A Comparison of Neuromorphic Classification Tasks

Reynolds, John J.; Plank, James S.; Schuman, Catherine D.; Bruer, Grant; Disney, Adam; Dean, Mark E.; Rose, Garrett S.

doi:10.1145/3229884.3229896

Cited by 10 publications

(5 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…[ 78 ] Thus, by extracting pertinent features, researchers can gain a more profound understanding of the brain's underlying activities. [ 79 ] The primary objective of this study is to demonstrate the effectiveness of our memcapacitor‐based RC system capable to solve classification problems in real‐time.…”

Section: Resultsmentioning

confidence: 99%

Biomembrane‐Based Memcapacitive Reservoir Computing System for Energy‐Efficient Temporal Data Processing

Hossain,

Mohamed,

Armendarez

et al. 2023

Advanced Intelligent Systems

View full text Add to dashboard Cite

Reservoir computing is a highly efficient machine learning framework for processing temporal data by extracting input features and mapping them into higher dimensional spaces. Physical reservoirs have been realized using spintronic oscillators, atomic switch networks, volatile memristors, etc. However, these devices are intrinsically energy‐dissipative due to their resistive nature, increasing their power consumption. Therefore, memcapacitive devices can provide a more energy‐efficient approach. Herein, volatile biomembrane‐based memcapacitors are leveraged as reservoirs to solve classification tasks and process time series in simulation and experimentally. This system achieves a 99.6% accuracy for spoken‐digit classification and a normalized mean square error of in a second‐order nonlinear regression task. Furthermore, to showcase the device's real‐time temporal data processing capability, a 100% accuracy for an epilepsy detection problem is achieved. Most importantly, it is demonstrated that each memcapacitor consumes an average of 41.5 fJ of energy per spike, regardless of the selected input voltage pulse width, while maintaining an average power of 415 fW for a pulse width of 100 ms, orders of magnitude lower than those achieved by state‐of‐the‐art devices. Lastly, it is believed that the biocompatible, soft nature of our memcapacitor renders it highly suitable for computing applications in biological environments.

show abstract

Section: Resultsmentioning

confidence: 99%

Biomembrane‐Based Memcapacitive Reservoir Computing System for Energy‐Efficient Temporal Data Processing

Hossain,

Mohamed,

Armendarez

et al. 2023

Advanced Intelligent Systems

View full text Add to dashboard Cite

show abstract

“…RoboNav is an autonomous navigation system for robotic applications and is meant to be deployed on a specific robot (Mitchell et al, 2017 ). We also used the Iris (Dua and Graff, 2017 ) and Radio (Reynolds et al, 2018 ) datasets for classification tasks. The former is a multivariate dataset of 50 samples from each of three species of the Iris flower, and the latter is a satellite radio signal classification problem.…”

Section: Methodology and Experimental Setupmentioning

confidence: 99%

“…For Spiking Neural Networks (SNNs), we consider both digital and mixed-signal hardware; DANNA2 (Mitchell et al, 2018), and mrDANNA (Chakma et al, 2017), respectively. Additionally, we select Pole-balance (Wieland, 1991;Gomez et al, 2006), and RoboNav (Mitchell et al, 2017) for experiments on control applications, and IRIS (Dua and Graff, 2017), and Radio (Reynolds et al, 2018) dataset for classification applications. In Figure 3, the experimental setup for ANN is shown in red, and for SNN in blue.…”

