Explains TENNs, combining spatial and temporal convolutions to process sequential data more efficiently than CNNs, RNNs, or transformers. It supports convolutional or recurrent operation, training on GPUs and inferring compactly at the edge. Reports state-of-the-art results in speech classification, vital-signs prediction, and automotive object detection with fewer parameters and ultra-low-power operation on 2nd-generation Akida.

Full access to tools, docs, models, and all future white papers. Free to join. No spam. Unsubscribe anytime.