We investigated the psychological mechanism of survival processing advantage from the perspective of false memory in two experiments.Using a DRM paradigm in combination with analysis based on signal detection theory,w...We investigated the psychological mechanism of survival processing advantage from the perspective of false memory in two experiments.Using a DRM paradigm in combination with analysis based on signal detection theory,we were able to separately examine participants’utilization of verbatim representation and gist representation.Specifically,in Experiment 1,participants rated semantically related words in a survival scenario for a survival condition but rated pleasantness of words in the same DRM lists for a non-survival control condition.The results showed that participants demonstrated more gist processing in the survival condition than in the pleasantness condition;however,the degree of item-specific processing in the two encoding conditions did not significantly differ.In Experiment 2,the control task was changed to a category rating task,in which participants were asked to make category ratings of words in the category lists.We found that the survival condition involved more item-specific processing than did the category condition,but we found no significant difference between the two encoding conditions at the level of gist processing.Overall,our study demonstrates that survival processing can simultaneously promote gist and item-specific representations.When the control tasks only promoted either item-specific representation or gist representation,memory advantages of survival processing occurred.展开更多
In this paper,we summarize recent progresses made in deep learning based acoustic models and the motivation and insights behind the surveyed techniques.We first discuss models such as recurrent neural networks(RNNs) a...In this paper,we summarize recent progresses made in deep learning based acoustic models and the motivation and insights behind the surveyed techniques.We first discuss models such as recurrent neural networks(RNNs) and convolutional neural networks(CNNs) that can effectively exploit variablelength contextual information,and their various combination with other models.We then describe models that are optimized end-to-end and emphasize on feature representations learned jointly with the rest of the system,the connectionist temporal classification(CTC) criterion,and the attention-based sequenceto-sequence translation model.We further illustrate robustness issues in speech recognition systems,and discuss acoustic model adaptation,speech enhancement and separation,and robust training strategies.We also cover modeling techniques that lead to more efficient decoding and discuss possible future directions in acoustic model research.展开更多
基金supported by the National Natural Science Foundation of China(31271202,30870763,J1103602)Beijing Natural Science Foundation(7093123)
文摘We investigated the psychological mechanism of survival processing advantage from the perspective of false memory in two experiments.Using a DRM paradigm in combination with analysis based on signal detection theory,we were able to separately examine participants’utilization of verbatim representation and gist representation.Specifically,in Experiment 1,participants rated semantically related words in a survival scenario for a survival condition but rated pleasantness of words in the same DRM lists for a non-survival control condition.The results showed that participants demonstrated more gist processing in the survival condition than in the pleasantness condition;however,the degree of item-specific processing in the two encoding conditions did not significantly differ.In Experiment 2,the control task was changed to a category rating task,in which participants were asked to make category ratings of words in the category lists.We found that the survival condition involved more item-specific processing than did the category condition,but we found no significant difference between the two encoding conditions at the level of gist processing.Overall,our study demonstrates that survival processing can simultaneously promote gist and item-specific representations.When the control tasks only promoted either item-specific representation or gist representation,memory advantages of survival processing occurred.
文摘In this paper,we summarize recent progresses made in deep learning based acoustic models and the motivation and insights behind the surveyed techniques.We first discuss models such as recurrent neural networks(RNNs) and convolutional neural networks(CNNs) that can effectively exploit variablelength contextual information,and their various combination with other models.We then describe models that are optimized end-to-end and emphasize on feature representations learned jointly with the rest of the system,the connectionist temporal classification(CTC) criterion,and the attention-based sequenceto-sequence translation model.We further illustrate robustness issues in speech recognition systems,and discuss acoustic model adaptation,speech enhancement and separation,and robust training strategies.We also cover modeling techniques that lead to more efficient decoding and discuss possible future directions in acoustic model research.