Section: Methodsmentioning

confidence: 99%

“…For artificial neural networks, we validated our approach using AlexNet (Krizhevsky et al, 2012) and VGG19 (Simonyan and Zisserman, 2014) on a Programmable Ultra-Efficient Memristor-based Accelerator (PUMA, Ankit et al, 2019). For spiking neuromorphic systems, we considered several control and classification tasks such as the canonical pole balancing (Gomez et al, 2006), autonomous robotic navigation (Mitchell et al, 2017), satellite radio signal classification (Reynolds et al, 2018), and Iris dataset classification (Dua and Graff, 2017) on both digital and mixed-signal memristor-based accelerators as the underlying hardware (Chakma et al, 2017;Mitchell et al, 2018;Plank et al, 2018). Hierarchical-PABO predicts the Pareto frontier for a three-objective (network performance, the accelerator's energy efficiency, and area) optimization with relatively few evaluations.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Bayesian Multi-objective Hyperparameter Optimization for Accurate, Fast, and Efficient Neural Network Accelerator Design

et al. 2020

Self Cite

View full text Add to dashboard Cite

In resource-constrained environments, such as low-power edge devices and smart sensors, deploying a fast, compact, and accurate intelligent system with minimum energy is indispensable. Embedding intelligence can be achieved using neural networks on neuromorphic hardware. Designing such networks would require determining several inherent hyperparameters. A key challenge is to find the optimum set of hyperparameters that might belong to the input/output encoding modules, the neural network itself, the application, or the underlying hardware. In this work, we present a hierarchical pseudo agent-based multi-objective Bayesian hyperparameter optimization framework (both software and hardware) that not only maximizes the performance of the network, but also minimizes the energy and area requirements of the corresponding neuromorphic hardware. We validate performance of our approach (in terms of accuracy and computation speed) on several control and classification applications on digital and mixed-signal (memristor-based) neural accelerators. We show that the optimum set of hyperparameters might drastically improve the performance of one application (i.e., 52–71% for Pole-Balance), while having minimum effect on another (i.e., 50–53% for RoboNav). In addition, we demonstrate resiliency of different input/output encoding, training neural network, or the underlying accelerator modules in a neuromorphic system to the changes of the hyperparameters.

show abstract

“…In this work we leverage an evolutionary optimization framework called EONS [13], [14] to train spiking neural networks. This evolutionary approach has been shown to lead to well-performing SNNs, as demonstrated on several machine learning and control tasks [6], [15], [16]. At the core of the evolutionary (genetic) algorithm is the fitness function which evaluates every network in a population.…”

Section: Introductionmentioning

confidence: 99%

Multi-Objective Optimization for Size and Resilience of Spiking Neural Networks

Dimovska

Johnston

Schuman

et al. 2019

2019 IEEE 10th Annual Ubiquitous Computing, Electronics &Amp; Mobile Communication Conference (UEMCON)

Self Cite

View full text Add to dashboard Cite

Inspired by the connectivity mechanisms in the brain, neuromorphic computing architectures model Spiking Neural Networks (SNNs) in silicon. As such, neuromorphic architectures are designed and developed with the goal of having small, low power chips that can perform control and machine learning tasks. However, the power consumption of the developed hardware can greatly depend on the size of the network that is being evaluated on the chip. Furthermore, the accuracy of a trained SNN that is evaluated on chip can change due to voltage and current variations in the hardware that perturb the learned weights of the network. While efforts are made on the hardware side to minimize those perturbations, a software based strategy to make the deployed networks more resilient can help further alleviate that issue. In this work, we study Spiking Neural Networks in two neuromorphic architecture implementations with the goal of decreasing their size, while at the same time increasing their resiliency to hardware faults. We leverage an evolutionary algorithm to train the SNNs and propose a multiobjective fitness function to optimize the size and resiliency of the SNN. We demonstrate that this strategy leads to well-performing, small-sized networks that are more resilient to hardware faults.

show abstract

A Comparison of Neuromorphic Classification Tasks

Cited by 10 publications

References 18 publications

Biomembrane‐Based Memcapacitive Reservoir Computing System for Energy‐Efficient Temporal Data Processing

Biomembrane‐Based Memcapacitive Reservoir Computing System for Energy‐Efficient Temporal Data Processing

Bayesian Multi-objective Hyperparameter Optimization for Accurate, Fast, and Efficient Neural Network Accelerator Design

Multi-Objective Optimization for Size and Resilience of Spiking Neural Networks

Contact Info

Product

Resources

About