#Tag · bonfire.cafe

#NeuralPlasticity & #learning are distinct but interrelated processes. #Plasticity denotes biological change in #NeuralSystems, while learning is its functional expression in #NetworkDynamics & #behavior. Learning arises from coordinated plastic processes, reshaping #NeuralStateSpace & #attractors to support stable yet flexible representations. Here's a new post on these concepts & their implications for #ComputationalNeuroscience:

🌍https://www.fabriziomusacchio.com/blog/2026-02-02-neural_plasticity_and_learning/

#CompNeuro #Neuroscience

3 media

Schematic illustration of neuronal plasticity across scales. (a) Schematic representation of key cellular elements (neuronal and glial cells) involved in the neuroplasticity (NP) process (Neuronal and Glia plasticity) as well as subcellular compartments (Synaptic, Dendritic, Axonal, Neuromuscular Plasticity). (b) Schematic diagram representing how repetitive synaptic stimulations repetitive LTPs are linked to molecular changes (Doted circle in (a)), and the generation of dendrite and spine remodeling. (c) Schematic illustration showing some of the changes in dendrite formations (dotted dendrite spines) following learning (synaptic re-wiring) and memory formation (synaptic stabilization). (d) Schematic illustration showing axonal mechanism of neuroplasticity and repair (rerouting and sprouting) following brain injury. Abbreviations: LTP, long-term potential. Source: Wikimediaꜛ (license: CC BY-SA 4.0; original figure from Gatto, 2020ꜛ).

Neural plasticity and learning are fundamental adaptive processes that enable the brain to modify its structure and function in response to experience. Shown is a simplified schematic illustration of plastic changes induced by intensive practice of a specific skill at multiple organizational levels. Top (behavior): Repeated practice improves behavioral performance, illustrated here by increased accuracy and reduced variability when practicing a motor task such as throwing darts. Middle (cortex): At the cortical level, practice leads to a reorganization of functional representations, with task relevant neural populations becoming more strongly engaged and more precisely tuned. Bottom (neuron): At the cellular level, these changes are supported by synaptic and structural plasticity, including modifications of synaptic strength and local connectivity at individual neurons. Together, these coordinated changes across scales illustrate how learning emerges from local plastic mechanisms to produce stable improvements in behavior. Source: Wikimedia Commonsꜛ (license: CC BY 3.0).

Neural population dynamics as an embedding of task structure into neural state space. (A) Behavioral paradigm illustrating a reaching movement to a single target. The monkey initiates a movement from a fixed starting position and reaches toward a specified goal. In this simple condition, the task is fully parameterized by time
t
along the movement trajectory. (B) Trial-averaged population firing rates of many simultaneously recorded neurons during the reach. Each trace represents the activity of one neuron as a function of time, illustrating how complex, heterogeneous single-neuron responses unfold during a structured behavior. (C) The same population activity represented as a trajectory in neural state space. At each time point, the joint activity of all recorded neurons defines a single point in a high-dimensional space, with neural dynamics corresponding to a continuous trajectory through this space. Temporal evolution of behavior is thus mapped onto a geometric path in population activity space. (D) Extension of the task to reaching movements in multiple directions. In this case, behavior is no longer described by time alone but by two task variables: Time within the movement and reach angle. (E) The resulting task manifold, here shown schematically as a low-dimensional cylinder parameterized by time and reach direction. Each point on this manifold corresponds to a specific behavioral state of the task. (F) The neural data manifold obtained from population activity. — Neural population dynamics as an embedding of task structure into neural state space. (A) Behavioral paradigm illustrating a reaching movement to a single target. The monkey initiates a movement from a fixed starting position and reaches toward a specified goal. In this simple condition, the task is fully parameterized by time t along the movement trajectory. (B) Trial-averaged population firing rates of many simultaneously recorded neurons during the reach. Each trace represents the activity of one neuron as a function of time, illustrating how complex, heterogeneous single-neuron responses unfold during a structured behavior. (C) The same population activity represented as a trajectory in neural state space. At each time point, the joint activity of all recorded neurons defines a single point in a high-dimensional space, with neural dynamics corresponding to a continuous trajectory through this space. Temporal evolution of behavior is thus mapped onto a geometric path in population activity space. (D) Extension of the task to reaching movements in multiple directions. In this case, behavior is no longer described by time alone but by two task variables: Time within the movement and reach angle. (E) The resulting task manifold, here shown schematically as a low-dimensional cylinder parameterized by time and reach direction. Each point on this manifold corresponds to a specific behavioral state of the task. (F) The neural data manifold obtained from population activity.

