| As the potential applications for artificial intelligence, and thus neural networks expand, and as the prevalence of big data increases, the need for improved training in neural networks that leverage data sets efficiently will soon surface. In addition, research in the field of human simulation has led to significant advancements in quality, time, and cost management for products like military and athletic equipment and vehicles. There is, however, a critical need for human simulation models to run in real time, especially those with large-scale problems like motion prediction (a single motion problem involves prediction of between 500-700 outputs). Hence, this work addresses both challenges by introducing a modified training process for an artificial neural network (ANN) that is capable of mitigating memory issues and improving accuracy with large scale problems that involve minimal taring data. The new modified ANN design is successfully tested on two common human-motion tasks, walking and going prone. Through comparison with a benchmark ANN, the results of the new network are shown to be accurate objectively and subjectively. |
