Clarification about α and β weights

[Stavros Azinas]

Hi developers, 

Firstly, thanks for releasing the code for this software and making it fairly straightforward to use. However, in anisotropy correction of the half-maps, could you help us clarify how to use the weights properly with some examples if you might? What would be "extreme" values, and how would those affect our data without introducing biases and artifacts?

Thanks again!

[YUNTAO LIU]

Hi Stavros,

Both α and β are ratios that can range from 0 to infinity, with extreme values being 0.01 and 100. These values 0.01 and 100 are considered equally extreme.

The α parameter balances between data consistency and rotation equivariance, while β determines the level of denoising to be applied during training. By default, α is set to 1 and β to 0.5. Supplementary Figure 1 provides examples of different values.

A larger α generally results in smoother outcomes. However, very small values, such as 0.1, might lead to overcompensation. While it is challenging to determine the optimal α, starting with the default value of 1 is probably good. (Think about balancing Tau value in RELION)

For β, the goal is to achieve effective denoising, larger means more denoising. In a slice view of the map, a good β value is indicated by suppressed background noise while preserving the structure.

I hope these help. 

--
You received this message because you are subscribed to the Google Groups "spIsoNet" group.
To unsubscribe from this group and stop receiving emails from it, send an email to spisonet+u...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/spisonet/f61e57bc-48da-448e-8bdf-f7cedc767231n%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Best Regards,

Yuntao Liu,  Postdoc.

California NanoSystem Institute

University of California Los Angeles

[Stavros Azinas]

It does! Thanks again!