A Multi-Level-Optimization Framework for FPGA-Based Cellular Neural Network Implementation

Liu, Zhongyang; Luo, Shaoheng; Xu, Xiaowei; Shi, Yiyu; Zhuo, Cheng

doi:10.1145/3273957

Cited by 19 publications

(4 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…which is referred to as the throughput on P for a network configuration θ [2], [20]. Due to possible quantization the commonly used metric floatingpoint operations per second (FLOP/s) is generalized to operations per second (OP/s) in this work similar to [17].…”

Section: B Metricsmentioning

confidence: 99%

BackboneAnalysis: Structured Insights into Compute Platforms from CNN Inference Latency

Hafner

Zeller

Schutera

et al. 2022

2022 IEEE Intelligent Vehicles Symposium (IV)

View full text Add to dashboard Cite

Customization of a convolutional neural network (CNN) to a specific compute platform involves finding an optimal pareto state between computational complexity of the CNN and resulting throughput in operations per second on the compute platform. However, existing inference performance benchmarks compare complete backbones that entail many differences between their CNN configurations, which do not provide insights in how fine-grade layer design choices affect this balance.We present BackboneAnalysis, a methodology for extracting structured insights into the trade-off for a chosen target compute platform. Within a onefactor-at-a-time analysis setup, CNN architectures are systematically varied and evaluated based on throughput and latency measurements irrespective of model accuracy. Thereby, we investigate the configuration factors input shape, batch size, kernel size and convolutional layer type.In our experiments, we deploy BackboneAnalysis on a Xavier iGPU and a Coral Edge TPU accelerator. The analysis reveals that the general assumption from optimal Roofline performance that higher operation density in CNNs leads to higher throughput does not always hold. These results highlight the importance for a neural network architect to be aware of platform-specific latency and throughput behavior in order to derive sensible configuration decisions for a custom CNN.

show abstract

Section: B Metricsmentioning

confidence: 99%

BackboneAnalysis: Structured Insights into Compute Platforms from CNN Inference Latency

Hafner

Zeller

Schutera

et al. 2022

2022 IEEE Intelligent Vehicles Symposium (IV)

View full text Add to dashboard Cite

show abstract

“…To further optimise and accelerate the DNNs, many works have been proposed and can be divided into four categories: matrix decomposition [42,43], network quantisation [44][45][46][47], knowledge distilling [48,49] and pruning [50][51][52][53]. In this paper, we mainly focus on model pruning.…”

Section: Light-weighted Dnn Optimisationmentioning

confidence: 99%

An efficient Industrial Internet of Things video data processing system for protocol identification and quality enhancement

Chen

Liu

Zhang

2022

IET Cyber-Phy Sys Theory & Ap

View full text Add to dashboard Cite

Video has become an essential medium to monitoring, identification and knowledge sharing. For industrial applications, especially Industrial Internet of Things (IIoT), videos encoded with specific protocols are transferred to smart gateways. In a typical IIoT scenario, the protocol of the video is firstly recognised, which prepares for subsequent video tasks. Due to the constrained resources in such scenarios, the video quality can be deteriorated during encoding and compression processes, which is challenging for IIoT. Recently, there have been extensive works focussing on the protocol identification (PI) and video quality enhancement (VQE) tasks on IIoT edge devices using deep neural networks (DNNs). Since DNNs often require high computational resources, complex networks can hardly be deployed on edge devices. An IIoT system which can efficiently identify the stream protocol and enhance the video quality is proposed in this study. The light‐weighted network designs and inference optimisation techniques have been proposed for PI and VQE to realise efficient deployments. Our proposed system employed on an IIoT edge device can achieve an accuracy of higher than 97.52% with fast inference speed for PI. For the VQE task, our system has demonstrated superior performance (15.230 FPS, 0.773 FPS/W) in comparison with the state‐of‐the‐art methods.

show abstract

“…Among the six literature, [66] focuses on the hybrid object types of multilayer and multi-core, and [67] hybrids multigrid and multi-core. In addition, "multi-layer" accounts for 2 papers [68,69] and each of the remaining "object type" is mentioned for once in the literature pool, that include multi-processor [70], multi-goal [71], multioutput [72], multi-constrained path [73], feature-rich [74], multi-phase [75], multiple Indexes [76], multiechelon [76], multi-heuristic [77], multi-Level [78], multi-view [79], multi-class [80], multigrid [81], many-field [82], multiple access [81], multi-population [64], multiple voltage regulators [83], multiple Smartphones [84].…”

Section: A Optimizationmentioning

confidence: 99%

A Comprehensive Review of Machine Learning in Multi-objective Optimization

Clausen

et al. 2021

2021 IEEE 4th International Conference on Big Data and Artificial Intelligence (BDAI)

View full text Add to dashboard Cite

In the real world, it is challenging to calculate a trade-off alternative with traditional classical methods for complex non-linear systems, which always involve multiple conflicting objectives. Such complicated systems urgently desire advanced methods to conquer the multi-objective optimization problems (MOPs). As a promising AI method, the development and application of Machine Learning (ML) attract increasingly more attention from researchers. The natures of ML methods, such as parallel computation possibility, no need for any priori assumptions, etc., ensure the effectiveness and efficiency for solving MOPs. However, as we know, there is no literature related to the comprehensive review of ML in multi-objective optimization domain until now. This literature review aims to provide researchers a global view of mainstream ML methods for MOO in a general domain and a reference for applying ML methods to solve a specific type of MOPs. In this paper, the general ML mainstream methods are summarized, based on which the literature relating to ML on MOPs are retrieved in comprehensive domains. The relevant literature is categorized according to the emphasis of object types, purposes and methods, and the categorization results are finally analyzed and discussed.

show abstract

A Multi-Level-Optimization Framework for FPGA-Based Cellular Neural Network Implementation

Cited by 19 publications

References 23 publications

BackboneAnalysis: Structured Insights into Compute Platforms from CNN Inference Latency

BackboneAnalysis: Structured Insights into Compute Platforms from CNN Inference Latency

An efficient Industrial Internet of Things video data processing system for protocol identification and quality enhancement

A Comprehensive Review of Machine Learning in Multi-objective Optimization

Contact Info

Product

Resources

About