Metastimuli for Human Learning via Machine Learning and Principal Component Analysis of Personal Information Graphs

A system for correlating information stimuli with a user's personal information management system (PIMS) via deep learning is presented for the purpose of improving human learning by way of the haptic bond effect. Studies on human learning have provided evidence that additional modalities of information transfer during the learning process improves human learning rates. An artificial neural network (ANN), trained with textual data labeled with their locations in a user's PIMS, a graph structure defined by the information architecture developed previously by Picone & Powell. This enables simultaneous PIMS-correlated stimuli (called metastimuli) and the original stimuli. The metastimuli is applied in this system via an actuator, typically haptic. Using principal component analysis, the dimensionality of the PIMS graph is reduced to m and the graph is represented in R^m, where m is also the actuator dimensionality. A suite of optimization is conducted on the ANN weights and biases, hyper-parameters, and meta-parameters. Optimization results suggest that atom embeddings created from a small dataset with a high specificity to the problem set yields improved results over a large dataset with a low specificity. ANN training results demonstrate effective prediction of R^m vectors for the testing set with high accuracy. These results demonstrate feasibility for further studies into the effectiveness of this system and metastimuli for human learning.