Fabrizio Musacchio

Neural plasticity and learning: A computational perspective

After discussing structural plasticity in the previous post, we now take a broader look at neural plasticity and learning from a computational perspective. What are the main forms of plasticity, how do they relate to learning, and how can we formalize these concepts in models of neural dynamics? In this post, we explore these questions and propose a unifying framework.

Federica Gazzelloni

@fgazzelloni@fosstodon.org · 3 days ago

📬 Issue No. 5 of Notes on Data and Learning highlights the skill of omission in education.

In teaching data, statistics, or modeling, deciding what to exclude is as crucial as what to include, preventing overwhelm while keeping depth.

Drawing from my Health Metrics and Infectious Diseases experience, I discuss how teaching and writing prioritize over completeness.

👉 Full issue on Substack: https://federicagazzelloni.substack.com

#Teaching #Learning #DataScience #Education #Substack

Ulrike Hahn boosted

Fabrizio Musacchio

@FabMusacchio@mastodon.social · 5 days ago

Incorporating structural #plasticity in #SpikingNeuralNetworks ( #SNN) enables dynamic #synaptic connectivity, reflecting the #brain's adaptability. By modeling synaptic growth and pruning based on #calcium concentration, we can simulate processes such as #learning and #MemoryFormation. In this post, I reproduce the #NESTSimulator tutorial on structural plasticity, demonstrating its impact on network stability and #homeostasis:

🌍 https://www.fabriziomusacchio.com/blog/2026-02-01-structural_plasticity/

#CompNeuro #Neuroscience #NeuralNetworks

2 media

Sketch illustrating structural plasticity during learning and memory formation. The sketch illustrates the dynamic remodeling of synaptic connectivity through dendritic spine turnover. Left: Under baseline conditions, synaptic networks exhibit continuous formation and elimination of dendritic spines, reflecting ongoing structural plasticity. Middle: During learning or learning related activity, this baseline turnover is transiently increased, leading to enhanced formation and pruning of synaptic contacts. Newly formed spines preferentially emerge near previously activated synapses, promoting the local clustering of synaptic inputs and enabling adaptive rewiring of circuits. Right: A subset of newly formed and activated synapses becomes selectively stabilized, providing a structural substrate for the long term retention of behaviorally relevant connections and memory traces. Source: Bernardinelli, Y., Nikonenko, I., Muller, D., Structural plasticity: mechanisms and contribution to developmental psychiatric disorders, Frontiers in Neuroanatomy, 2014, 8:123, DOI 10.3389/fnana.2014.00123ꜛ (license: CC BY 4.0).

Simulation results of the structural plasticity model. The plot shows the temporal evolution of the mean calcium concentration of excitatory and inhibitory neurons (blue and red lines, respectively) and the total number of connections of excitatory and inhibitory neurons (magenta and black dashed lines, respectively). The horizontal lines represent the growth curves for excitatory (blue) and inhibitory (red) neurons. The model demonstrates how the network’s connectivity changes over time based on the calcium concentration in the neurons. The growth curves determine the threshold at which synapses are created or pruned.

Fabrizio Musacchio

Incorporating structural plasticity in neural network models

In standard spiking neural networks (SNN), synaptic connections between neurons are typically fixed or change only according to specific plasticity rules, such as Hebbian learning or Spike-Timing Dependent Plasticity (STDP). However, the brain’s connectivity is not static. Neurons can grow and retract synapses in response to activity levels and environmental conditions. A phenomenon known as structural plasticity. This process plays a crucial role in learning and memory formation in the brain. To illustrate how structural plasticity can be modeled in spiking neural networks, in this post, we will use the NEST Simulator and replicate the tutorial on ‘Structural Plasticity.

Fabrizio Musacchio

@FabMusacchio@mastodon.social · 5 days ago

🌍 https://www.fabriziomusacchio.com/blog/2026-02-01-structural_plasticity/

#CompNeuro #Neuroscience #NeuralNetworks

2 media

Fabrizio Musacchio

Incorporating structural plasticity in neural network models

Chip Butty and 2 others boosted

Calligrafae

@welshpixie@mastodon.art · 2 months ago

I've published my digital Celtic Knotwork Workshop Booklet, yay!

https://ko-fi.com/s/f1a74eaf4b

It's 'Pay what you want over £5', and it's 10 worksheets (four of them are re-usable printable template sheets) that guide you through the basics to more complex techniques.

Huge thanks to the few of you who helped me test this and provided awesome feedback on earlier versions <3

#MastoArt #CelticKnot #ArtTutorial #Learning

4 media

Crop of one of the worksheets talking about knotwork construction with demonstrations on a grid

Crop of one of the template pages with grids ready to fill in

Crop of one of the tutorial pages talking about knot formation

Crop of one of the tutorial pages talking about knot formation, there are illustrations in purple and orange

Hacker News

@h4ckernews@mastodon.social · 3 weeks ago

Design Thinking Books You Must Read

https://www.designorate.com/design-thinking-books/

#HackerNews #DesignThinking #Books #MustRead #Innovation #Learning #Creativity

Designorate

Design Thinking Books You Must Read (updated)

An updated list of design thinking books and papers to understand the principles underpinning the design thinking process and its practice.

