Hungarian Layer: Logics Empowered Neural Architecture
This addresses the problem of integrating logical algorithms into neural networks for researchers in AI and NLP, representing a novel method rather than an incremental improvement.
The paper tackles the limitation of neural architectures lacking logic flow by introducing a dynamic calculus graph guided by logic, enabling the embedding of traditional algorithms like the Hungarian algorithm into neural networks. For sentence matching, it reformulates the problem as task-assignment and shows substantial outperformance over state-of-the-art baselines in experiments.
Neural architecture is a purely numeric framework, which fits the data as a continuous function. However, lacking of logic flow (e.g. \textit{if, for, while}), traditional algorithms (e.g. \textit{Hungarian algorithm, A$^*$ searching, decision tress algorithm}) could not be embedded into this paradigm, which limits the theories and applications. In this paper, we reform the calculus graph as a dynamic process, which is guided by logic flow. Within our novel methodology, traditional algorithms could empower numerical neural network. Specifically, regarding the subject of sentence matching, we reformulate this issue as the form of task-assignment, which is solved by Hungarian algorithm. First, our model applies BiLSTM to parse the sentences. Then Hungarian layer aligns the matching positions. Last, we transform the matching results for soft-max regression by another BiLSTM. Extensive experiments show that our model outperforms other state-of-the-art baselines substantially.