Marcel Keller, Ke Sun
We have not lifted any functions out of this paper's repositories yet, so there is nothing we have run. The repositories linked to it are listed below.
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Softmax is widely used in deep learning to map some representation to a probability distribution. As it is based on exp/log functions that is relatively expensive in multi-party computation, Mohassel and Zhang (2017) proposed a simpler replacement based on ReLU to be used in secure computation. However, we could not reproduce the accuracy they reported for training on MNIST with three fully connected layers. Later works (e.g., Wagh et al., 2019 and 2021) used the softmax replacement not for computing the output probability distribution but for approximating the gradient in back-propagation. In this work, we analyze the two uses of the replacement and compare them to softmax, both in terms of accuracy and cost in multi-party computation. We found that the replacement only provides a significant speed-up for a one-layer network while it always reduces accuracy, sometimes significantly. Thus we conclude that its usefulness is limited and one should use the original softmax function instead. Changelog: We fixed a bug in our software affecting the accuracy. The new figures support our conclusion more strongly.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2011.11202")
get_code_for_paper("2011.11202")
have("2011.11202")
The run record, dated, one paper per request, free:
curl https://syntology.ai/api/ran/2011.11202.json
A badge for a README (the split and the date, never a ratio):
[](https://syntology.ai/paper/2011.11202)
Connect an agent — have() is free.