Conclusion.tex

\section{Conclusion}\label{sec:Conclusion}

This project looked into how we can use the SMCDEL library to better understand and model Gossip. To the first point, we wrote \texttt{gsi}, our \textit{Gossip Scene Investigation} function, to better understand SMCDEL's Knowledge Scenes as they pertain to The Gossip Problem. To the latter point, we used the existing notion of a Knowledge Transformer in SMCDEL to write a Knowledge Transformer for the Transparent Gossip Problem. Both of these aforementioned processes helped us build an understanding of how SMCDEL approaches Gossip, and specifically what makes it so computationally intensive. With this in mind, we tried using a pre-existing optimize function within SMCDEL's library to reduce complexity, as well as writing our own Simple Transformer, to target the blow up in vocabulary that the Classic Transformer implemented in SMCDEL encrues. 

There are two ways we can analyze our work: by considering its speed and accuracy. In Section \ref{sec:Benchmarks}, we explore the first point. The Transparent Transformer and the Simple Transformer both had large improvements on computation time, specifically as the number of calls increased. 

On the other hand, the correctness of our code still has room for improvement. We believe knowing the true differences between what information these transformers codify requires mathematics outside the scope of a programming project. However, it is our belief that the Simple Transformer makes \textit{fewer} propositions true than the Classic Transformer, and therefore it doesn't recognize all instances of higher-order knowledge that are satisfied when the same instance of the problem is run on the Classic implementation. 
% fixme (wouter?): should we mention the fact that the classic transformer also might not be correct? 

In terms of further work, besides correctness, readability is a big focus. The Gossip Problem is a specific example within the area of DEL, and is therefore quite hard to work with a lack of background knowledge. This is part of the reason we wrote the \texttt{gsi} function, and we advise the reader to test its usability by running \texttt{main}. However, although we decode the propositional variables and observations, the state law is still a large BDD, and uninterpretable by the user. Future work could be done to make this more user friendly, perhaps by way of the tool \textit{graphviz}.