| Map Unavailable |
KEGS members and non-members are encouraged to attend. Geosoft is the technology sponsor for the online portion of this event. Geosoft is delighted to host the online portion of the Canadian Exploration Geophysical Society meeting.
The talk will be presented on 8 April 2014 between 4:30 pm and 5:30 pm in Toronto, Canada. Australian times are:
Melbourne, Sydney, Brisbane: April 9, 6:30 AM to 7:30 AM (as shown in the events calendar)
Adelaide and Darwin: April 9, 6:00 AM to 7:00 AM
Perth: April 9, 4:30 AM.
Registration
To register for this event click here
Geophysical survey contractors routinely offer multi-parameter data to clients. For example, a helicopter-borne survey may acquire Time-domain electromagnetic (TDEM), magnetic gradiometer and even gamma-ray spectrometer data (i.e., VTEMplus, Geotech LTD). Exploration geophysicists can certainly take advantage some of the readily available multi-disciplinary (geology, geophysics and remote sensing) and multi-parameter (potential field, EM, gamma-ray spectrometry, and others) datasets for mineral exploration. However, the integration and interpretation of these datasets can be time-consuming and even challenging, especially for large-scale datasets covering large areas with diverse geological conditions. The Supervised Neural Network (NN) Targeting and Classification technique for mineral exploration described and demonstrated by Reford, Lipton and Ugalde, 2004, “Predictive Ore Deposit Targeting Using Neural Network Analysis” (http://www.pgw.on.ca/downloads.html), can be a useful and promising tool for the analysis of multi-disciplinary and multi-parameter data.
In this presentation, the properties or responses of the two feed-forward multilayer Neural Networks, Levenberg-Marquardt (NN with LM training) and Fast Classification (FCNN), as implemented in the current version by PGW, are studied in detail. The supervised NN simulations are performed on specially constructed synthetic data. Intended as a tutorial and the NNs treated as black boxes, the objectives of the exercise are twofold, to demonstrate the targeting as well as classification capabilities of the Neural Networks, and at the same time to show one of the known limitations and to suggest a way to get around it. The utility of the NN tool is demonstrated again with real cases from the Republic of Niger.
