nasa caltech_jpl
GTOC X
10th Global Trajectory Optimisation Competition

May 21, 2019, 8 p.m. UTC

June 12, 2019, 8 p.m. UTC

Timeline

The competition is over.

Post-Competition Leaderboard?

avatar

Anastassios@JPL

I wanted to generate a bit of further discussion on the post-competition leaderboard idea - if there is sufficient interest, we will try to make it happen in the very near future (so teams will also be able to validate new results for the conference). The purpose of it would be to use the luxury of time and potentially inter-team collaboration to foster further research in identify fruitful avenues of optimisation of this class of problems. If we do it, I would propose B=1 and we will make clear that it is "post competition". What are your thoughts, one way or another?

avatar

shenhongxin

I think it is a good idea facilitating future research on gtops. My personal most impressive post-competition solution is "48+1" from Dietmar for GTOC4!
Maybe I will be a user to the post-competition leaderboard, thank you!

avatar

darioizzo

I would then create a snapshot of the final leaderboard and add it as a new static page maybe called "Results" ... You can create exactly the same look as the dynamic leaderboard.

avatar

IngoA

Dear Shen, dear friends,

> My personal most impressive post-competition solution
> is "48+1" from Dietmar for GTOC4!

The winning solution in GTOC4 had 44 asteroids for flyby + 1 for landing.
And the best solution Dietmar found afterwards (in April 2018)
has 49 + 1 asteroids. You may have a look at
https://www.youtube.com/watch?v=7QxikroB-6Q&t=2s
Dietmar reached it over the years in several "small" steps: 45+1, 46+1, ... 49+1.

Let me use the opportunity to tell a little anecdote around this development.
It was in Summer 2009, when I learned about the GTOC for
the first time. GTOC4 had just finished, with the winning
team coming from Moscow. I found that stuff so supercool
that I got the idea to participate in the scene by some
way or the other.

However, I would need a strong programmer with background
both in combinatorial and continuous optimization. I
remembered the Eternity Puzzle (from the years 1999 and 2000)
where Dietmar Wolz had done a fantastic job in the after
competition: After several months of engagement, his standard
Pentium III PC generated another new Eternity solution every five
days in average.

So, Dietmar had a strong background in combinatorial optimization
(but not in the continuous counterpart). Nevertheless, he took
fire and started to write GTOC programs. He found 44+1 soon
and was convinced to get 45+1 and 46+1 records within weeks. We
contacted Dario, and I told him about Dietmar's progress.
Unfortunately it turned out several times that Dietmar's solutions
were not correct or not numerically stable. However, he never
lost faith and tried again and again. And then the 4x+1 solutions
started rolling in.

And my role, you may ask: I introduced Dietmar to the scene,
became his manager, was his sparring partner in discussions,
and also sort of his jester. My heydays as a programmer I
had had in the late 1970's, exploiting the programmable TI 59
pocket calculator to the extremes.

Cheers, Ingo.

avatar

ignacio.torralba

Dear Anastassios,

Thank you very much for considering seriously the idea of a Post-competition Leaderboard.
Let me clarify that the intention is NOT to diminish the solutions found during the competition and the level of excellence of their creators. It is obvious to everyone that the competition is extremely difficult not only due to the difficulty of the problem but also due to the time constraint, therefore the nominal Leaderboard must be clearly highlighted and the winners recognized for it.

I believe, the presented idea is of great interest for small teams (small in terms of people or computational power or time available or new participants,...) and for researchers.

Having this tool available will allow to verify that your result is valid (avoiding the issues Ingo was mentioning) and to keep researching, no only, to improve your own best score but also to compete for a better position in the Post-competition Leaderboard (only a competition can provide such a push).

I strongly believe that having the Post-competition Leaderboard available will make GTOC11 much more competitive and will allow the creation of new algorithms with great performance for global optimization.

avatar

ignacio.torralba

PS. Even JPL should be allowed to participate.

avatar

IngoA

Hello,
I very much like the postings by Ignacio, and in
particular his "JPL encouragement".

And thanks again to Anastassios for the website with the
solutions by participants.

Cheers, Ingo.

avatar

Anastassios@JPL

Ignacio, now that's a challenge - we have 1 very intense month of catching up to do :)
I think there is sufficient interest for a post-competition leaderboard, so I will set it up, hopefully in the coming week.

avatar

ignacio.torralba

Anastassios, I am glad this idea has been found interesting for different teams and, I think, we all are looking forward to seeing JPL results :)

avatar

DietmarW

Solving/improving GTOC results long after the competitions ended is one of my hobbies. There was never the possibility to verify these solutions before, if we could change that it would be great. No one wants to diminish the results during the competition, therefore I suggested in another thread that we could use a "bonus factor" of 0.1 - 0.3 to "protect" the best solutions so far. This would somehow "separate" the post competition results from the existing ones.

avatar

DietmarW

After competition improvements have a long tradition at GTOC, to show a few examples see http://www.esa.int/gsp/ACT/doc/AI/pub/ACT-RPR-AI-2013-jupiter_tour.pdf improving GTOC6 which introduced "Lazy Race Tree Search" and further back at 2007 https://www.sciencedirect.com/science/article/pii/S0094576507001750 which shows improvements of the submitted GTOC1 result. Much later https://scientificcomputation.blogspot.com/2018/12/how-to-solve-gtoc1-today.html refers to this paper and tries to replace search by optimisation to achieve a similar result.

Please log in to leave a comment.