Deep Learning with PAI: a Case Study of AliMe
We introduce a machine learning platform in Alibaba called Platform of AI (PAI) and its application in AliMe assistant bot. PAI deep learning (DL) framework uses Caffe and TensorFlow as our base software, and supports training and deploying rich DL algorithms in distributed CPUs and GPUs. We have implemented several optimizations to ensure efficiency in distributed mode of PAI DL. With our optimization techniques, some model achieves the fastest performance in distributed mode so far as we know, and has over 10X speed-up against the default distributed implementation. Furthermore, we showcase an application of PAI DL in building the intention mining and open-domain conversation modules in AliMe assistant bot. Results show our modules have very good performance and scale up well for industrial applications.
Minghui Qiu is a senior algorithm engineer in Alibaba, working on deep learning and reinforcement learning for many NLP tasks. He hold a PhD degree in Singapore Management University. He has published more than 20 papers in top journals and conferences, including a co-authored paper that was a runner-up for the best paper award in SocInfo’13. He has been selected to participate in the Carnegie Mellon University Exchange Program. He did an internship in Google Mountain View, working with Google Strategic Technologies team in summer 2014.