## Conclusion

I presented here the application of the recursive least squares to linear contextual bandits. In my opinion, this is a simple to implement and scalable algorithm. Its main advantage is the direct use of the inverse of sum of a square matrix, which makes it computationally efficient. It also enables us to explore interesting future use cases:

- We can use this model for multi-objective optimisation which will have a non-Bernoulli reward (unit interval or monetary values).
- We will be able to solve very complex tasks with a lot of arms and interactions while reducing the complexity by updating only relevant regions of the feature space when we receive feedback. In the same spirit as [4], we can have an ever-changing model in which we can add and remove arms.

I hope this blog was a useful entry and can be applied as a cookbook recipe. Stay tuned for future extensions, simulations and use cases!