#differentiable — Public Fediverse posts

🧠 New paper by Deistler et al: #JAXLEY: differentiable #simulation for large-scale training of detailed #biophysical #models of #NeuralDynamics.

They present a #differentiable #GPU accelerated #simulator that trains #morphologically detailed biophysical #neuron models with #GradientDescent. JAXLEY fits intracellular #voltage and #calcium data, scales to 1000s of compartments, trains biophys. #RNNs on #WorkingMemory tasks & even solves #MNIST.

🌍 https://doi.org/10.1038/s41592-025-02895-w

#Neuroscience #CompNeuro

🤔 Oh look, Alice is having mathematical tea with the Cheshire Cat in a "differentiable" wonderland, because the real world just isn't complex enough! 🚀 Meanwhile, arXiv, the Wikipedia of nerds, wants a #DevOps wizard to keep the scientific cat videos running... because, clearly, science can't disseminate itself! 🧙‍♂️✨
https://arxiv.org/abs/2404.17625 #AliceInWonderland #Differentiable #Math #ScienceNerds #CheshireCat #HackerNews #ngated

Alice's Adventures in a Differentiable Wonderland

https://arxiv.org/abs/2404.17625

#HackerNews #Alice #Adventures #Differentiable #Wonderland #AI #Research

Differentiable Logic Cellular Automata — https://google-research.github.io/self-organising-systems/difflogic-ca/?hn
#HackerNews #Differentiable #Logic #Cellular #Automata #research #AI #machinelearning #selforganizing-systems

@schmitz (left) explaining his recent work on making #dftk algorithmically #differentiable at the #cecam workshop on #dft and #ai (https://www.cecam.org/workshop-details/1281). With his work derivatives of key density-functional theory quantities like forces or band structures wrt. model parameters can now be easily computed.

Deep Learning with Jax
https://www.manning.com/books/deep-learning-with-jax
#ycombinator #neural_networks #Python #Google #Flax #Haiku #gradient_processing #optimization #numerical #computations #federated_learning #parallelized #distributed #primitives #differentiable

The #Zhukovsky #Aerofoil (sometimes transliterated as #Joukowsky from #Russian), is a 2D model of #streamlined #Airflow past a #wing. It uses #ComplexVariable and is an #AnalyticFunction (i.e. #Differentiable everywhere, save at isolated #Singularities). Take a circle in the #ComplexPlane which is not quite centred at the #origin but passes through the #coordinate (1,0) or (z=1+0i).

#MyWork #CCBYSA #AppliedMathematics #WxMaxima #FreeSoftware #Aeronautics #Aerodynamics #LaminarFlow

Whenever I walk to/from home, I have to walk up/down an inclined street; I noticed that the asphalt floor has different curvatures depending on how near it is of a bend, and I try to find a less steep incline while walking.

This got me inspiration for the few questions below. Any simple explanations, and related links, are welcome.

Given a #differentiable surface within R^3, and two distinct points in it, there are infinitely many differentiable paths from one point to another, remaining on the surface. At each point of the #path, one can find the path's local #curvature. Then:

- Find a path that minimizes the supreme of the curvature. In other words, find the "flattest" path.

- Find a path that minimizes the variation of the curvature. In other words, find a path that "most resembles" a circle arc.

Are these tasks always possible within the given conditions? Are any stronger conditions needed? Are there cases with an #analytic solution, or are they possible only with numerical approximations?

#Analysis #DifferentialGeometry #Calculus #DifferentialEquations #NumericalMethods

#Juliacon #workshops are out :julia: !

So many cool stuff! I am particularly looking forward to

- #webdev with #genieframework

- #differentiable #programming on #gpu

- #imageprocessing with #julialang

what are your favorites?

https://pretalx.com/juliacon2023/featured/

see you in #boston in a few months!