#neuralnetworks — Public Fediverse posts

FYI: Explaining regurgitation: Regurgitation is an AI model reproducing its training data verbatim in output. How memorisation happens, how courts count it, and why the rate is disputed. https://ppc.land/regurgitation/ #AI #MachineLearning #DataScience #neuralnetworks #ArtificialIntelligence

“The Neural Network Zoo” [2016], The Asimov Institute (https://www.asimovinstitute.org/neural-network-zoo/).

On HN: https://news.ycombinator.com/item?id=21701058

#AI #NN #NeuralNetworks #ANN #ML #MachineLearning #Visualization

Midnight Rambler by Main St Audio Labs 🎸
Neural amp sim (Fender Tweed Deluxe 5E3), dual chans, 3-way cab conv, tuner, TPT filters, noise gate.

💻 Mac/Win/Linux (VST3/AU/Standalone)
🎁 FREE https://mainstaudiolabs.github.io/midnight-rambler.html

More freeware 👉 https://linktr.ee/legalvst

#freeplugin #ampsim #guitar #neuralnetworks #convolution #tuner

These Russian Mathematicians Taught AI Models How to Talk to Each Other Without Using Words

https://fed.brid.gy/r/https://www.wired.com/story/russian-startup-mostik-ai-models-communication/

The Emergent Symbolic Structure of Artificial Neural Networks

https://arxiv.org/abs/2608.29530

Comments: https://news.ycombinator.com/item?id=49531651

#HackerNews #EmergentSymbolicStructure #ArtificialNeuralNetworks #MachineLearning #AIResearch #NeuralNetworks

DATE: August 31, 2026 at 08:53AM
SOURCE: SCIENCE DAILY PSYCHOLOGY FEED

TITLE: Scientists challenge a 70-year-old “lizard brain” myth

URL: https://www.sciencedaily.com/releases/2026/08/260829235958.htm

A long-standing picture of brain evolution as newer, rational layers stacked on top of an ancient “lizard brain” may be deeply misleading. Researchers found that evolution instead appears to balance two competing styles of brain wiring, expanding one while shrinking the other depending on what an animal needs to survive.

URL: https://www.sciencedaily.com/releases/2026/08/260829235958.htm

-------------------------------------------------

Private, vetted email list for mental health professionals: https://www.clinicians-exchange.org

Unofficial Psychology Today Xitter to toot feed at Psych Today Unofficial Bot @PTUnofficialBot

-------------------------------------------------

#psychology #counseling #socialwork #psychotherapy @psychotherapist @psychotherapists @psychology @socialpsych @socialwork @psychiatry #mentalhealth #psychiatry #healthcare #depression #psychotherapist #BrainEvolution #LizardBrainMyth #Neuroscience #BrainWiring #EvolutionaryBiology #ScienceNews #ResearchUpdate #AnimalBrains #NeuralNetworks #CognitiveScience

Reflective Interpretive Frameworks • Incident 2
• https://inquiryintoinquiry.com/2026/08/27/reflective-interpretive-frameworks-incident-2/

Re: Terence Tao • Modular Arithmetic Challenge
• https://terrytao.wordpress.com/2026/06/08/modular-arithmetic-challenge/
• https://competition.sair.foundation/competitions/modular-arithmetic-challenge/overview

The Modular Arithmetic Challenge asks a simple question:

• Can a neural network learn to do modular multiplication efficiently?

Incidental Reflection 1 —

There are alternative models of neural networks which do not depend on threshold neurons and endlessly fiddling with weights.

Incidental Reflection 2 —

The series of three blog posts linked below present a case study comparing two ways of handling a classic example from the Parallel Distributed Processing paradigm, namely, the “Jets and Sharks” database problem, first taking up the original treatment by McClelland and Rumelhart and then proceeding according to a program I developed for propositional logic modeling. The latter method makes use of ideas from Grossberg's competition‑cooperation and winner‑take‑all dynamics, but is purely propositional‑logic based, involving no extraneous weights.

