FAULT-TOLERANCE VIA WEIGHT NOISE IN ANALOG VLSI IMPLEMENTATIONS OF MLP - A CASE-STUDY WITH EPSILON

Training with weight noise has been shown to be an effective means of improving the fault tolerance of multilayer perceptrons (MLP's), This paper investigates the use of weight noise used during MLP training to compensate for the inherent errors encountered in VLSI implementation s. Weight adaptation is conducted solely in software, eliminating the need for costly in-the-loop training. The particular VLSI implementati on considered here is the EPSILON processor card developed at Edinburg h University, Both software and hardware experiments demonstrate the e ffectiveness of this approach. This case study with the EPSILON proces sor card highlights what we believe to be a number of common inadequac ies with custom designed hardware. In particular, the limitations of E PSILON in terms of its dynamic range performance has been shown to be a problem. In summary, we show that networks trained with weight noise are fault-tolerant but also require an increased dynamic range to exp loit this property.