Shakedown Social

j_bertolotti<a href="https://mathstodon.xyz/tags/PhysicsJournalClub" class="mention hashtag" rel="nofollow noopener" target="_blank">#PhysicsJournalClub</a> "Model-free estimation of the Cramér–Rao bound for deep learning microscopy in complex media" by I. Starshynov et al.Nat. Photon. (2025) <a href="https://doi.org/10.1038/s41566-025-01657-6" rel="nofollow noopener" translate="no" target="_blank">https://doi.org/10.1038/s41566-025-01657-6</a>As everybody who ever tried to orient themselves while immersed in thick fog knows, scattering scrambles information. The question "how much information is still there?" is not particularly interesting as the answer is "essentially all of it", as elastic scattering can't destroy information. A much more interesting question is "how much information can we retrieve?" In order to even try to give an answer we need to be a bit more specific, so the authors placed a small reflective surface behind a scattering layer and asked how much information about its transverse position could be retrieved. This is a well-posed question, and the answer takes the form of a "Cramér–Rao bound" (<a href="https://en.wikipedia.org/wiki/Cram%C3%A9r%E2%80%93Rao_bound" rel="nofollow noopener" translate="no" target="_blank">https://en.wikipedia.org/wiki/Cram%C3%A9r%E2%80%93Rao_bound</a>). After estimating this upper bound, the authors investigate how well a trained neural network can do at this task, and show that a specifically built convolutional neural network can almost reach the theoretical bound.[Conflict of interest: Ilya Starshynov (the first author) did his PhD in my group.]<a href="https://mathstodon.xyz/tags/Physics" class="mention hashtag" rel="nofollow noopener" target="_blank">#Physics</a> <a href="https://mathstodon.xyz/tags/InformationTheory" class="mention hashtag" rel="nofollow noopener" target="_blank">#InformationTheory</a> <a href="https://mathstodon.xyz/tags/Optics" class="mention hashtag" rel="nofollow noopener" target="_blank">#Optics</a> <a href="https://mathstodon.xyz/tags/MachineLearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#MachineLearning</a>

j_bertolotti<a href="https://mathstodon.xyz/tags/PhysicsJournalClub" class="mention hashtag" rel="nofollow noopener" target="_blank">#PhysicsJournalClub</a> "Three-dimensional holographic imaging of incoherent objects through scattering media" by Y. Baek, H. de Aguiar and <a href="https://mastodon.social/@sylvaingigan" class="u-url mention" rel="nofollow noopener" target="_blank">@sylvaingigan</a> <a href="https://arxiv.org/abs/2502.01475" rel="nofollow noopener" translate="no" target="_blank">https://arxiv.org/abs/2502.01475</a> <a href="https://mathstodon.xyz/tags/optics" class="mention hashtag" rel="nofollow noopener" target="_blank">#optics</a> <a href="https://mathstodon.xyz/tags/physics" class="mention hashtag" rel="nofollow noopener" target="_blank">#physics</a> <a href="https://mathstodon.xyz/tags/imaging" class="mention hashtag" rel="nofollow noopener" target="_blank">#imaging</a> As you daily experience anytime you look at anything, light scattering severely impairs your ability to image (mild scattering like mist makes things in a distance fuzzy, strong scattering like your own body makes it completely impossible to see what is happening inside or behind it). On one hand this is good, as it allows us to see where (e.g.) trees are so we don't bump into them. On the other hand there are a LOT of situations where you would really like to see what is going on behind a scattering medium (surely it would save a lot of exploratory surgeries). The problem of imaging through a scattering medium is largely unsolvable in its most general form, but there are a lot of special cases where you can go surprisingly far, and people (me included) have spent a lot of time checking exactly how far.In this paper the authors consider a set of small fluorescent objects behind a not-too-thick scattering medium, and look for a way to retrieve their 3D arrangement. Problem: fluorescent emission means incoherent emission, so the phase information (which encodes a lot of information about position) is lost. Still, we can rely on the assumption that there is a finite (ideally not too large) amount of point emitters. Since each emitter is point-like, if we only measure the light that reaches us through the scattering medium at a single frequency (to be more realistic, a small bandwidth), we will see the incoherent sum of a speckle pattern per fluorescent emitter. 1/2

j_bertolotti<a href="https://mathstodon.xyz/tags/PhysicsJournalClub" class="mention hashtag" rel="nofollow noopener" target="_blank">#PhysicsJournalClub</a> "Anderson Transition for Light in a Three-Dimensional Random Medium" by A. Yamilov, H. Cao, and S. Skipetrov Phys. Rev. Lett. 134, 046302 (2025)Anderson localization is a phenomenon where wave interference in a disordered scattering medium results in the wave being unable to propagate. It is a fascinating, complex, and VERY niche topic. It is also the centre of never-ending arguments.Anderson localization was originally proposed as a mechanism to explain why certain spin states in metals full of impurities would take longer to relax than expected (the answer was: their quantum state could not propagate due to the disorder). [<a href="https://www.nobelprize.org/uploads/2018/06/anderson-lecture-1.pdf" rel="nofollow noopener" translate="no" target="_blank">https://www.nobelprize.org/uploads/2018/06/anderson-lecture-1.pdf</a>] It was soon realized that electron-electron interaction would dominate over a pure "Anderson localization" in most solid state systems, where a similar effect ("Mott insulation") would dominate.The topic had a resurgence when it became clear that Anderson localization was a general wave phenomenon, and it wasn't restricted to electrons, but could in principle happen to any wave, e.g. sound or light. [<a href="https://doi.org/10.1103/PhysRevLett.58.2486" rel="nofollow noopener" translate="no" target="_blank">https://doi.org/10.1103/PhysRevLett.58.2486</a>] Due to its nature, Anderson localization is easier to achieve the lower the dimensionality of the system is. Anderson localization of (ultra)sound in 1D and 2D were quickly discovered, with light coming a few years later.1/

Recent searches

Search options

Administered by:

Server stats:

#physicsjournalclub