Sparser spiking activity can be better: Feature Refine-and-Mask spiking neural network for event-based visual recognition

Yao, Man; Zhang, Hengyu; Zhao, Guiping; Wang, Dingheng; Cao, Gang; Li, Guoqi

doi:10.1016/j.neunet.2023.07.008

Cited by 10 publications

(2 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Inspired by this, the basic idea of the proposed dynamic framework is to modify the spiking response by optimizing the membrane potential because spiking neurons determine whether to fire based on whether the membrane potential exceeds a threshold. In our previous work 25 , 27 – 29 , we implemented the attention through additional plug-and-play modules, including independent or coupled three dimensions of temporal, channel, and spatial, to learn “when”, “what” and “where” to focus on. These attention modules first capture global information of different dimensions and then use them to model the relative importance between different input moments, channels, or locations.…”

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Spike-based dynamic computing with asynchronous sensing-computing neuromorphic chip

Yao,

Richter,

Zhao

et al. 2024

Nat Commun

Self Cite

View full text Add to dashboard Cite

By mimicking the neurons and synapses of the human brain and employing spiking neural networks on neuromorphic chips, neuromorphic computing offers a promising energy-efficient machine intelligence. How to borrow high-level brain dynamic mechanisms to help neuromorphic computing achieve energy advantages is a fundamental issue. This work presents an application-oriented algorithm-software-hardware co-designed neuromorphic system for this issue. First, we design and fabricate an asynchronous chip called “Speck”, a sensing-computing neuromorphic system on chip. With the low processor resting power of 0.42mW, Speck can satisfy the hardware requirements of dynamic computing: no-input consumes no energy. Second, we uncover the “dynamic imbalance” in spiking neural networks and develop an attention-based framework for achieving the algorithmic requirements of dynamic computing: varied inputs consume energy with large variance. Together, we demonstrate a neuromorphic system with real-time power as low as 0.70mW. This work exhibits the promising potentials of neuromorphic computing with its asynchronous event-driven, sparse, and dynamic nature.

show abstract

Section: Resultsmentioning

confidence: 99%

“…To facilitate the deployment of attention SNNs on neuromorphic chips, we summarize existing methods 25 , 27 – 29 into a general attention-based dynamic framework. Figure 3 b, c show the LIF-SNN layer and dynamic SNN architecture, respectively.…”

Section: Resultsmentioning

confidence: 99%