“Hive Mind” Shared Deep Learning For SQ Users

I have been submitting a lot of new ideas for the development of SQX lately but there has been one issue I could not figure out how to approach. After digging through a lot of the forums it seems as if many people are trying to achieve the same goal which is finding a way to benefit from strategies other people produce. The core issue with this is that no one wants to share their strategies once they find them ;). This means if there was ever an infrastructure built for people to provide free strategies, they would more than likely consist of strategies that aren’t that great… or at least not a great as they can be. They are more or less throwaways that devs provide to help newcomers. However, I believe there is a solution to this that allows everyone to win without having access to anyone else’s code and allows us all to benefit from the best strategies we can find.

 

An in-depth explanation is here: https://roadmap.strategyquant.com/tasks/sq4_4042

 

The concept is relatively simple:

1. Create a database where users can submit strategies to directly from the SQX platform
2. Implement a Neural Network that can learn from all the strategies submitted
3. Provide an updated Neural Network to all SQ users on each SQX update

 

And that’s all there is to it. The Neural Network would be embedded into each new build of SQ and would get implemented directly into genetic evolution. Its job would essentially be to make edits to bad strategies based on its learning from the collective dataset we create. This would mean SQ users would be incentivized to submit their best work as it would help everyone build better strategies in the future. The more the database grows, the more efficient the Neural Network should be at creating better strategies out of the random parameters you throw at it. No one would have access to the strategies stored in the database. We would all benefit from each iterative update where the AI would learn new patterns based on the new strategies we’re submitting. Essentially we should all be able to generate better strategies much faster one day using this method and no one’s strategy would be exactly the same as someone else’s.

 

Feel free to read the more in-depth explanation of how this would work in the link above and vote for it if you think its a good idea. I just wanted to start this thread for anyone who wanted to contribute their thoughts/critiques or add to this idea. Normally I wouldn’t go out of my way to talk about new features and ideas, especially given the fact that this would be a very long term project. However, I think what this could accomplish in the end would dramatically help everyone. Feel free to let me know what you think about this.

 

Thanks for your contribution. That’s an interesting concept of sharing knowledge inside the SQ community while still “keeping the secrets” of somebody’s hard work. I can understand why some might be unwilling to share their work. Financial markets are probably the most competitive environment you can ever participate but your solution makes sense and would basically solve the conflict and also help to add value into StrategyQuant. Mark will surely check this out and consider what could be done and how it could be possibly implemented into future StrategyQuant