MnnFast: A Fast and Scalable System Architecture for Memory-Augmented Neural Networks | IEEE Conference Publication | IEEE Xplore
Recently, neural networks have risen as new information processing paradigms in various fields. Among them, newly emerging neural networks called memory-augmented neural networks (MemNNs) [69], [78] are getting increasing attention from the researchers, thanks to their powerful context-aware information processing capability. In contrast to feedforward neural networks (e.g., CNNs, DNNs), MemNNs, which exploit their dedicated memory components to process sequences of inputs, are known for their powerful reasoning capability. Different from other neural networks, MemNN can discretely read and write all contents by exploiting a flexible mechanism similar to the human's working memory [44]. (a) Shows an example story with a question. (b) Shows where the memory network stores the story, and how it processes the question to derive the answer. Show All Figure 1 shows how MemNN processes a story with a question and an answer. In this example, MemNN first receives a four-sentence story and stor
Explore this link on the map →