A Low-cost Fault Corrector for Deep Neural Networks through Range Restriction

The adoption of deep neural networks (DNNs) in safety-critical domains has\nengendered serious reliability concerns. A prominent example is hardware\ntransient faults that are growing in frequency due to the progressive\ntechnology scaling, and can lead to failures in DNNs.\n This work proposes Ranger, a low-cost fault corrector, which directly\nrectifies the faulty output due to transient faults without re-computation.\nDNNs are inherently resilient to benign faults (which will not cause output\ncorruption), but not to critical faults (which can result in erroneous output).\nRanger is an automated transformation to selectively restrict the value ranges\nin DNNs, which reduces the large deviations caused by critical faults and\ntransforms them to benign faults that can be tolerated by the inherent\nresilience of the DNNs. Our evaluation on 8 DNNs demonstrates Ranger\nsignificantly increases the error resilience of the DNNs (by 3x to 50x), with\nno loss in accuracy, and with negligible overheads.\n