No abstract

No abstract

No abstract

Increasing demand for on-device Automatic Speech Recognition (ASR) systems has resulted in renewed interests in developing automatic model compression techniques. Past research have shown that AutoML-based Low Rank Factorization (LRF) technique, when applied to an end-to-end Encoder-Attention-Decoder style ASR model, can achieve a speedup of up to 3.7×, outperforming laborious manual rank-selection approaches. However, we show that current AutoML-based search techniques only work up to a certain compression level, beyond which they fail to produce compressed models with acceptable word error rates (WER). In this work, we propose an iterative AutoML-based LRF approach that achieves over 5× compression without degrading the WER, thereby advancing the state-of-the-art in ASR compression.

This paper presents a novel customer malling behavior modeling framework for an urban shopping mall. As an automated computing framework using smartphones, it is designed to provide comprehensive understanding of customer behavior. We prototype the framework in a real-world urban shopping mall. Development consists of three steps; customer data collection, customer trace extraction, and behavior model analysis. We extract customer traces from a collection of 701-hour sensor data from 195 in-situ customers who installed our logging application at Android Market. The practical behavior model is created from the real traces. It has a multi-level structure to provide the holistic understanding of customer behavior from physical movement to service semantics. As far as we know, it is the first work to understand complex customer malling behavior in offline shopping malls.