News for hackers. Stuff that matters.

jarekd

Re:Improving KAN with CDF normalization to quantiles on Wednesday July 23, 2025 @02:18AM (#44656193)
In ML there is usually used normalization by subtracting the mean and dividing by standard deviation - I haven't seen by CDF in ML (?, they are popular in finance for copulas: https://en.wikipedia.org/wiki/Copula_(statistics) ), which provides more uniform distributions, allowing for better description with smaller models, what seems beneficial for generalization (e.g. description with low degree polynomials in this arXiv).

For which tasks CDF/EDF normalization could be beneficial in ML? Any reasons it seems unknown in ML?

Any other interesting nonstandard normalizations?

Just represent joint density for each neuron as a linear combination - then you can inexpensively propagate in both directions e.g. as E[X|Y,Z] or E[Y,Z|X] by substituting and normalizing ... the formulas turn out quite simple - could be hidden in dynamics of (bidirectional) biological NN ...

And for pairwise distribution becomes ~KAN, which turned out quit successful ... so we are talking about its extension: adding more possibilities, like triplewise dependencies and multidirectional propagation.

Sure biological NN are much more complicated, but basically action propagation can travel in both directions, and evolution should optimize for that.

In contrast, current ANNs are focused on unidirectional propagation, and are much worse at training from single samples - to reach abilities of biological, maybe it is worth to start thinking about multidirectional?

Neurons containing joint distribution model can do propagate conditional distributions in various direction, and it is not that difficult to represent - maybe something like that could be hidden in biological (?)

(Multidirectional) biological neural networks are no longer superior in MNIST benchmark ... but e.g. consciousness, or being able to learn from single examples.

And no, recreating it is not a task a single person can complete.

Which benchmarks for multidirectional neurons? To compare with which approaches?

Multidirectional are biological neurons, but I don't know how to compare with them?

There is a dozen of papers in this methodology (e.g. end of https://community.wolfram.com/groups/-/m/t/3017754 ), but not as ANN.

However, it degenerates to ~KAN if restring to pairwise dependencies (can consciously add triplewise and higher), and gives many new possibilities, like multidirectional propagation, of values or probability distributions, with novel additional training approaches like through tensor decomposition.

While ANNs are rather trained for unidirectional propagation, action potential propagation in biological neurons is symmetric e.g. ”it is not uncommon for axonal propagation of action potentials to happen in both directions” ( https://journals.aps.org/pre/abstract/10.1103/PhysRevE.92.03... ).

Also, while current ANNs use guessed parametrizations, objectively available is joint distribution - biological neuron should be evolutionarily optimized to exploit, and it is relatively simple in approach from this arXiv.

Such joint distribution neurons bring additional training approaches - maybe some of them are used by biological neural networks?

Talk: https://www.youtube.com/watch?v=pv95hvSdA3c

This is enhancement of standard one-way quantum computers (1WQC) by adding missing operation: "reversed" version (CPT symmetry analog) of state preparation (e.g. pull/push, negative/positive pressure, stimulated emission/absorption). There are many arguments it should be possible, but needs experimental confirmation.

In theory this enhancement allows to attack any NP problems, but it likely will be more difficult than Shor's factorization - so might be completely impractical (?)

There are many interesting new question - like of theoretical complexity class for such 2WQC, somewhere between NP and PSPACE?

There might appear also different approaches to NP, so maybe it would be safer to start thinking of cryptography based on PSPACE? https://en.wikipedia.org/wiki/PSPACE-complete

Slides: https://www.dropbox.com/scl/fi/1fymjwd7obq8sdv1e661f/two-way...

Article: https://www.researchgate.net/publication/372677599_Two-way_q...

This is enhancement of standard one-way quantum computers (1WQC) by adding missing operation: "reversed" version (CPT symmetry analog) of state preparation (e.g. pull/push, negative/positive pressure, stimulated emission/absorption). There are many arguments it should be possible, but needs experimental confirmation.

In theory this enhancement allows to attack any NP problems, but it likely will be more difficult than Shor's factorization - so might be completely impractical (?)

There are many interesting new question - like of theoretical complexity class for such 2WQC, somewhere between NP and PSPACE?

