Action Recognition Using Supervised Spiking Neural Networks
The main idea and innovation: using pulsed neural networks Spiking neural network, the motion residual plots for pattern classification
And advantages
97.2% recognition accuracy
Knowledge Point
Pulse neural network is a kind of imitation human neural stimulation, there is a stimulation threshold, only above this threshold will activate the current neurons.
Own thinking
Reference blog
- DESCRIPTION pulse neural network SNN: Neural Network Next Generation
- Spiking neural network learning notes (review)
Code and data
Code: https://github.com/ArefMq/action-recognition-via-snn/blob/master/SNN-test1.ipynb
Data: http://research.ibm.com/dvsgesture/
Guided weak supervision for action recognition with scarce data to assess skills
of children with autism
The main idea and innovation: weak supervised learning, combined with maximum likelihood estimation, to achieve the Video category.
And advantages
Autism dataset Autism Dataset 75.1% recognition accuracy
Knowledge Point
Weak supervision, the maximum likelihood estimate
Own thinking