Since my last report we've made progress in several areas.

Eric's been focused on generatingbirdie signals in a way that matches as closely as possible what the SETI@home client would actually return if that birdie were present. For example:

• The client returns no more than 8 spikes per workunit, and the particular 8 it returns depends on the order in which it scans chirp rates.
  
• When a signal has smaller bandwidth than the FFT frequency resolution, the detection power depends on where it lies in the FFT bin. The power is greatest in the center and drops off significantly at the edges.
  

Birdie signal generation now models both of these effects.

We noticed that many high-scoring multiplets consisted of signals from a brief time period - a minute or two. These multiplets are less interesting than ones with signals that are widely separated in time. The scoring function didn't reflect this. So we added another term to the scoring function, which we call "time entropy" - a measure of how separated in time the multiplet's signals are. The web pages for multiplets now show this value, together with the other components of the score.

We've been working with Manik Varma and Pranshu Jain, machine-learning experts from Microsoft. One of the things we've been looking at is using ML to make our multiplet scoring algorithm better at finding birdie multiplets (and hence better at finding ET). The scoring formula is a weighted sum of several factors - signal power, variance in position and frequency, time entropy. Manik and Pranshu used ML techniques to combine these factors in a way that produces higher scores for birdies.

And of course there have been various bug fixes and enhancements to the web interface. You can see all the changes here.

Maybe a month. The main remaining item, I think, is to incorporate the ML results in multiplet scoring.

In my test runs thus far I've been scoring 100K pixels (out of 16M total). Doing a full run will take more time and produce a lot more data. This may create some issues.

+1

---

Longer-term, is the plan to keep collecting SETI data and re-run Nebula at intervals? Or once this Nebula run is done, is that the end of the project?

---

Thanks for all the hard work to you and the entire team David.

---