Hacker News

@h4ckernews@mastodon.social · 3 weeks ago

GenAI, the Snake Eating Its Own Tail

https://www.ybrikman.com/blog/2026/01/21/gen-ai-snake-eating-its-own-tail/

#HackerNews #GenAI #Snake #Eating #Its #Own #Tail #Technology #Innovation #AI #Ethics #Machine #Learning

Yevgeniy Brikman

GenAI, The Snake Eating Its Own Tail: How tools like ChatGPT and Claude are destroying the ecosystems they rely on, and what to do about it

Generative artificial intelligence (GenAI) tools such as ChatGPT and Claude have two superpowers. The first superpower is a boon: they can dramatically increase ...

Chip Butty and 1 other boosted

The Kid Should See This 🌈🪐✨

@tksst@fediscience.org · 3 weeks ago

🐈👅 When a cat licks you, they aren't just trying to clean you – they’re designating you as a "preferred associate."

Animal behaviorists explain that this social grooming, or allogrooming, is a #behavior learned from their mothers to strengthen bonds and show trust.

👉 https://www.popsci.com/environment/why-cats-lick-you/

#science #cats #biology #animals #stem #pets #nature #learning #education

Popular Science

Why do cats lick you? An expert explains.

Grooming is only one way cats say, "I love you.”

The Kid Should See This 🌈🪐✨

@tksst@fediscience.org · 3 weeks ago

🐈👅 When a cat licks you, they aren't just trying to clean you – they’re designating you as a "preferred associate."

Animal behaviorists explain that this social grooming, or allogrooming, is a #behavior learned from their mothers to strengthen bonds and show trust.

👉 https://www.popsci.com/environment/why-cats-lick-you/

#science #cats #biology #animals #stem #pets #nature #learning #education

Popular Science

Why do cats lick you? An expert explains.

Grooming is only one way cats say, "I love you.”

Hacker News

@h4ckernews@mastodon.social · 3 weeks ago

LLM Structured Outputs Handbook

https://nanonets.com/cookbooks/structured-llm-outputs

#HackerNews #LLM #Structured #Outputs #Handbook #LLM #Handbook #Structured #Outputs #Machine #Learning #AI #Resources

Structured LLM outputs

A practical handbook for developers using LLMs to get structured outputs.

Hacker News

@h4ckernews@mastodon.social · 4 weeks ago

Choosing learning over autopilot

https://anniecherkaev.com/choosing-learning-over-autopilot

#HackerNews #ChoosingLearning #OverAutopilot #Learning #Mindset #PersonalGrowth #ContinuousLearning #SelfImprovement

choosing learning over autopilot

I use ai coding tools a lot. I love them. I’m all-in on ai tools. They unlock doors that let me do things that I cannot do with my human hands alone.

Hacker News

@h4ckernews@mastodon.social · 4 weeks ago

Reproducing DeepSeek's MHC: When Residual Connections Explode

https://taylorkolasinski.com/notes/mhc-reproduction/

#HackerNews #Reproducing #DeepSeek #MHC #Residual #Connections #Explode #Deep #Learning #AI #Research

Taylor Kolasinski

DeepSeek's mHC: When Residual Connections Explode - Taylor Kolasinski

Taylor Kolasinski - Engineering at FlowMode. ML systems & research, reinforcement learning, robotics. Based in Brooklyn, NY.

hypebot boosted

Kushal Das :python: :tor:

@kushal@toots.dgplug.org · last month

Introducing #ektupy https://kushaldas.in/posts/introducing-ektupy.html #python #learning #kids #scratch #browser #django #community #pyscript #teaching #programming

Please share it across and let me know what you made using this :)

3 media

Kushal Das

FOSS and life. Kushal Das talks here.

Fabrizio Musacchio

@FabMusacchio@mastodon.social · last month

🧠 New preprint by Shervani-Tabar, Brincat & @ekmiller on emergent #TravelingWaves in #RNN.

By aligning RNN dynamics to an empirically measured #NeuralManifold, they show that task-relevant TW can emerge through #learning, w/o hard-coding wave dynamics or connectivity. The cool thing here is that the waves are not imposed or engineered, but emerge naturally from learning under #BiologicallyPlausible constraints:

🌍 https://doi.org/10.64898/2026.01.08.698281

#Neuroscience #CompNeuro #NeuralDynamics #WorkingMemory