Efficient Time Series Classification using Spiking Reservoir

Dey, Sounak; Banerjee, Dighanchal; George, Arun M.; Mukherjee, Arijit; Pal, Arpan

doi:10.1109/ijcnn55064.2022.9892728

Cited by 3 publications

(10 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We also mention the State-of-The-Art (SoTA) results obtained with the non-spiking methods [e.g., LSTM-FCN (Karim et al, 2017) and HIVE-COTE (Lines et al, 2018)] for completeness, but compare ours with only the other spiking results for fairness. As can be seen in the Table 5, the Max acc results obtained with the LSNN model completely outperforms the latest spiking results on the Wafer, Ford-A, Ford-B, and the Earthquakes dataset reported by Dey et al (2022). In fact, with the LSNN model, considering the Max acc results, we get an improvement of 0.668%, 16.412%, 28.607%, and 11.802% in classification accuracy (over Dey et al, 2022) for Wafer, Ford-A, Ford-B, and Earthquakes datasets, respectively.…”

Section: Lsnn Model's Results Analysismentioning

confidence: 75%

“…As can be seen in the Table 5, the Max acc results obtained with the LSNN model completely outperforms the latest spiking results on the Wafer, Ford-A, Ford-B, and the Earthquakes dataset reported by Dey et al (2022). In fact, with the LSNN model, considering the Max acc results, we get an improvement of 0.668%, 16.412%, 28.607%, and 11.802% in classification accuracy (over Dey et al, 2022) for Wafer, Ford-A, Ford-B, and Earthquakes datasets, respectively. On the ECG5000 dataset, the LSNN model obtains a maximum accuracy of 98.49% which interestingly outperforms the current SoTA 98.43% obtained by a non-spiking model (Pereira and Silveira, 2019), although, by a small margin.…”

Section: Lsnn Model's Results Analysismentioning

confidence: 75%

“…Considering the Max acc and Mean acc accuracy results obtained with the LSNN nhdn model (where the network is still spiking, but with no HDN layer), we see that they too outperform the results obtained by Dey et al (2022), with 0.536%, 11.509%, 20.927%, and 10.801% improvement (w.r.t. Max acc results under LSNN nhdn ) on Wafer, Ford-A, Ford-B, and Earthquakes datasets, respectively.…”

Section: Lsnn Model's Results Analysismentioning

confidence: 84%

“…Next, we revisit the Table 5 and make a subtle remark with respect to the LSNN's results on the Ford-B dataset. We see that authors (Dey et al, 2022) achieve 64.32% accuracy on Ford-B with an LSM based model, where unlike the training data, the test samples are noisy. However, the LSNN model and its derivatives achieve far better accuracy on Ford-B (28.607% max improvement).…”

Section: Discussion On Energy Consumptionmentioning

confidence: 89%

“…stands for accuracy; Maxacc and Meanacc stand for the maximum accuracy and mean of the maximum accuracies (mean ± std) over 3 runs. We compare our results with only the SoTA spiking ones (Dey et al, 2022). The italicized values indicate accuracy values from non-spiking models.…”

Section: Slrc Model's Results Analysismentioning

confidence: 99%

See 4 more Smart Citations

Reservoir based spiking models for univariate Time Series Classification

Gaurav

Stewart

2023

Front. Comput. Neurosci.

View full text Add to dashboard Cite

A variety of advanced machine learning and deep learning algorithms achieve state-of-the-art performance on various temporal processing tasks. However, these methods are heavily energy inefficient—they run mainly on the power hungry CPUs and GPUs. Computing with Spiking Networks, on the other hand, has shown to be energy efficient on specialized neuromorphic hardware, e.g., Loihi, TrueNorth, SpiNNaker, etc. In this work, we present two architectures of spiking models, inspired from the theory of Reservoir Computing and Legendre Memory Units, for the Time Series Classification (TSC) task. Our first spiking architecture is closer to the general Reservoir Computing architecture and we successfully deploy it on Loihi; the second spiking architecture differs from the first by the inclusion of non-linearity in the readout layer. Our second model (trained with Surrogate Gradient Descent method) shows that non-linear decoding of the linearly extracted temporal features through spiking neurons not only achieves promising results, but also offers low computation-overhead by significantly reducing the number of neurons compared to the popular LSM based models—more than 40x reduction with respect to the recent spiking model we compare with. We experiment on five TSC datasets and achieve new SoTA spiking results (—as much as 28.607% accuracy improvement on one of the datasets), thereby showing the potential of our models to address the TSC tasks in a green energy-efficient manner. In addition, we also do energy profiling and comparison on Loihi and CPU to support our claims.

show abstract