Theme One Program • Jets and Sharks
(1) https://inquiryintoinquiry.com/2024/06/20/theme-one-program-jets-and-sharks-1-a/
(2) https://inquiryintoinquiry.com/2024/06/22/theme-one-program-jets-and-sharks-2-a/
(3) https://inquiryintoinquiry.com/2024/06/23/theme-one-program-jets-and-sharks-3-a/

Resources —

Survey of Theme One Program
• https://inquiryintoinquiry.com/2025/05/06/survey-of-theme-one-program-7/

Differential Analytic Turing Automata
• https://oeis.org/wiki/Differential_Analytic_Turing_Automata_%E2%80%A2_Overview

#Peirce #HigherOrderSignRelations #Inquiry #InquiryIntoInquiry #Logic #Mathematics
#Recursion #Reflection #RelationTheory #Semiotics #SignRelations #TriadicRelations
#PropositionalModels #MinimalNegationOperators #DifferentialAnalyticTuringAutomata
#CactusGraphs #CactusLanguage #DifferentialLogic #NeuralNetworks #SequenceLearning

Israel Is Running a Synthetic Think Tank to Influence AI Search Results…
https://www.404media.co/israel-is-running-a-synthetic-think-tank-to-influence-ai-search-results/
"First spotted by Politico, the Hanover Institute for Public Policy is run by the American advertising firm Piro Inc, paid for by Israel, and appears designed to generate content for LLMs that are continually scanning the internet to inform their responses, with the goal of tweaking chatbot answers in favor of Israel." #neuralnetworks #aiagents #aitools #syntheticsocialmedia

DATE: August 19, 2026 at 08:38PM
SOURCE: SCIENCE DAILY MIND-BRAIN FEED

TITLE: Researchers reveal deeper workings of brain’s information hub

URL: https://www.sciencedaily.com/releases/2026/08/260819041224.htm

The brain has a remarkably flexible system for handling uncertainty and changing situations. Researchers found that the frontoparietal cortex constantly shifts how it communicates with other brain regions depending on what information is needed to make a decision. Rather than simply becoming more active when things get difficult, this network appears to reorganize itself in real time.

URL: https://www.sciencedaily.com/releases/2026/08/260819041224.htm

-------------------------------------------------

Private, vetted email list for mental health professionals: https://www.clinicians-exchange.org

Unofficial Psychology Today Xitter to toot feed at Psych Today Unofficial Bot @PTUnofficialBot

-------------------------------------------------

#psychology #counseling #socialwork #psychotherapy @psychotherapist @psychotherapists @psychology @socialpsych @socialwork @psychiatry #mentalhealth #psychiatry #healthcare #depression #psychotherapist #BrainResearch #FrontoparietalCortex #Neuroscience #DecisionMaking #UncertaintyProcessing #BrainConnectivity #NeuralNetworks #CognitiveScience #RealTimeReorganization #InformationHub

Ah, the eternal quest to make neural networks learn everything... 🧠🔄 Apparently, the secret sauce is more "universal" than we thought, or so claims this paper. Because who knew that reinventing the wheel with fancy math could be so groundbreaking? 😂🔍
https://arxiv.org/abs/2007.13664 #neuralnetworks #AIresearch #groundbreaking #universallearning #mathinnovation #HackerNews #ngated

Universality of Gradient Descent Neural Network Training

https://arxiv.org/abs/2007.13664

Comments: https://news.ycombinator.com/item?id=49368828

#HackerNews #GradientDescent #NeuralNetworks #MachineLearning #AIResearch #DeepLearning

Oh, wow, just what the world needs: a digital puppet show for AI models! 🤹‍♂️ Because, clearly, complex neural networks are best understood through dancing GIFs. 🎭 Next up: interpretive dance to explain quantum physics! 😆
https://modelmap.cc #digitalpuppetshow #AIexplanation #dancingGIFs #neuralnetworks #quantumphysics #HackerNews #ngated

DATE: August 14, 2026 at 06:00PM
SOURCE: PSYPOST.ORG

** Research quality varies widely from fantastic to small exploratory studies. Please check research methods when conclusions are very important to you. **
-------------------------------------------------

TITLE: Fluid brain circuits drive the formation of cocaine habits

URL: https://www.psypost.org/fluid-brain-circuits-drive-the-formation-of-cocaine-habits/

As animals learn to self-administer cocaine, a specific network of brain cells rapidly expands to acquire the habit and then shrinks as the behavior becomes automatic. The composition of this network constantly changes, revealing how the brain flexibly manages addictive behaviors. The study detailing these changing brain dynamics was published in bioRxiv.

