User Tools

Site Tools


en:faultydnn

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
en:faultydnn [2018/05/12 22:15]
francoislp
en:faultydnn [2018/05/12 22:17] (current)
francoislp
Line 1: Line 1:
 ===Project: Deep Neural Networks on Faulty Hardware=== ===Project: Deep Neural Networks on Faulty Hardware===
  
-Machine learning algorithms based on deep neural networks (DNNs) have now become extremely good at solving a variety of signal processing or optimization challenges such as image recognition,​ speech processing, complex game solvers, etc. However, ​most powerful ​algorithms also require significant computing power for training but also for the inference step, in which the trained system makes decisions based on new inputs. It is crucial to reduce the energy consumption of DNN implementations,​ so that powerful ​AI algorithms can be used in low-power embedded devices, without requiring costly (in terms of energy and latency) communication back to the cloud.+Machine learning ​(ML) algorithms based on deep neural networks (DNNs) have now become extremely good at solving a variety of signal processing or optimization challenges such as image recognition,​ speech processing, complex game solvers, etc. However, powerful ​DNNs require significant computing power for training but also for the inference step, in which the trained system makes decisions based on new inputs. It is crucial to reduce the energy consumption of DNN implementations,​ so that powerful ​ML algorithms can be used in low-power embedded devices, without requiring costly (in terms of energy and latency) communication back to the cloud.
  
 It turns out that vanilla DNNs are quite fault tolerant. For instance, in an implementation that relies on stochastic computing, accepting occasional timing violations in the circuit can further reduce energy consumption [[http://​ieeexplore.ieee.org/​document/​7839313/​|(see paper)]]. It turns out that vanilla DNNs are quite fault tolerant. For instance, in an implementation that relies on stochastic computing, accepting occasional timing violations in the circuit can further reduce energy consumption [[http://​ieeexplore.ieee.org/​document/​7839313/​|(see paper)]].
en/faultydnn.txt ยท Last modified: 2018/05/12 22:17 by francoislp