Temporal Attention-Gated Model for Robust Sequence Classification

摘要: Typical techniques for sequence classification are designed well-segmented sequences which have been edited to remove noisy or irrelevant parts. Therefore, such methods cannot be easily applied on expected in real-world applications. In this paper, we present the Temporal Attention-Gated Model (TAGM) integrates ideas from attention models and gated recurrent networks better deal with unsegmented sequences. Specifically, extend concept of model measure relevance each observation (time step) a sequence. We then use novel network learn hidden representation final prediction. An important advantage our approach is interpretability since temporal weights provide meaningful value salience time step demonstrate merits TAGM approach, both prediction accuracy interpretability, three different tasks: spoken digit recognition, text-based sentiment analysis visual event recognition.