Approximate Computing Methods for Embedded Machine Learning

@article{Ibrahim2018ApproximateCM,
  title={Approximate Computing Methods for Embedded Machine Learning},
  author={Ali Ibrahim and Mario Osta and Mohamad Gabriel Alameh and Moustafa Saleh and Hussein Chible and Maurizio Valle},
  journal={2018 25th IEEE International Conference on Electronics, Circuits and Systems (ICECS)},
  year={2018},
  pages={845-848}
}
  • A. Ibrahim, M. Osta, M. Valle
  • Published 1 December 2018
  • Computer Science
  • 2018 25th IEEE International Conference on Electronics, Circuits and Systems (ICECS)
Embedding Machine Learning enables integrating intelligence in recent application domains such as Internet of Things, portable healthcare systems, and wearable devices. This paper presents an assessment of approximate computing methods at algorithmic, architecture, and circuit levels and draws perspectives for further developments and applications. The main goal is to investigate how approximate computing may reduce the complexity and enable the feasibility of embedded Machine Learning (ML… 
View on IEEE
doi.org
14 Citations
Background Citations
5
Methods Citations
5

Figures and Tables from this paper

14 Citations

Energy Efficient Implementation of Machine Learning Algorithms on Hardware Platforms
TLDR
An overview about state of the art techniques enabling efficient implementation of Machine and Deep learning (ML/DL) algorithms aiming to improve the energy efficiency is presented and an assessment of the algorithms suitable for embedded implementation is provided.
Near Sensors Computation based on Embedded Machine Learning for Electronic Skin
Random-based networks with dropout for embedded systems
TLDR
A novel training procedure for random-based neural networks, which combines ensemble techniques and dropout regularization is presented, which limits the computational complexity of the training phase without affecting classification performance significantly; the method best fits Internet of Things (IoT) applications.
Algorithmic-Level Approximate Tensorial SVM Using High-Level Synthesis on FPGA
TLDR
This paper presents the first FPGA implementation of an approximate tensorial Support Vector Machine (SVM) classifier with algorithmic level ACTs using High-Level Synthesis (HLS).
Data Oriented Approximate K-Nearest Neighbor Classifier for Touch Modality Recognition
TLDR
Experimental results demonstrate that the use of software level approximation techniques reduce the execution time and memory usage up to 38% and 55% respectively, at the cost of accuracy loss less than 10% for the target application.
Latency Control for Distributed Machine Vision at the Edge Through Approximate Computing
TLDR
This paper shows how approximate computation techniques can be used to design a latency controller that uses multiple video frame image quality control knobs to simultaneously satisfy latency and accuracy requirements for machine vision applications involving object detection, and human pose estimation.
An Energy Efficient System for Touch Modality Classification in Electronic Skin Applications
  • M. Osta, A. Ibrahim, M. Valle
  • Computer Science
    2019 IEEE International Symposium on Circuits and Systems (ISCAS)
  • 2019
TLDR
The design and implementation of an energy efficient smart system for tactile sensing based on a RISC-V parallel ultra-low power platform (PULP), called Mr. Wolf, performs the on-board classification of different touch modalities and achieves 15× better energy efficiency than the classification done on the STM32F40.
Mez: An Adaptive Messaging System for Latency-Sensitive Multi-Camera Machine Vision at the IoT Edge
TLDR
Experimental evaluation on an IoT Edge testbed with a pedestrian detection machine vision application indicates that Mez is able to tolerate latency variations of up to 10x with a worst-case reduction of 4.2% of the application accuracy F1 score metric.
An On-Device Learning System for Estimating Liquid Consumption from Consumer-Grade Water Bottles and Its Evaluation
TLDR
A detailed characterization of sip detection is performed to understand the accuracy-complexity tradeoffs by developing and implementing a variety of different ML models with varying complexities.
Mez: A Messaging System for Latency-Sensitive Multi-Camera Machine Vision at the IoT Edge
TLDR
Experimental evaluation on an IoT Edge testbed with a pedestrian detection machine vision application indicates that Mez is able to tolerate latency variations of up to 10x with a worst-case reduction of 4.2\% in the application accuracy F1 score metric.
...
...

References

SHOWING 1-10 OF 29 REFERENCES
Designing Hardware for Machine Learning: The Important Role Played by Circuit Designers
  • V. Sze
  • Computer Science
    IEEE Solid-State Circuits Magazine
  • 2017
TLDR
A short overview of the key concepts in machine learning is given, its challenges particularly in the embedded space are discussed, and various opportunities where circuit designers can help to address these challenges are highlighted.
Exploiting approximate computing for deep learning acceleration
TLDR
Based on earlier studies demonstrating that DNNs are resilient to numerical errors from approximate computing, techniques to reduce communication overhead of distributed deep learning training via adaptive residual gradient compression (AdaComp), and computation cost for deep learning inference via Prameterized clipping ACTivation (PACT) based network quantization are presented.
Energy Efficient Neural Computing: A Study of Cross-Layer Approximations
TLDR
The experiments on recognition benchmarks show that cross-layer approximation provides substantial improvements in energy efficiency for different accuracy/quality requirements, and a synergistic framework for combining the approximation techniques to achieve maximal energy benefits from approximate DNNs is proposed.
Adaptive and Energy-Efficient Architectures for Machine Learning: Challenges, Opportunities, and Research Roadmap
TLDR
The challenges and opportunities for building energy-efficient and adaptive architectures for machine learning are discussed, and brain-inspired emerging computing paradigms, such as approximate computing; that can further reduce the energy requirements of the system are focused on.
ApproxANN: An approximate computing framework for artificial neural network
TLDR
This work proposes a novel approximate computing framework for ANN, namely ApproxANN, which characterizes the impact of neurons on the output quality in an effective and efficient manner, and judiciously determine how to approximate the computation and memory accesses of certain less critical neurons to achieve the maximum energy efficiency gain under a given quality constraint.
Inexact Arithmetic Circuits for Energy Efficient IoT Sensors Data Processing
TLDR
The proposed inexact arithmetic circuits achieve up to 80% power saving when compared to the exact one, and similar solutions presented in literature with a maximum loss of 1.39 dB in output SNR when applied to FIR filters.
Real-Time Embedded Machine Learning for Tensorial Tactile Data Processing
  • A. Ibrahim, M. Valle
  • Computer Science
    IEEE Transactions on Circuits and Systems I: Regular Papers
  • 2018
TLDR
This paper presents the hardware architectures and implementation of a real time ML method based on tensorial kernel approach dealing with multidimensional input tensors and demonstrates the feasibility of the proposed implementations for real time classification.
AxNN: Energy-efficient neuromorphic systems using approximate computing
TLDR
A new approach to design energy-efficient hardware implementations of large-scale neural networks (NNs) using approximate computing and a programmable and quality-configurable neuromorphic processing engine (qcNPE), which utilizes arrays of specialized processing elements that execute neuron computations with dynamically configurable accuracies and can be used to execute AxNNs from diverse applications.
Bio-Inspired Imprecise Computational Blocks for Efficient VLSI Implementation of Soft-Computing Applications
TLDR
It is shown that these proposed Bio-inspired Imprecise Computational blocks (BICs) can be exploited to efficiently implement a three-layer face recognition neural network and the hardware defuzzification block of a fuzzy processor.
Leveraging the error resilience of machine-learning applications for designing highly energy efficient accelerators
TLDR
This paper proposes to expand the application scope, error tolerance as well as the energy savings of inexact computing systems through neural network architectures, and demonstrates that the proposed inexact neural network accelerator could achieve 43.91%-62.49% savings in energy consumption.
...
...