An Automatic Sound ClassifiCation Framework …
435
the SNN classifier, a fully SNN-based framework when implemented with NVM
devices would translate to significant power saving and speed up.
Besides emulating synaptic functionalities, several works also explored using
NVM to emulate the neuronal dynamics, including membrane potentials maintenance, transient dynamic, spike generation and transmission [62–65]. Neuromorphic systems implemented fully with PCM have been reported in [64, 65], in which
the neurons are integrated seamlessly with dense synaptic arrays using PCM. For
NVM implemented synapses, each conductance state of an NVM is important and
impacts the synaptic integration performed at the post-synaptic neurons. In contrast, the requirements for continuum and linearity of conductance states are less
stringent when implementing leaky integrate-and-fire neurons. The biological neuron fires whenever the membrane potential crosses the firing threshold, hence, it
is sufficient that accumulated electrical pulses trigger the ‘firing’ after receiving a
certain number of pulses, without explicit conductance update for every electrical
pulse. Furthermore, the non-volatility property of certain materials is beneficial to
implement the leaky dynamic. It also worth noting that large changes on conductance state, appeared in the neuron ‘firing’ process, pose severe challenges to device
endurance. While the sparse activation demonstrated with the proposed framework
would greatly improve the system lifetime.
Although NVM devices offer attractive prospects for energy efficient, distributed
and parallel neuromorphic computing systems, large-scale system implementation
with NVM devices are still lacking. Substantial research efforts are demanded to
overcome the engineering challenges, including device variability and non-ideality,
power dissipation, limited conductance dynamic range and asymmetric response of
conductance[10].
References
1. G. Guo, S.Z. Li, Content-based audio classification and retrieval by support vector machines.
IEEE Trans. Neural Networks 14(1), 209–215 (2003). Jan
2. A. Rabaoui, M. Davy, S. Rossignol, N. Ellouze, Using one-class SVMS and wavelets for audio
surveillance. IEEE Trans. Inf. Forensics Secur. 3(4), 763–775 (2008). Dec
3. J. Dennis, H.D. Tran, H. Li, Spectrogram image feature for sound event classification in mismatched conditions. IEEE Signal Process. Lett. 18(2), 130–133 (2011). Feb
4. C. Kwak, O.W. Kwon, Cardiac disorder classification by heart sound signals using murmur
likelihood and hidden Markov model state likelihood. IET Signal Proc. 6(4), 326–334 (2012).
June
5. R.V. Sharan, T.J. Moir, An overview of applications and advancements in automatic sound
recognition. Neurocomputing 200, 22–34 (2016)
6. P.A. Merolla, J.V. Arthur, R. Alvarez-Icaza, A.S. Cassidy, J. Sawada, F. Akopyan, B.L. Jackson,
N. Imam, C. Guo, Y. Nakamura, B. Brezzo, I. Vo, S.K. Esser, R. Appuswamy, B. Taba, A. Amir,
M.D. Flickner, W.P. Risk, R. Manohar, D.S. Modha, A million spiking-neuron integrated circuit
with a scalable communication network and interface. Science 345(6197), 668–673 (2014)
7. F.F. Li, R. Fergus, P. Perona, One-shot learning of object categories. IEEE Trans. Pattern Anal.
Mach. Intell. 28(4), 594–611 (2006). April
435
the SNN classifier, a fully SNN-based framework when implemented with NVM
devices would translate to significant power saving and speed up.
Besides emulating synaptic functionalities, several works also explored using
NVM to emulate the neuronal dynamics, including membrane potentials maintenance, transient dynamic, spike generation and transmission [62–65]. Neuromorphic systems implemented fully with PCM have been reported in [64, 65], in which
the neurons are integrated seamlessly with dense synaptic arrays using PCM. For
NVM implemented synapses, each conductance state of an NVM is important and
impacts the synaptic integration performed at the post-synaptic neurons. In contrast, the requirements for continuum and linearity of conductance states are less
stringent when implementing leaky integrate-and-fire neurons. The biological neuron fires whenever the membrane potential crosses the firing threshold, hence, it
is sufficient that accumulated electrical pulses trigger the ‘firing’ after receiving a
certain number of pulses, without explicit conductance update for every electrical
pulse. Furthermore, the non-volatility property of certain materials is beneficial to
implement the leaky dynamic. It also worth noting that large changes on conductance state, appeared in the neuron ‘firing’ process, pose severe challenges to device
endurance. While the sparse activation demonstrated with the proposed framework
would greatly improve the system lifetime.
Although NVM devices offer attractive prospects for energy efficient, distributed
and parallel neuromorphic computing systems, large-scale system implementation
with NVM devices are still lacking. Substantial research efforts are demanded to
overcome the engineering challenges, including device variability and non-ideality,
power dissipation, limited conductance dynamic range and asymmetric response of
conductance[10].
References
1. G. Guo, S.Z. Li, Content-based audio classification and retrieval by support vector machines.
IEEE Trans. Neural Networks 14(1), 209–215 (2003). Jan
2. A. Rabaoui, M. Davy, S. Rossignol, N. Ellouze, Using one-class SVMS and wavelets for audio
surveillance. IEEE Trans. Inf. Forensics Secur. 3(4), 763–775 (2008). Dec
3. J. Dennis, H.D. Tran, H. Li, Spectrogram image feature for sound event classification in mismatched conditions. IEEE Signal Process. Lett. 18(2), 130–133 (2011). Feb
4. C. Kwak, O.W. Kwon, Cardiac disorder classification by heart sound signals using murmur
likelihood and hidden Markov model state likelihood. IET Signal Proc. 6(4), 326–334 (2012).
June
5. R.V. Sharan, T.J. Moir, An overview of applications and advancements in automatic sound
recognition. Neurocomputing 200, 22–34 (2016)
6. P.A. Merolla, J.V. Arthur, R. Alvarez-Icaza, A.S. Cassidy, J. Sawada, F. Akopyan, B.L. Jackson,
N. Imam, C. Guo, Y. Nakamura, B. Brezzo, I. Vo, S.K. Esser, R. Appuswamy, B. Taba, A. Amir,
M.D. Flickner, W.P. Risk, R. Manohar, D.S. Modha, A million spiking-neuron integrated circuit
with a scalable communication network and interface. Science 345(6197), 668–673 (2014)
7. F.F. Li, R. Fergus, P. Perona, One-shot learning of object categories. IEEE Trans. Pattern Anal.
Mach. Intell. 28(4), 594–611 (2006). April