UCLA

The spiking neural networks (SNNs) are often described as the "third generation" of neural networks, and they are expected to improve the existing deep neural networks. Recent advancements of SNNs mainly focused on processing and learning the visual signals, while SNNs' potential in classifying non-image data is rarely tested. In this thesis, we extended the functionality of BindsNET, a popular SNN simulation software, to allow it to process and classify non-image data. We built an SNN that can efficiently classify the embedding data of 51,278 online communities ("subreddits") on Reddit.com in an unsupervised fashion. With the classification result, we further analyzed the social network structure of the subreddit clusters of video games, using the exponential-family random graph model (ERGM). We discovered that communities of the same video game genre or same platform are more likely to be hostile towards each other. The number of subscribers and the availability of online mode are also significant factors in the hostility of a subreddit.