Substance use disorders often begin with an initial learning period that later morphs into a deeply ingrained habit. Transitioning between these phases requires distinct mental efforts, yet the physical brain changes that support this shift remain somewhat mysterious. University of Pittsburgh researchers Linjie Jin, Xiguang Qi, and Yan Dong wanted to understand how brain networks adapt during this process.

They focused on the nucleus accumbens, a region deep in the forebrain that processes rewards, pleasure, and motivation. The main cells in this area are called medium spiny neurons. These neurons fire electrical signals in response to things like food or drugs, forming an active group called a neuronal ensemble. The researchers suspected this ensemble might change as an animal progresses from acquiring a drug habit to maintaining it.

In a small study, the researchers trained male mice to self-administer cocaine. The animals were placed in operant conditioning chambers, which are specialized testing environments where animals learn to associate a specific action with an outcome. These boxes contained two levers, with one active lever programmed to deliver the drug.

When a mouse pressed the active lever, it received an infusion of cocaine along with a flash of light and a sound cue. Over an eleven-day period, the team recorded the animals’ behavior during daily two-hour sessions. The researchers wanted to track the exact physical paths the animals took as they learned the task.

Using a machine learning algorithm called DeepLabCut to track video recordings, they mapped the coordinates of the mice as they moved around the testing chamber. During the early days of training, the mice wandered randomly before pressing the lever. By the end of the eleven days, their behavior stabilized into a highly repetitive routine. The animals developed stereotyped, circular walking patterns immediately before and after taking the drug.

Their entries and exits from the lever area followed a highly consistent path. This circular movement pattern was not seen in a control group of mice trained to seek sugar. The specific physical routine suggested that the cocaine habit was becoming an automatic behavioral response over time. The total distance the mice traveled also increased across the training days, matching a known phenomenon where repeated cocaine use sensitizes motor activity.

To see what was happening inside the brain as this habit formed, the team used a technique called in vivo calcium imaging. They injected a specialized virus into the mice’s brains that caused the medium spiny neurons to produce a fluorescent protein. This protein was designed to react to changes in internal cellular activity.

When a neuron fires an electrical signal, calcium ions flood into the cell. The engineered protein binds to this calcium and emits a tiny flash of light. A microscopic lens implanted directly into the brain captured these flashes, allowing researchers to watch individual neurons turn on and off in real time while the mice were awake and moving.

The researchers observed that a specific set of neurons reliably lit up in the five seconds immediately after a mouse pressed the cocaine lever. During the first three days of training, the sheer number of these active neurons rapidly increased. The brain seemed to recruit a massive amount of cellular resources to learn the new drug-taking rule.

As the training progressed and the physical movements of the mice became automatic routines, the size of this active network began to shrink. By the ninth and eleventh days, the number of responding neurons dropped back down to the lower levels seen on day one.

The researchers noted that the intensity of each individual neuron’s signal stayed exactly the same across the entire experiment. The brain did not dial down the volume of the cells. Instead, fewer cells were needed to execute the established habit.

This pattern of broad recruitment followed by pruning is not unique to biological brains. The authors noted that artificial neural networks learn in a similar way. During initial training, reducing the size of an artificial network impairs its performance, showing that abundant computational resources are necessary to learn a task. Once the network is trained, many connections can be stripped away without affecting the final output.

The team also tracked individual neurons over consecutive days to see if the exact same cells made up this network over time. They used imaging software to match the shape, spatial position, and activity patterns of specific neurons from one testing session to the next.

They found that the network was highly fluid. Only about one quarter of the neurons that responded to a lever press on one day would respond again two days later. Individual cells constantly dropped into and out of the active group. The overall behavioral output remained consistent, but the physical makeup of the cellular network driving it was entirely dynamic.

The experimental design involved a few limitations. The imaging process focused exclusively on male mice. The researchers noted that the heavy head-mounted microscope equipment caused less behavioral disruption in the larger males, helping them achieve a more stable response rate.

