Max Pfingsthorn

Applied AI Leader | Von der Forschung zur Wirkung | Robotics · Computer Vision · AI Strategy

[Max Pfingsthorn @maxint]

@amy_tabb Neat! I had an eye on Jax for general numerical opt on GPU for a while. Have you seen JaxOpt? They have a BFGS and CG implementations. But I have not really seen proper second order methods, which is weird since Jax can do autodiff‘d hessians.

[Max Pfingsthorn @maxint]

@amy_tabb @david_picard @ducha_aiki RA-L has a conference option for multiple venues, including IROS, ICRA, CASE, RoboSoft, and BioRob: https://t.co/7ji52WgcFJ

[Max Pfingsthorn @maxint]

@jbhuang0604 Would supervisors just tell students to decline review requests then? I think instead of resulting in better reviews this will make it harder to find reviewers in the first place.

[Max Pfingsthorn @maxint]

@amy_tabb I also (don’t) recommend Optimization Algorithms on Matrix Manifolds by Absil, Mahoney and Sepulchre. Super dense, I needed a day per page. But great stuff after reading Nocedal and Wright! https://t.co/1XzDlvmYLD

[Max Pfingsthorn @maxint]

@facontidavide I am attending. Would love to thank you in person for BehaviorTree.CPP! Great work we use all the time in my group.

[Max Pfingsthorn @maxint]

@amy_tabb @kevindoherty95 Hi Amy, I jumped on the first Twitter hype a long time ago when some cool handles were not yet taken, but then I stopped using it. I’m back though, making sure Twitter does not overflow 😉

[Max Pfingsthorn @maxint]

@kevindoherty95 @amy_tabb Very nice work, outlier rejection in SLAM does not get enough attention anymore. Do check out our previous work on solving the discrete problem first and the continuous Problem second: https://t.co/7z0FpTq8ZH

[Max Pfingsthorn @maxint]

@marcmagus @ashwindollar @tom_swiss @chromakode „The comparison is apt.“ I see what you did there.

[Max Pfingsthorn @maxint]

@GoogleAI So is this like a half Huber loss but as an activation? https://t.co/WAeWMNrOjl

[METRICS project @METRICSproject1]

The Sim2Real Challenge under ADAPT Competition is already in the test phase🪧
Learn more about the competition here: https://t.co/MCH0Il9TKi
#metricsproject #adapt #sim2real

[Max Pfingsthorn @maxint]

@DeblockBlock And please normalize by current total funding amount (eg papers/dollar or euro) when using research output as KPI to automate (low) barriers to tiers of lotteries. Also make transfer efforts a KPI.

[Max Pfingsthorn @maxint]

@DeblockBlock It is obviously inefficient. It seems also ingrained in the publish/perish culture: need established record (papers, grants) to get grant, need grants to write papers, need both for tenure, etc. I wonder if all has to change at once to make it work.

[Max Pfingsthorn @maxint]

@amy_tabb @rosorg Interesting! My goto library for Lie stuff was Sophus (https://t.co/aFGqMVbYRP), since it was easy to use also with Ceres Solver. I will give this one a try!

[METRICS project @METRICSproject1]

We have prolonged the Testing and Validation phase for the Sim2Real Challenge under the ADAPT competition to January 25, 2022.
More info on the challenge: https://t.co/XfQ3kHrgjg

[Max Pfingsthorn @maxint]

@jmhessel @amy_tabb Would „three times the number of papers I submit“ be an appropriate answer?

[Max Pfingsthorn @maxint]

@EmtiyazKhan Ghosting during these activities is of course upsetting and unprofessional. But I wonder if a deadline is well defined if you are asked to write emergency reviews before it? Shouldn’t there be an escalation after and not before?

[Max Pfingsthorn @maxint]

@amy_tabb @ducha_aiki To render good novel views, the system has to know quite well where things are and how they look under lighting, etc. The information seems to be there. Might be a data access issue? For measuring I mean.

[Max Pfingsthorn @maxint]

@MarkSchrammVR Very nice! The head crab is a nice touch, must be the Index?

[Max Pfingsthorn @maxint]

@maththrills @duckietown_coo @RSS_Foundation Is this possibly by design, to discourage repeat submissions of recently rejected papers to game the reviewer lottery? A bit more evenly distributed schedule would ease the editorial burden and allow for actual improvement of initially rejected papers for resubmission.

[Max Pfingsthorn @maxint]

@arturokkboss33 Any industrial cameras with external synch, like from IDS or Basler, work great. You can choose the sensor with a good fit for frame rate and resolution.