Fig 6 - uploaded by Eda Okur
Content may be subject to copyright.
Source publication
Understanding passenger intents and extracting relevant slots are important building blocks towards developing contextual dialogue systems for natural interactions in autonomous vehicles (AV). In this work, we explored AMIE (Automated-vehicle Multi-modal In-cabin Experience), the in-cabin agent responsible for handling certain passenger-vehicle int...
Contexts in source publication
Context 1
... first (i.e., level-1), and then feeding only the words that are predicted as intent keywords & valid slots (i.e., not the ones that are predicted as None/O) as an input sequence to various separate sequence-to-one models (described above) to recognize final utterance-level intents (i.e., level-2). A sample network architecture is given in Fig. 6 (a). The idea behind filtering out non-slot and non-intent keywords here resembles providing a summary of input sequence to the upper levels of the network hierarchy, where we actually learn this summarized sequence of keywords using another RNN layer. This would potentially result in focusing the utterance-level classification problem on ...
Context 2
... Hierarchical & Joint Models. Proposed hierarchical models detect/extract intent keywords & slots using sequence-to-sequence networks first, and then only the words that are predicted as intent keywords & valid slots (i.e., not the ones that are predicted as None/O) are fed as input to the joint sequence-to-sequence models (described above). See Fig. 6 (b) for sample network architecture. After the filtering or summarization of sequence at level-1, <BOU> and <EOU> tokens are appended to the shorter input sequence before level-2 for joint learning. Note that in this case, using Joint-1 model (jointly training annotated slots & utterance-level intents) for the second level of the ...
Context 3
... first (i.e., level-1), and then feeding only the words that are predicted as intent keywords & valid slots (i.e., not the ones that are predicted as None/O) as an input sequence to various separate sequence-to-one models (described above) to recognize final utterance-level intents (i.e., level-2). A sample network architecture is given in Fig. 6 (a). The idea behind filtering out non-slot and non-intent keywords here resembles providing a summary of input sequence to the upper levels of the network hierarchy, where we actually learn this summarized sequence of keywords using another RNN layer. This would potentially result in focusing the utterance-level classification problem on ...
Context 4
... Hierarchical & Joint Models. Proposed hierarchical models detect/extract intent keywords & slots using sequence-to-sequence networks first, and then only the words that are predicted as intent keywords & valid slots (i.e., not the ones that are predicted as None/O) are fed as input to the joint sequence-to-sequence models (described above). See Fig. 6 (b) for sample network architecture. After the filtering or summarization of sequence at level-1, <BOU> and <EOU> tokens are appended to the shorter input sequence before level-2 for joint learning. Note that in this case, using Joint-1 model (jointly training annotated slots & utterance-level intents) for the second level of the ...
Similar publications
Understanding passenger intents and extracting relevant slots are crucial building blocks towards developing contextual dialogue systems for natural interactions in autonomous vehicles (AV). In this work, we explored AMIE (Automated-vehicle Multi-modal In-cabin Experience), the in-cabin agent responsible for handling certain passenger-vehicle inter...
Understanding passenger intents from spoken interactions and car's vision (both inside and outside the vehicle) are important building blocks towards developing contextual dialog systems for natural interactions in autonomous vehicles (AV). In this study, we continued exploring AMIE (Automated-vehicle Multimodal In-cabin Experience), the in-cabin a...
Understanding passenger intents from spoken interactions and car’s vision (both inside and outside the vehicle) are important building blocks towards developing contextual dialog systems for natural interactions in autonomous vehicles (AV). In this study, we continued exploring AMIE (Automated- vehicle Multimodal In-cabin Experience), the in- cabin...
