurate mathematical modelling, as well as precise control has been achieved through the proposed methodology.
△ Less
Submitted 26 August, 2021;
originally announced August 2021.
-
Low Cost 3D Printing for Rapid Prototy** and its Application
Authors:
Taha Hasan Masood Siddique,
Iqra Sami,
Malik Zohaib Nisar,
Mashal Naeem,
Abid Karim,
Muhammad Usman
Abstract:
In the recent years of industrial revolution, 3D printing has shown to grow as an expanding field of new applications. The low cost solutions and short time to market makes it a favorable candidate to be utilized in the dynamic fields of engineering. Additive printing has the vast range of applications in many fields. This study presents the wide range of applications of the 3D printers along with…
▽ More
In the recent years of industrial revolution, 3D printing has shown to grow as an expanding field of new applications. The low cost solutions and short time to market makes it a favorable candidate to be utilized in the dynamic fields of engineering. Additive printing has the vast range of applications in many fields. This study presents the wide range of applications of the 3D printers along with the comparison of the additive printing with the traditional manufacturing methods have been shown. A tutorial is presented explaining the steps involved in the prototype printing using Rhinoceros 3D and Simplify 3D software including the detailed specifications of the end products that were printed using the Delta 3D printer.
△ Less
Submitted 25 November, 2019;
originally announced November 2019.
