Timezone: »

Kernelized Bayesian Softmax for Text Generation
Ning Miao · Hao Zhou · Chengqi Zhao · Wenxian Shi · Lei Li

Thu Dec 12 05:00 PM -- 07:00 PM (PST) @ East Exhibition Hall B + C #136

Neural models for text generation require a softmax layer with proper token embeddings during the decoding phase. Most existing approaches adopt single point embedding for each token. However, a word may have multiple senses according to different context, some of which might be distinct. In this paper, we propose KerBS, a novel approach for learning better embeddings for text generation. KerBS embodies two advantages: (a) it employs a Bayesian composition of embeddings for words with multiple senses; (b) it is adaptive to semantic variances of words and robust to rare sentence context by imposing learned kernels to capture the closeness of words (senses) in the embedding space. Empirical studies show that KerBS significantly boosts the performance of several text generation tasks.

Author Information

Ning Miao (ByteDance AI Lab)
Hao Zhou (Bytedance)
Chengqi Zhao (Bytedance)
Wenxian Shi (Bytedance)
Lei Li (ByteDance)

Research Scientist at ByteDance AI Lab

More from the Same Authors