Asymptotic Sequential Learning.

Review of Omer Tamuz's Paper

In brief, this paper¹ focuses on the convergance rate and its dependencies on actions and signals. They try to prove that the convergence occurs quickly with respect to signals' external observations rather than actions using a two state stochastic process.

Taking the case of Gaussian distribution as the signal distribution they prove that the convergence with respect to actions occurs by $(\sqrt{\log t})$ times faster. They also discuss the definitivity of the convergence placing an emphasis on the sublinerity of the actions' influence.

Note: Sublinearity means $(\lim\limits_{n \to \infty} \frac{x_n}{n})$

Subscribe!

If you find the above content helpful/interesting and wish to read more such articles, then do subscribe to Random Proof to never miss an update.

PS: Don’t hesitate to comment or leave a message

References: