Discover and read the best of Twitter Threads about #arxiv

Most recents (15)

🧩Three intriguing concepts in the world of #DeepLearning: #BackwardCompatibility, #StationaryRepresentation, and #NeuralCollapse. Their connection may hold the key to better deep learning models with possibly novel ways to use them
🤝Backward Compatibility enables matching internal feature representations from different neural networks. Stationary Representation maintains feature spatial configurations fixed during learning. How does Neural Collapse fit in?
💥Neural Collapse is when feature representations converge towards a simpler structure during training. This fascinating phenomenon is strongly related to compatibility and stationary representation.
Read 7 tweets
Fully differentiable Hückel model using #JAX.
i) We optimize the atom types for different molecular frameworks to find the molecule with the targeted property (HOMO-LUMO) 🧵 (1/3)
@kjelljorner @robpollice @A_Aspuru_Guzik @chemuoft
ii) Parameter optimization a la machine learning for diff. molecular property, for example, polarizability
30% improvement with only 100 DFT training data.
🧵 (2/3) Image
#arxiv #github
(inverse molecular design)…
(parameter optimization)…
🧵 (3/3)
Read 3 tweets
New paper on #ArXiv!
"Tracking moving objects through scattering media via speckle correlations"
With @YJaureguiS, and Harry Penketh.

Short(?) thread explaining what it is about.
Scattering is a major problem for imaging, and it doesn't take very much of it before we can't see essentially anything of what it is happening.
Not surprisingly, imaging in the presence of scattering is a very active field of research.
There isn't a single best way on how to deal with scattering, and the answer depends a LOT on how much scattering we are talking about and its properties.
As a rule of thumb, the most complicated situation is where all light is multiply scattered.
Read 13 tweets
#arXiv Emergent Cosmology from Matrix Theory Matrix theory is a proposed non-perturbative definition of superstring theory in which space is emergent. It leads to a non-singular cosmology solving the horizon problem of Standard Big Bang. cc @GastonGiribet
"The new scenario does not involve a period of inflationary expansion. Since it is based on a nonperturbative approach to superstring theory, the scenario is free from any swampland constraints, consistency conditions which rule out many inflationary models."
"There are other approaches to obtaining space-time and cosmology from matrix theory. The new work is different in that space-time is obtained directly from the matrix theory."
Read 4 tweets
#arXiv Three Impossible Theories, by Leonard Susskind "Black hole central dogma: as seen from the outside, a black hole can be described in terms of a quantum system with Area/(4 G) degrees of freedom, which evolves unitarily under time evolution."
"Analogously, cosmological central dogma: as seen from a causal patch a cosmological space-time can be described in terms of an isolated quantum system with Area/(4 G) degrees of freedom, which evolves unitarily under time evolution."
"Area refers to the area of the cosmological horizon (for de Sitter space metric Area = 4πR²). Hence, the second law of thermodynamics must be respected by all observations within a causal patch. Naively the second law means the entropy never decreases."
Read 7 tweets
#FacebookAIResearch #arXiv Deep Learning for Symbolic Mathematics "Results that outperform commercial Computer Algebra Systems such as Matlab or Mathematica."
#FacebookAIResearch Learning advanced mathematical computations from examples "Neural networks nar perfect prediction of qualitative characteristics, and good approximations of numerical features of differential systems."
#QuantaMagazine Symbolic Mathematics Finally Yields to Neural Networks. After translating some of math’s complicated equations, researchers have created an AI system that they hope will answer even bigger questions.… by @StephenOrnes
Read 6 tweets
1/n How can we identify and measure ‘emergent’ processes in complex systems? Here’s a new #arXiv preprint addressing this question, by Lionel Barnett (@LionelBarnett2) and me. It’s all about ‘Dynamical Independence’. Thread:
2/n Emergence is a popular but difficult to grasp concept. We all have strong intuitions about when macroscopic processes ‘emerge’ from the micro-level constituents – as in flocks (murmurations) of starlings.
3/n One context of particular interest is how, or whether, mental states – perhaps even conscious states – might ‘emerge’ from the collective dynamics of their underlying neural populations.
Read 14 tweets
#arXiv Stellarators as a Fast Path to Fusion Seventeen years after the beginning of the fusion program, sentiments changed in 1968 when tokamaks were shown to have confinement properties far beyond those of any other fusion concept.
Nonetheless, the absence of robust stability was concerning since that had been known from early tokamak experiments. An invention is required for a tokamak to have sufficient passive stability for a power plant.
Is the stellarator the required invention that makes the least change in the fundamental concept? Stellarators are intrinsically strongly non-axisymmetric, which means careful design is required for adequate confinement; the power plant construction requires roughly five years.
Read 5 tweets
🚨🚨 New working paper 🚨🚨