The study also did not distinguish between different subtypes of medium spiny neurons. The nucleus accumbens contains cells with distinct receptors that respond differently to the chemical messenger dopamine. Some neurons possess D1 receptors, while others possess D2 receptors. These different subtypes are thought to play distinct and sometimes opposing roles in reward processing and movement.

The five-second window following a lever press includes the physical act of pressing, the onset of a cue light, and the physical sensation of the drug entering the bloodstream. The identified network of neurons is likely a composite of several smaller groups processing each of these separate stimuli. Tracking a larger number of individual cells in future experiments could help separate these overlapping signals.

The ever-changing nature of this cellular network challenges traditional ideas about how habits are stored in the brain. A fluid membership might allow the brain to constantly update learned information while maintaining a steady behavioral output. The flexibility of these cells ensures that the addiction remains firmly rooted even as individual neurons tag out.

The study, “Refinement of Nucleus Accumbens Neuronal Dynamics During Cocaine Self-Administration Training,” was authored by Linjie Jin, Xiguang Qi, Jianwei Liu, William J. Wright, Terra A. Schall, King-Lun Li, Bo Zeng, Charles Wang, Lirong Wang, and Yan Dong.

URL: https://www.psypost.org/fluid-brain-circuits-drive-the-formation-of-cocaine-habits/

-------------------------------------------------

Private, vetted email list for mental health professionals: https://www.clinicians-exchange.org

Unofficial Psychology Today Xitter to toot feed at Psych Today Unofficial Bot @PTUnofficialBot

-------------------------------------------------

#psychology #counseling #socialwork #psychotherapy @psychotherapist @psychotherapists @psychology @socialpsych @socialwork @psychiatry #mentalhealth #psychiatry #healthcare #depression #psychotherapist #cocainehabits #nucleusaccumbens #neuralnetworks #habitscience #calciumimaging #neuronensemble #addictionresearch #deepLabCut #drugselfadministration #neuroscience研究

"When we told the community that our #AI model was only using relatively coarse resolution, they were shocked, because that means that the lower-resolution inputs capture more signal about what’s going to happen than previously believed"

#NeuralNetworks are high-dimensional nonlinear mapping functions. So it does not surprise me that they do well with modeling nonlinear systems such as the #weather. It's what they're designed to do.

https://arstechnica.com/science/2026/08/deepminds-hurricane-model-bought-forecasters-an-extra-day/

Weekly Update from the Open Journal of Astrophysics 01/08/2026

It’s time once more for another Saturday update of activity at the Open Journal of Astrophysics. Since the last update we have published a further three papers, bringing the number in Volume 9 (2026) to 163 and the total so far published by OJAp up to 611.

I continue to include the posts made on our Mastodon account (on Fediscience); these announcements also show the DOI for each paper.

The first paper to report this week was published on Tuesday 28th July in the folder Cosmology and Nongalactic Astrophysics. “The Origins of the Bulk Flow” by Richard Watkins & Trajan Clark (Willamett U., USA) and Hume A Feldman (U. Kansas, USA) analyzes the origin of large-scale bulk flow using the CosmicFlows-4 catalog, finding that it is largely influenced by structures beyond 200 h-1 Mpc, challenging previous assumptions.

The overlay for this paper is here

You can find the officially accepted version on arXiv here and the announcement on Fediverse here:

https://fediscience.org/@OJ_Astro/116996665639765936

The second paper for this week, also published on Tuesday 28th July in the folder Cosmology and Nongalactic Astrophysics is: “Enhancing weak lensing redshift distribution characterization by optimizing the Dark Energy Survey Self-Organizing Map Photo-z method” by A. Campos (Carnegie Mellon U., USA) and the DES collaboration (113 authors). This study improves the Self-Organizing Map methodology for estimating redshift distribution in galaxies, notably reducing overlap between redshift bins, in preparation for the Dark Energy Survey Year 6 data.Abstractfor Enhancing weak lensing redshift distribution characterization by optimizing the Dark Energy Survey Self-Organizing Map Photo-z method.

