Reconfigurable biological signal co-processor for feature extraction dedicated to implantable biomedical microsystems

Memristor-Based Hierarchical Attention Network for Multimodal Affective Computing in Mental Health Monitoring

“…Notably, the structures of the fast-path layer and the slow-path layer mainly conclude four components, i.e., the LSTM module, the self-attention module, the dense layer module, and the feature extraction module. Particularly, the circuit implementation of the dense layer module and the feature extraction module is relatively simple and easy to achieve [19,20]. The specific implementation scheme can be found in [19,20], and we will not repeat any details here.…”

“…Particularly, the circuit implementation of the dense layer module and the feature extraction module is relatively simple and easy to achieve [19,20]. The specific implementation scheme can be found in [19,20], and we will not repeat any details here. This section mainly focuses on the circuit design of LSTM module and self-attention module, due to the fact that the relative research is insufficient and challenging.…”

“…Notably, the visual signal contains two parts here, i.e., the image part and the text part. Then, the hierarchical feature representation (i.e., the image features, text features, and audio features) can be achieved by the feature extraction module [20]. Next, these hierarchical features are fused using the self-attention module and further injected to the LSTM module.…”