There might appear also different approaches to NP, so maybe it would be safer to start thinking of cryptography based on PSPACE? https://en.wikipedia.org/wiki/PSPACE-complete

Re:The Standard Model – Sheldon Lee Glashow on Sunday July 23, 2023 @06:50AM (#36834193)
For more than half a century they unsuccessfully try to unify Standard Model with gravity ... doesn't it mean that at least one of them is inaccurate?

https://en.wikipedia.org/wiki/Quantum_gravity#Nonrenormaliza...

Re:The Standard Model – Sheldon Lee Glashow on Sunday July 23, 2023 @03:21AM (#36833257)
Indeed baryogenesis is one example of baryon number violation, another is Hawking radiation - baryons forming stars and finally ending as massless radiation.

In both these examples conditions are completely extreme - what might be the missing factor for human attempts to directly observe it: in room temperature water.

I would rather expect it e.g. in the centers of some neutron stars - especially those thousands times brighter than possible with standard explanations, like: https://www.space.com/bizare-object-10-times-brighter-than-s...

While electric charge is ultimately conserved due to Gauss law, there is no Gauss law for baryon number.

Re:Hopfions in Modern Physics on Sunday July 23, 2023 @02:19AM (#36832995)
Lots of talks about topological defects including hopfions: http://solitonsatwork.net/?display=archive

Framework for Liquid Crystal Based Particle Models: https://community.wolfram.com/groups/-/m/t/2856493

Re:The Standard Model – Sheldon Lee Glashow on Sunday July 23, 2023 @02:16AM (#36832974)
Thanks, I don't have blog only update my webpage: http://th.if.uj.edu.pl/~dudaj/ , sometime post something on LI: https://www.linkedin.com/in/jarek-duda-7b56213/

E.g. recently I have proposed (and search for collaboration) two-way quantum computers - enhancement with CPT analogue of state preparation, potentially solving NP problems:

https://quantumcomputing.stackexchange.com/questions/33411/t...

Re:The Standard Model – Sheldon Lee Glashow on Sunday July 23, 2023 @01:05AM (#36832696)
Maybe historically, now there is large group around Nick Manton, also in my Jagiellonian University, and the central topic are 3D models of nuclei.

Lots of talks: http://solitonsatwork.net/?display=archive

Some article: https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.12...

They see topological defects as baryon numbers, while we with Manfried Faber interpret them as electric charges for charge quantization.

They use trivial vacuum (potential with single minimum) getting only short-range interactions (nuclear), while we use S^2, SO(3), SO(1,3) vacuum due to Higgs-like potential - allowing for long-range interactions like Coulomb (+QM+GEM for SO(1,3) Lorentz group vacuum).

This is similar to liquid crystals situation (I use similarity with) - for which they get long-range e.g. Coulomb-like interactions, see e.g.: https://pubs.rsc.org/en/content/articlelanding/2019/sm/c9sm0... , https://journals.aps.org/pre/abstract/10.1103/PhysRevE.76.01... , https://www.nature.com/articles/s41598-017-16200-z

Re:The Standard Model – Sheldon Lee Glashow on Sunday July 23, 2023 @12:49AM (#36832626)
Hi, I am the author of ANS and of these diagrams you don't like - made more than a decade ago. Sure, animations are nicer, but you cannot put them into articles ... and indeed somebody else could make them through this decade.

Regarding the Standard Model, its development is mostly new experimental surprises and fitting new corrections to the Lagrangian - currently more than a hundred. It would be great to see them as kind of Taylor expansion of some simpler more fundamental model, not guessing but deriving new terms.

Where to search for such more fundamental model? Maybe in topological defects like recently growing in popularity skyrmion models. For charge quantization: by interpreting curvature of some deeper field as EM field, Gauss law counts topological charge - which has to be quantized.

I am developing this kind of approach since 2009, and seems quite promising, e.g. naturally unifying EM+QM+GEM gravity vacuum dynamics, has 3 leptons, baryons e.g. with proton lighter than neutron ...

Slides with links: https://www.dropbox.com/s/9dl2g9lypzqu5hp/liquid%20crystal%2...

Wikipedia: https://en.wikipedia.org/wiki/Maximal_entropy_random_walk

Hundreds of articles: https://scholar.google.pl/scholar?q=maximal+entropy+random+w...

This is one of topics I work with - since 2006, would gladly discuss.

Older »