I examine what @PeoplesMomentum actually did for the @UKLabour during the 2017 Election #Facebook campaign - I develop further the concept of 'satellite campaigning' and originate the idea of 'Janus-faced campaigns' #arxiv 1/…
I find the #2017GE fascinating because it disrupts the targeted ads are everything narrative. Something was in the water and it was visible via Facebook. Since Facebook's audience has changed, but I still feel the case highlights the vitality of organic campaigns/satellites 2/ Image
#Methods: Quantitative content analysis of the organic posts sent by the pages one month before election day (excluding election day). I used over 800 variables, only some are in the paper. Others will be used later on. 3/
Read 11 tweets
The size of #NLP models have increased enormously, growing to millions, or even billions, of parameters, along with a significant increase in the financial cost and carbon emissions. ASAPP Reducing the High Cos...
The cost associated with training large models limits the #AIresearch community's ability to innovate, because a research project often needs a lot of experimentation.
Consider training a top-performing #LanguageModel on the Billion Word benchmark. A single experiment would take 384 GPU days (6 days*64 V100 #GPUs, or as much as $36,000 using AWS on-demand instances)
Read 21 tweets
📢🗳️ #poll is over. Check-out the results about double-blind (DBP) vs. single-blind (SBP) policy preference for #CVPR2021. Thanks 4 participating ❤️

🧵 With the reason behind it, *my* #takeaways, and open #questions

c.c. @CVPR @eccvconf @ICCV_2021

TLDR; I feel that #reviewer judgement is RUINed by #oveR #enthUsiam, #PR & #clIeNtelism. RUIN got into our #peer_review system by using #arxiv. Our #players weren't ready for this. The community is more diverse, bigger and the scale of the conferences is unprecedented.
🙋🏽🙋🏿 if it's OK to run a startup in this way. We become viral & need to keep up the pace.
Actions. Increase the number of employees, (hopefully) hiring ppl with the bare minimum badges. Provide no training whatsoever, only updating the company guidelines.

Read 12 tweets
After years of efforts,

The Energy-Participation Ratio (EPR) for the design and quantization of a #quantumcomputing chip now on the #arXiv

Glad the EPR method seems to already be useful to some 25+ groups…

Send comments/questions
“Where is the energy?”

is the only question you need to ask & answer to design both the dissipative budget and the quantum #Hamiltonian of any #quantum device

— irrelevant of its topology or the constitution of its nonlinear elements.
The answer to this question is the Energy-Participation Ratio (EPR) p — a number between 0 and 1,

which says how much of the energy of a mode is stored in a nonlinear element (e.g., Josephson junction).

The EPR is the key to the classical-to-quantum bridge for #quantumcircuits
Read 8 tweets
Anoche apareció en el #arxiv… el modelo epidemiológico usado por el gobierno federal. Va una breve explicacion.
Este es un modelo compartamental. Si no han oído de ellos, la idea fundamental es dividir a la población en varias "cajas" dependiendo de su estado de enfermedad.
Ejemplo sencillo: el modelo SIR. En este, hay una población [S]usceptible, una [I]nfectada, y otra [R]ecuperada.
Hay funciones, que dependen del tiempo, del número de elementos en otros compartimentos, y de parámetros de la enfermedad que nos pasa de S->I y de I->R. Image
Read 28 tweets
¿Pueden las estrellas regular la formación de generaciones futuras de estrellas? En este nuevo artículo #MNRAS de #astronomia presentamos evidencia directa de ese efecto. La explicación a todo público en el hilo👇 1/7 @inaoe_mx Image
Las estrellas nacen de la contracción de nubes frías de gas molecular. Si no hay algún mecanismo que pare la contracción, toda la masa de gas daría lugar a una primera generación de estrellas y poco más. Pero en las galaxias vemos estrellas de diferentes generaciones. 2/7 Image
Durante años hemos debatido sobre si una inyección de energía de las propias estrellas, en forma de vientos estelares o explosiones de supernova, podría ralentizar la formación estelar. La alternativa es que sean vientos producidos por el hoyo negro supermasivo central. 3/7 Image
Read 9 tweets
Gravitational corrections to the value of the muon #gminus2 - a quick roundup.

It all started with this paper - - the 3rd in a trilogy. Claimed that a correction due to Earth's gravity should alleviate tension between theory and experiment:
In a blog post -… - @lumidek points out (among other things) that if you're considering corrections due to the gravitational potential, the Sun's potential should dominate, not the Earth's
On twitter (and in a blog post) -… - @PitifulRed offers a short proof of why the gravitational correction must be zero.
Read 7 tweets

Related hashtags

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3.00/month or $30.00/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!