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.


Impact of different rounding methods during training on network performance with reduced weight bit precision. (A) Effectiveness of the dual-copy rounding weight training paradigm. Training at full precision and later rounding performs consistently worse than the dual-copy rounding method introduced in this paper. Rounding the weights during training can prevent learning entirely at low-precision regimes. The results show averages of five independent runs with different random seeds. (B) Increase in classification accuracy of a spiking DBN with Q3.1 precision weights due to the dual-copy rounding method for input rates of 100 and 1500 Hz. Results over four trials.
© Copyright Policy
Related In: Results  -  Collection

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

Figure 9: Impact of different rounding methods during training on network performance with reduced weight bit precision. (A) Effectiveness of the dual-copy rounding weight training paradigm. Training at full precision and later rounding performs consistently worse than the dual-copy rounding method introduced in this paper. Rounding the weights during training can prevent learning entirely at low-precision regimes. The results show averages of five independent runs with different random seeds. (B) Increase in classification accuracy of a spiking DBN with Q3.1 precision weights due to the dual-copy rounding method for input rates of 100 and 1500 Hz. Results over four trials.

Mentions: As there was no performance loss in the Q3.12 representation compared to full double-precision, this was taken as the full-precision reference point. Figure 9 shows the effect of the three investigated training methods on the classification accuracy, when testing the weight matrix at different levels of bit precision. Rounding a high-precision weight matrix does work effectively, but can fail for lower-precision weights. Unfortunately, the iterative rounding method of training works extremely poorly for low-precision cases; the weight update size is simply less than the precision of the weight, so learning halts entirely after the error gradient falls below a certain threshold.


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)

Impact of different rounding methods during training on network performance with reduced weight bit precision. (A) Effectiveness of the dual-copy rounding weight training paradigm. Training at full precision and later rounding performs consistently worse than the dual-copy rounding method introduced in this paper. Rounding the weights during training can prevent learning entirely at low-precision regimes. The results show averages of five independent runs with different random seeds. (B) Increase in classification accuracy of a spiking DBN with Q3.1 precision weights due to the dual-copy rounding method for input rates of 100 and 1500 Hz. Results over four trials.
© Copyright Policy
Related In: Results  -  Collection

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

Figure 9: Impact of different rounding methods during training on network performance with reduced weight bit precision. (A) Effectiveness of the dual-copy rounding weight training paradigm. Training at full precision and later rounding performs consistently worse than the dual-copy rounding method introduced in this paper. Rounding the weights during training can prevent learning entirely at low-precision regimes. The results show averages of five independent runs with different random seeds. (B) Increase in classification accuracy of a spiking DBN with Q3.1 precision weights due to the dual-copy rounding method for input rates of 100 and 1500 Hz. Results over four trials.
Mentions: As there was no performance loss in the Q3.12 representation compared to full double-precision, this was taken as the full-precision reference point. Figure 9 shows the effect of the three investigated training methods on the classification accuracy, when testing the weight matrix at different levels of bit precision. Rounding a high-precision weight matrix does work effectively, but can fail for lower-precision weights. Unfortunately, the iterative rounding method of training works extremely poorly for low-precision cases; the weight update size is simply less than the precision of the weight, so learning halts entirely after the error gradient falls below a certain threshold.

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.