Broader Impact

  • Our implementation of automatic differentiation provides a fast and accurate way of calculating derivatives. Our SPLAD Tool package is handy and straightforward to apply in many fields. When handling large-scale computation, utilizing our package will relieve calculation workload and avoid computational errors. Besides, the package is also helpful in dealing with a wide range of mathematical problems. For example, by adding the implementation of loss functions, we were able to apply our spladtool_reverse to construct a simple data classifier, which is demonstrated in detailed under the exmaple directory. Furthermore, our package can also be used to construct root-finding algorithms based on Newton’s method.

  • While our automatic differentiation package provides many conveniences and can be applied widely in many fields, it might also be misused in some conditions. As a convenient tool for calculating derivatives automatically, our package might hinder students or beginners from thoroughly learning and understanding the basic theory behind the mechanism. This misuse contradicts our original intentions of helping people study and work more efficiently.

Software Inclusivity

  • In order to make our package to be as inclusive as possible, we intend to publish our package as an open-source resource online. By distributing over Github and PyPI, we allow people from all kinds of backgrounds to be able to download, use and coordinate with us. Furthermore, we also encourage other developers from all communities to contribute to our codebase by enabling people to create new pull requests, leave comments in our repository on Github. All of our group members will continue monitoring new comments and pull requests and schedule meetings at any time to discuss further improvement and optimization if needed.

  • Furthermore, to eliminate the barrier to underrepresented groups, we expect to implement new features in the future concerning different communities respectively. For example, to help eliminate the language barrier to non-native English speakers, we expect to provide detailed instructions in multiple languages other than English. Besides, if possible, we may build a GUI that can visualize the trace of automatic differentiation to help users better understand the working flow of automatic differentiation.