2019/10/02 by Deepan Muthirayan, Muthirayan, Deepan, Scott Nivison +3 · 1 citation
Computer Science · Engineering · #Advanced Memory and Neural Computing #Domain Adaptation and Few-Shot Learning #FOS: Electrical engineering #Neural Networks and Applications #Systems and Control (eess.SY) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1910.01189
openalex publication_date 2019/10/02 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28
We introduced a it working memory augmented adaptive controller in our\nrecent work. The controller uses attention to read from and write to the\nworking memory. Attention allows the controller to read specific information\nthat is relevant and update its working memory with information based on its\nrelevance. The retrieved information is used to modify the final control input\ncomputed by the controller. We showed that this modification speeds up\nlearning. In the above work, we used a soft-attention mechanism for the\nadaptive controller. Controllers that use soft attention or hard attention\nmechanisms are limited either because they can forget the information or fail\nto shift attention when the information they are reading becomes less relevant.\nWe propose an attention mechanism that comprises of (i) a hard attention\nmechanism and additionally (ii) an attention reallocation mechanism. The\nattention reallocation enables the controller to reallocate attention to a\ndifferent location when the relevance of the location it is reading from\ndiminishes. The reallocation also ensures that the information stored in the\nmemory before the shift in attention is retained which can be lost in both soft\nand hard attention mechanisms. We illustrate through detailed simulations of\nvarious scenarios for two link robot and three link robot arm systems we\nillustrate the effectiveness of the proposed attention mechanism.\n