Low-power neuromorphic circuits for unsupervised spike based learning
This article introduces a novel multi-layer Winner- Take-All (ML-WTA) spiking neural network (SNN) architecture using neurons with nonlinear dendrites and binary synapses. The network is trained by an unsupervised spike based learning rule that modifies the network connections. Inspired by the multi...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project (FYP) |
| Language: | English |
| Published: |
2016
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/68169 |
| _version_ | 1811685040432611328 |
|---|---|
| author | He, Tong |
| author2 | Arindam Basu |
| author_facet | Arindam Basu He, Tong |
| author_sort | He, Tong |
| collection | NTU |
| description | This article introduces a novel multi-layer Winner- Take-All (ML-WTA) spiking neural network (SNN) architecture using neurons with nonlinear dendrites and binary synapses. The network is trained by an unsupervised spike based learning rule that modifies the network connections. Inspired by the multi-layer models of human visual cortex and chunking learning, the proposed architecture contains multiple layers of neurons for learning the data by sequence. I show that if I increase the synaptic neuron time constant of the layers of the system in succession, the ML-WTA network is capable of inspecting the incoming patterns for a longer duration of time before providing a decision. Moreover, the decision could be made near the end of whole pattern with the aid of adaptive threshold mechanism. After the training is complete, a unique neuron of the last layer emits a spike for the same class of patterns. The results of three different benchmarks discussed in this article show that the proposed structural plasticity based WTA network is capable of classifying Poisson spike trains and the two layer structure has better performance in all three different benchmarks when sufficient neurons are employed. |
| first_indexed | 2024-10-01T04:38:12Z |
| format | Final Year Project (FYP) |
| id | ntu-10356/68169 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T04:38:12Z |
| publishDate | 2016 |
| record_format | dspace |
| spelling | ntu-10356/681692023-07-07T15:58:47Z Low-power neuromorphic circuits for unsupervised spike based learning He, Tong Arindam Basu School of Electrical and Electronic Engineering DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Biometrics This article introduces a novel multi-layer Winner- Take-All (ML-WTA) spiking neural network (SNN) architecture using neurons with nonlinear dendrites and binary synapses. The network is trained by an unsupervised spike based learning rule that modifies the network connections. Inspired by the multi-layer models of human visual cortex and chunking learning, the proposed architecture contains multiple layers of neurons for learning the data by sequence. I show that if I increase the synaptic neuron time constant of the layers of the system in succession, the ML-WTA network is capable of inspecting the incoming patterns for a longer duration of time before providing a decision. Moreover, the decision could be made near the end of whole pattern with the aid of adaptive threshold mechanism. After the training is complete, a unique neuron of the last layer emits a spike for the same class of patterns. The results of three different benchmarks discussed in this article show that the proposed structural plasticity based WTA network is capable of classifying Poisson spike trains and the two layer structure has better performance in all three different benchmarks when sufficient neurons are employed. Bachelor of Engineering 2016-05-24T07:41:47Z 2016-05-24T07:41:47Z 2016 Final Year Project (FYP) http://hdl.handle.net/10356/68169 en Nanyang Technological University 60 p. application/pdf |
| spellingShingle | DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Biometrics He, Tong Low-power neuromorphic circuits for unsupervised spike based learning |
| title | Low-power neuromorphic circuits for unsupervised spike based learning |
| title_full | Low-power neuromorphic circuits for unsupervised spike based learning |
| title_fullStr | Low-power neuromorphic circuits for unsupervised spike based learning |
| title_full_unstemmed | Low-power neuromorphic circuits for unsupervised spike based learning |
| title_short | Low-power neuromorphic circuits for unsupervised spike based learning |
| title_sort | low power neuromorphic circuits for unsupervised spike based learning |
| topic | DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Biometrics |
| url | http://hdl.handle.net/10356/68169 |
| work_keys_str_mv | AT hetong lowpowerneuromorphiccircuitsforunsupervisedspikebasedlearning |