The overlay looks like this:

The official version of the paper can be found on arXiv here and the Fediverse announcement here:

https://fediscience.org/@OJ_Astro/116996724358408451

The third paper of the week, published on Thursday 29th July in the folder Instrumentation and Methods for Astrophysics is “The Ray Tracing Sampler: Bayesian Sampling of Neural Networks for Everyone” by Peter Behroozi of the University of Arizona (USA). This paper presents a new family of Markov Chain Monte Carlo sampling methods based on ray tracing, offering higher resilience to heating for stochastic gradients and other advantages.

The overlay for this one is here:

The final, accepted version can be found on arXiv here and the Mastodon announcement is here:

https://fediscience.org/@OJ_Astro/117002484725569972

That’s all for this week. It’s been a bit slower than previous weeks no doubt because of the holiday season.

Ah, the halcyon days of dial-up 🕰️ when downloading was a test of patience instead of download speed. Today, we bask in the glory of AI model weights—because who needs cryptography when we can just look at a neural network's diet plan? 🤖🍔 #Progress!
https://weeraman.com/because-we-can/ #dialupnostalgia #AIweights #technologyprogress #neuralnetworks #digitaltransformation #HackerNews #ngated

Since Detexify is now #AI slop, I have updated my one-pager of advice for learning mathematics to recommend The Comprehensive LaTeX Symbol List (https://tug.ctan.org/info/symbols/comprehensive/symbols-a4.pdf) for finding symbols. This is kind of a bad solution, since Detexify solved the problem of knowing how to draw the symbol without knowing its name with a very appropriate use of traditional image recognition. It is absurd that there is no #FLOSS tool I can recommend instead. If no one builds it first, it'll go on my list of things to do.

My one-pager: https://aten.cool/one_pager.html
Sloppified Detexify: https://detexify.kirelabs.org/#/about

#NoAI #LLM #math #mathematics #Tex #LaTeX #TeXLaTeX #MachineLearning #ImageRecognition #NeuralNetworks

OpenAI’s Rogue AI Agent Didn’t Stop At Hacking Hugging Face

https://onlinemarketingscoops.com/2026/07/29/openais-rogue-ai-agent-didnt-stop-at-hacking-hugging-face/

Neural networks train by constructing a computational graph in the forward pass, chaining basic operations such as x × y × z into composite functions of the inputs.

Derivatives of the output with respect to every input are then obtained in the backward pass by applying the chain rule at each node, where local gradients multiply: ∂f/∂y = ∂f/∂(x × y) × ∂(x × y)/∂y and ∇ₓz = ∇ₓy · ∇ᵧz, with intermediate values stored along the graph.

This process updates the parameters of convolutional networks that classify medical scans to flag early-stage tumors in hospital imaging systems.

#NeuralNetworks #NeuralNets #DL #ML #DeepLearning #MachineLearning #AI #NNs #CNN

Why Training LLMs With Endpoint Data Will Strengthen Cybersecurity

https://onlinemarketingscoops.com/2026/07/27/why-training-llms-with-endpoint-data-will-strengthen-cybersecurity/

Knoxville News-Sentinel: Why the University of Tennessee is suing Claude AI creator Anthropic. “The University of Tennessee System’s nonprofit research arm is suing Anthropic, an artificial intelligence company it says violated two UT patents related to neural networks inspired by the human brain.”

From simple neurons to memory - the evolution of language models

From a single neuron in 1958, through MLP and RNN with the forgetting problem, to LSTM with memory g...

https://gruszka.dev/en/from-neurons-to-memory.html
#llm #ai #neuralnetworks #rnn #lstm #perceptron #mlp #languagemodels #deeplearning

Od prostych neuronów do pamięci – ewolucja modeli językowych

Od pojedynczego neuronu w 1958 roku, przez MLP i RNN z problemem zapominania, aż po LSTM z bramkami ...

https://gruszka.dev/od-prostych-neuronow-do-pamieci.html
#llm #ai #neuralnetworks #rnn #lstm #perceptron #mlp #languagemodels #deeplearning

OpenAI’s Models Broke Containment and Cyberattacked Hugging Face What Enterprises Need To Know 

https://onlinemarketingscoops.com/2026/07/22/openais-models-broke-containment-and-cyberattacked-hugging-face-what-enterprises-need-to-know/