HNHacker News
TopNewBestAskShowJobs

ks1723

75 karma · joined July 25, 2016

submissionscomments
ks1723··on The boundary of neural network trainability is fractal
Just to add to this, only the two learning rates are changed, everything else including initialization and data is fixed. From the paper:

Training consists of 500 (sometimes 1000) iterations of full batch steepest gradient descent. Training is performed for a 2d grid of η0 and η1 hyperparameter values, with all other hyperparameters held fixed (including network initialization and training data).

ks1723··on CERN aims to build €20B collider
No, it is not wrong. I think your statement comes from a rather short sighted view on scientific and technological developments. We need progress as a society and as humankind to develop and further advance our (currently rather poor) state of affairs (on a global scale). Be it technological (where new tech can actually improve stuff considerably) or 'just' on the understanding of things, i.e. knowledge'.

The comparatively small research budgets would not have any noteworthy impact on dysfunctional systems such as healthcare anyway.

EDIT: Actually, I am not pro this new CERN machine. I think, there are other scientific things where it would make more sense to fund them.

ks1723··on Americans are fake and the Dutch are rude (2022)
Ha, it’s just that I am old ;-) Thanks for the link!
ks1723··on Americans are fake and the Dutch are rude (2022)
Hmm, interesting. I really wonder how I could miss this all my life. (Would have made them here as well, but not anymore)

Edit: spelling

ks1723··on Americans are fake and the Dutch are rude (2022)
Interestingly I became aware of a similar cultural difference regarding punctuation: I frequently use three dots ‚…‘ in more informal emails at the end of a sentence to indicate that I am not sure about something yet or still wondering about it. Like „I guess we could discuss this tomorrow…“ or „Yeah, I might be able to do this…“

Some month ago an Italian PhD student of mine (I’m from Germany) asked me if I was upset or mad at him. I had no idea why he was asking and was completely taken aback by this suggestion. But turns out, in Italy (at least the northern part of it), the three dots typically indicate that you are upset or at least impatient with the receiver.

I still wonder how many of my Italian collaborators I have sent strange impatient emails in the past…

ks1723··on Did a 1997 merger ruin Boeing?
Ok, then I misunderstood, and we can only agree to disagree. I think there need to be people caring about organizational stuff which you cannot fully formulate as engineering problem: setting deadlines, putting teams together, determining team leads, determining how much budget is spent on which branch or discipline or even which part development. In addition, even more technical problems like calculating and comparing costs resulting from parts-Design over the full product lifecycle (design, manufacturing, usability, maintenance, repair, disposal) is not feasible today. And using approximations amounts to making assumption and thereby decisions up-front, where you usually cannot oversee all their implications. Designing complex machinery and shaping the organization which builds them is hellishly complex.
ks1723··on Did a 1997 merger ruin Boeing?
I agree with you and you basically make the same point as wanted to make: you can’t just frame it as an engineering problem, it involves accountants and bosses. So your original statement „If you frame it properly to engineers, they can solve cost problems too.“ is too simplistic in my view.
ks1723··on Did a 1997 merger ruin Boeing?
This is too simple. For large endeavors (building a plane) it is impossible to formulate all contributions to all quality criteria/cost and constraints into a technical engineering optimization problem. Let alone when re-structuring and optimizing the company organization and processes to build those planes at the same time.
ks1723··on Planting Undetectable Backdoors in Machine Learning Models
As a side question: What is the NIST post-quantum debacle? Could you give some references?
ks1723··on How Bell’s Theorem proved ‘spooky action at a distance’ is real
Sorry, this is just wrong. Bell‘s inequality and the very related Bell-Kochen-Specker theorem [1] state that local hidden variables (one ball is blue, one is red, but we just don’t know it) are not consistent with QM.

[1] https://en.m.wikipedia.org/wiki/Kochen–Specker_theorem

ks1723··on Reality as a Vector in Hilbert Space
I fully agree to the last point. He had (and probably/hopefully continues to have) very interesting discussions on all kinds of topics ranging from physics to biology to philosophy to psychology to social topics to ... in his mindscape podcast https://www.preposterousuniverse.com/podcast/
ks1723··on Gradient Descent Models Are Kernel Machines
This is a little too simplistic/misleading

- yes, QM is linear in terms of states, vector spaces and operators acting on states (its just functional analysis [1])

- it’s not linear in terms of how probability distributions and states behave under the action of forces and multi-body interactions (see, eg correlated phenomena like phase transitions, magnetism, superconductivity etc)

[1] https://en.m.wikipedia.org/wiki/Functional_analysis

ks1723··on TimeTagger: Open-Source Time Tracker
First of all this is quite interesting and I appreciate your effort, thanks!

And since I am currently looking for a lean time tracker which runs locally for a single user without any internet and cloud stuff, this is very useful.

I fully agree with earlier the feedback of r4tn4l (most importantly, that a delete button would be really helpful) And here are some additional comments:

- the dashed line indicating 'now' is really easy to overlook and it should be visually more prominent (other color, thicker, ...)

- In the overview (this week, this month, ...) the records should be display somehow. Indicating records visually in the overviews makes it easier to grasp the work distribution over longer periods of time. Additionally, if I mess up the month/year of a record (which I frequently do ;-) ) I have no chance to find it again except that I scroll through every day and look for it. So maybe assigning a separate color to each tag and display them in the big grey fields next to the time line would work... (edit: formatting)

ks1723··on Europe says 737 Max won't fly until it completes it own design review
I am not so sure about this since i read this Insider view: https://www.reddit.com/r/videos/comments/bdfqm4/the_real_rea...
ks1723··on Gaussian Distributions Are Soap Bubbles
What i found quite surprising in that context, is that the volume of the n-dimensional ball for any finite fixed radius goes to zero as n goes to infinity (see, for example, section "high dimensions" of https://en.m.wikipedia.org/wiki/Volume_of_an_n-ball)
← PreviousPage 2 of 2