Limits...
Robustness of spiking Deep Belief Networks to noise and reduced bit precision of neuro-inspired hardware platforms.

Stromatias E, Neil D, Pfeiffer M, Galluppi F, Furber SB, Liu SC - Front Neurosci (2015)

Bottom Line: Increasingly large deep learning architectures, such as Deep Belief Networks (DBNs) are the focus of current machine learning research and achieve state-of-the-art results in different domains.The on-going work on design and construction of spike-based hardware platforms offers an alternative for running deep neural networks with significantly lower power consumption, but has to overcome hardware limitations in terms of noise and limited weight precision, as well as noise inherent in the sensor signal.Our results demonstrate that spiking DBNs can tolerate very low levels of hardware bit precision down to almost two bits, and show that their performance can be improved by at least 30% through an adapted training mechanism that takes the bit precision of the target platform into account.

View Article: PubMed Central - PubMed

Affiliation: Advanced Processor Technologies Group, School of Computer Science, University of Manchester Manchester, UK.

ABSTRACT
Increasingly large deep learning architectures, such as Deep Belief Networks (DBNs) are the focus of current machine learning research and achieve state-of-the-art results in different domains. However, both training and execution of large-scale Deep Networks require vast computing resources, leading to high power requirements and communication overheads. The on-going work on design and construction of spike-based hardware platforms offers an alternative for running deep neural networks with significantly lower power consumption, but has to overcome hardware limitations in terms of noise and limited weight precision, as well as noise inherent in the sensor signal. This article investigates how such hardware constraints impact the performance of spiking neural network implementations of DBNs. In particular, the influence of limited bit precision during execution and training, and the impact of silicon mismatch in the synaptic weight parameters of custom hybrid VLSI implementations is studied. Furthermore, the network performance of spiking DBNs is characterized with regard to noise in the spiking input signal. Our results demonstrate that spiking DBNs can tolerate very low levels of hardware bit precision down to almost two bits, and show that their performance can be improved by at least 30% through an adapted training mechanism that takes the bit precision of the target platform into account. Spiking DBNs thus present an important use-case for large-scale hybrid analog-digital or digital neuromorphic platforms such as SpiNNaker, which can execute large but precision-constrained deep networks in real time.

No MeSH data available.


Weight distributions for different bit precision levels and DBN layers. Each row represents different fixed-point weight precisions, while each column represents a layer of the DBN, starting from Layer 1 (left), Layer 2 (middle) to the Output Layer (right). Despite the different discretization levels, the overall shape of the weight distribution is conserved.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4496577&req=5

Figure 5: Weight distributions for different bit precision levels and DBN layers. Each row represents different fixed-point weight precisions, while each column represents a layer of the DBN, starting from Layer 1 (left), Layer 2 (middle) to the Output Layer (right). Despite the different discretization levels, the overall shape of the weight distribution is conserved.

Mentions: There are a few tools that can be employed to investigate how the distribution of the reduced bit precision weights nonetheless manages to maintain a substantial amount of the network's classification performance. Firstly, the initial question is to investigate whether this reduction in bit precision qualitatively maintains the same weight distribution as the original. Figure 5 shows that the quasi-continuous distribution of weights obtained for double-precision becomes increasingly discretized as the precision f decreases. In the extreme case of a Q3.1 representation, the weight values are quantized to ± 0.5, ± 1, and 0, but nonetheless seem to reflect the shape of the original distribution.


Robustness of spiking Deep Belief Networks to noise and reduced bit precision of neuro-inspired hardware platforms.

Stromatias E, Neil D, Pfeiffer M, Galluppi F, Furber SB, Liu SC - Front Neurosci (2015)

Weight distributions for different bit precision levels and DBN layers. Each row represents different fixed-point weight precisions, while each column represents a layer of the DBN, starting from Layer 1 (left), Layer 2 (middle) to the Output Layer (right). Despite the different discretization levels, the overall shape of the weight distribution is conserved.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4496577&req=5

Figure 5: Weight distributions for different bit precision levels and DBN layers. Each row represents different fixed-point weight precisions, while each column represents a layer of the DBN, starting from Layer 1 (left), Layer 2 (middle) to the Output Layer (right). Despite the different discretization levels, the overall shape of the weight distribution is conserved.
Mentions: There are a few tools that can be employed to investigate how the distribution of the reduced bit precision weights nonetheless manages to maintain a substantial amount of the network's classification performance. Firstly, the initial question is to investigate whether this reduction in bit precision qualitatively maintains the same weight distribution as the original. Figure 5 shows that the quasi-continuous distribution of weights obtained for double-precision becomes increasingly discretized as the precision f decreases. In the extreme case of a Q3.1 representation, the weight values are quantized to ± 0.5, ± 1, and 0, but nonetheless seem to reflect the shape of the original distribution.

Bottom Line: Increasingly large deep learning architectures, such as Deep Belief Networks (DBNs) are the focus of current machine learning research and achieve state-of-the-art results in different domains.The on-going work on design and construction of spike-based hardware platforms offers an alternative for running deep neural networks with significantly lower power consumption, but has to overcome hardware limitations in terms of noise and limited weight precision, as well as noise inherent in the sensor signal.Our results demonstrate that spiking DBNs can tolerate very low levels of hardware bit precision down to almost two bits, and show that their performance can be improved by at least 30% through an adapted training mechanism that takes the bit precision of the target platform into account.

View Article: PubMed Central - PubMed

Affiliation: Advanced Processor Technologies Group, School of Computer Science, University of Manchester Manchester, UK.

ABSTRACT
Increasingly large deep learning architectures, such as Deep Belief Networks (DBNs) are the focus of current machine learning research and achieve state-of-the-art results in different domains. However, both training and execution of large-scale Deep Networks require vast computing resources, leading to high power requirements and communication overheads. The on-going work on design and construction of spike-based hardware platforms offers an alternative for running deep neural networks with significantly lower power consumption, but has to overcome hardware limitations in terms of noise and limited weight precision, as well as noise inherent in the sensor signal. This article investigates how such hardware constraints impact the performance of spiking neural network implementations of DBNs. In particular, the influence of limited bit precision during execution and training, and the impact of silicon mismatch in the synaptic weight parameters of custom hybrid VLSI implementations is studied. Furthermore, the network performance of spiking DBNs is characterized with regard to noise in the spiking input signal. Our results demonstrate that spiking DBNs can tolerate very low levels of hardware bit precision down to almost two bits, and show that their performance can be improved by at least 30% through an adapted training mechanism that takes the bit precision of the target platform into account. Spiking DBNs thus present an important use-case for large-scale hybrid analog-digital or digital neuromorphic platforms such as SpiNNaker, which can execute large but precision-constrained deep networks in real time.

No MeSH data available.