Applying Advanced Learning Algorithms to ALVINN

摘要: ALVINN (Autonomous Land Vehicle in a Neural Net) is Backpropagation trained neural network which capable of autonomously steering vehicle road and highway environments. Although fairly robust, one the problems with it has been time takes to train. As on-line learning, driver drive car for about 2 minutes before autonomous operation. One reason this use Backprop. In report, we describe original system, then look at three alternative training methods Quickprop, Cascade Correlation, 2. We run series trials using Correlation Cascade2, compare them BackProp baseline. Finally, hidden unit analysis performed determine what learning. Applying Advanced Learning Algorithms Parag H. Batavia Dean A. Pomerleau Charles E. Thorpe 9 October 1996 CMU-RI-TR-96-31 This research was partially supported under National Science Foundation Graduate Fellowship, Highway Traffic Safety Administration (NHTSA) contract no. DTNH22-93-C-07023, by USDOT Cooperative Agreement Number DTFH61-94-X-00001 as part Automated System Consortium