Distributed Neural Networks for Internet of Things: The Big-Little Approach

Coninck, Elias De; Verbelen, Tim; Vankeirsbilck, Bert; Bohez, Steven; Simoens, Pieter; Demeester, Piet; Dhoedt, Bart

doi:10.1007/978-3-319-47075-7_52

Cited by 35 publications

(36 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Fuzzy logic and artificial neural networks are important techniques in data fusion, in which the data from many sensors is combined in various ways [56,57]. This is an important function in the IoT.…”

Section: Fuzzy Systems Theory/artificial Neural Networkmentioning

confidence: 99%

Research Challenges for the Internet of Things: What Role Can OR Play?

Ryan

Watson²

2017

Systems

View full text Add to dashboard Cite

Abstract:The Internet of Things (IoT) is an extension of the Internet in which large numbers of "things", including sensors, actuators and processors, in addition to human users, are networked and able to provide high resolution data on their environment and exercise a degree of control over it. It is still at an early stage of development, and many problems/research challenges must be solved before it is widely adopted. Many of these are technical, including interoperability and scalability, as billions of heterogeneous devices will be connected, but deciding on how to invest in the IoT is a challenge for business, and there are also major social, legal and ethical challenges, including security and privacy of data collection, which must be resolved. As the future IoT will be a multi-national, multi-industry, multi-technology infrastructure, the paper reviews the global standardization efforts that are underway to facilitate its worldwide creation and adoption. The main purpose of the paper is to give a broad survey, based on published literature, of the methods of Operations Research (OR), both the mathematical tools and techniques of "hard" OR, and the various approaches of Systems Thinking, including "soft" OR, which may assist in dealing with these problems. A subset of these is described in greater depth to better convey what might be involved in applying OR and Systems Thinking to the IoT. It is suggested that OR has a role to play in balancing the technical and non-technical research challenges which confront the IoT.

show abstract

Section: Fuzzy Systems Theory/artificial Neural Networkmentioning

confidence: 99%

Research Challenges for the Internet of Things: What Role Can OR Play?

Ryan

Watson²

2017

Systems

View full text Add to dashboard Cite

show abstract

“…Based on a network connection, the cloud comes with variable and high latency and a limited upload bandwidth [9]. Furthermore, the external processing includes additional costs [10] and can make the data accessible to third parties. This leads to the current approach to use embedded systems for a preprocessing of the data [10], which can reduce the data size and therefore lowers the costs for server and cloud capacity.…”

Section: Discussionmentioning

confidence: 99%

“…In [10] a so-called Big-Little neural network architecture is presented. The little neural network only classifies a subset of the output classes and can be executed locally with limited processing performance on an embedded system.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

A PLC-Based Measuring System for Machine Crosslinking and Monitoring

Küfner¹,

Reger²,

Schönig³

2018

dtetr

View full text Add to dashboard Cite

Rising energy costs, a green image and fulfilling of political demands are some reasons, why companies try to reduce their energy consumption. Biggest challenge is to identify potentials, derive measures and realize them. Although, measuring instruments are available in a great variety there exists no cheap and flexible system, which uses the measured line for supply and data transmission. This innovation, introduced in this paper, delivers a new possibility for post-crosslinking and smart energy monitoring. Based on current powerlinecommunication (PLC) technologies and modulation methods, a hardware was developed. A program handles the PLC and manages the measuring. The system has sensors for alternating voltage and current and calculates frequency, power factor, active, reactive and apparent power. For visualization and further analysis, a software was established. The complete system was evaluated in a production facility. The accuracy shows a difference of 1.5 % to a reference.

show abstract

“…Generally, in PULP-based processors, the cluster domain is used for applications where a large amount of computations is required, while the FC is usually engaged in I/O handling or other kinds of scheduling. However, for an application scenario where, for example, a small network is used to detect the onset and, once the onset is detected, a deeper network is used for classification [44], both domains (SoC and Cluster) have its own advantage: the FC continuously reads the sensory data and executes the onset detection algorithm, while the cluster domain is activated once the onset is detected to perform the classification with a deep NN. In this case, our proposed framework stores the small network into the private L2 memory for the FC, while for the classification task, the DMA unit transfers the network using the double-buffering technique into the L1 memory for the computation in cluster domain.…”

Section: B the Fann-on-mcu Deployment Frameworkmentioning

confidence: 99%

FANN-on-MCU: An Open-Source Toolkit for Energy-Efficient Neural Network Inference at the Edge of the Internet of Things

Wang

Magno

Cavigelli

et al. 2020

IEEE Internet Things J.

130

View full text Add to dashboard Cite

The growing number of low-power smart devices in the Internet of Things is coupled with the concept of "Edge Computing", that is moving some of the intelligence, especially machine learning, towards the edge of the network. Enabling machine learning algorithms to run on resource-constrained hardware, typically on low-power smart devices, is challenging in terms of hardware (optimized and energy-efficient integrated circuits), algorithmic and firmware implementations. This paper presents FANN-on-MCU, an open-source toolkit built upon the Fast Artificial Neural Network (FANN) library to run lightweight and energy-efficient neural networks on microcontrollers based on both the ARM Cortex-M series and the novel RISC-Vbased Parallel Ultra-Low-Power (PULP) platform. The toolkit takes multi-layer perceptrons trained with FANN and generates code targeted to low-power microcontrollers. This paper also presents detailed analyses of energy efficiency across the different cores, and the optimizations to handle different network sizes. Moreover, it provides a detailed analysis of parallel speedups and degradations due to parallelization overhead and memory transfers. Further evaluations include experimental results for three different applications using a self-sustainable wearable multi-sensor bracelet. Experimental results show a measured latency in the order of only a few microseconds and power consumption of a few milliwatts while keeping the memory requirements below the limitations of the targeted microcontrollers. In particular, the parallel implementation on the octa-core RISC-V platform reaches a speedup of 22x and a 69% reduction in energy consumption with respect to a single-core implementation on Cortex-M4 for continuous real-time classification.

show abstract

Distributed Neural Networks for Internet of Things: The Big-Little Approach

Cited by 35 publications

References 11 publications

Research Challenges for the Internet of Things: What Role Can OR Play?

Research Challenges for the Internet of Things: What Role Can OR Play?

A PLC-Based Measuring System for Machine Crosslinking and Monitoring

FANN-on-MCU: An Open-Source Toolkit for Energy-Efficient Neural Network Inference at the Edge of the Internet of Things

Contact Info

Product

Resources

About