脳楽

NouraKu is a visual educational app that allows you to learn about the mechanisms of neuroscience and neural computation by "seeing" them.

This is the 14th installment in the series, following Math, Objects, Electronics, Waves, Heat, Stars, Light, Quantity, Flow, Confirmation, Geometry, Chemistry, and Germ. This time, the theme is "Brain and Learning." From single-neuron action potentials to synaptic plasticity, neural circuits, cognitive illusions, and reinforcement learning—you can learn about 10 areas that run vertically through neurobiology (wet) and neural computation (dry) theory all at once with beautiful simulations and interactions.

[10 Functions]

- Neuronal Firing (Action Potential)
Generates a single-neuron action potential using the Hodgkin-Huxley equation. When the stimulation current exceeds the threshold, you can observe the chain reaction of Na+/K+ channel opening and closing, creating an "all or nothing" spike, using membrane potential plots and gate variable bars.

- Synaptic Transmission (Synapse): Visualizes how presynaptic neuron spikes generate alpha-function conductance, which is then added to postsynaptic LIF neurons as EPSP/IPSP. Switch between excitability and inhibition with a single tap.

- Neural Circuit: Arranges 100 LIF neurons in a grid, manipulating the excitation/inhibition balance and connection strength with sliders. Experience the rich synchronous/asynchronous patterns that arise from critical balance, and the silence/runaway behavior when this balance is disrupted.

- Brain Waves: Synthesizes 5-band oscillators (δ, θ, α, β, γ) using an arousal level slider. Observe the shift in dominant rhythms and the synthesized waveform in real time, from deep sleep to a focused state.

- Hebb Learning (STDP): Displays the Δw vs Δt curve of spike-timing-dependent plasticity. Moving the marker with the Δt slider reveals the asymmetry that determines whether the presynaptic firing order leads to synaptic strengthening (LTP) or weakening (LTD).

- Receptive Field
V1: A Gabor filter, a mathematical model of simple cells, is applied to the synthesized image in real time. Changing the orientation θ, spatial frequency, and σ reveals how the human visual system decomposes the world into "lines" and "directions."

- Perceptron
A Rosenblatt perceptron that linearly separates two Gaussian clusters. The weights are updated step by step, allowing you to track the movement of the decision boundary step by step.

- Reinforcement Learning
Q: A learning agent searches for rewards in a 6x6 maze. Adjusting the learning rate, search rate, and discount rate reveals the optimal action for each cell with arrows, and the episode-average reward and TD error δ are displayed in real time.

- Cognitive Illusions
Switch between three types of illusions: the Müller-Lyer arrow illusion, the Stroop task (reaction time measurement), and motor aftereffect. You can use your own brain as the experimental subject.

- Memory
Associative memory using the Hopfield network. When a noisy input is input to a weight matrix containing three internal patterns, the original pattern is recalled in one step update.

[Features]
- Includes explanation sheets for formulas and principles for all functions
- Implements textbook-level formulas such as Hodgkin-Huxley, STDP, TD error, and Hopfield
- Supports Japanese / English
- Supports Dark Mode / Light Mode
- Intuitive and beautiful UI
- Advertisements are limited to banners at the bottom of the screen

[Recommended for:]
- Students studying neuroscience, cognitive science, and machine learning
- Those who want to intuitively understand the relationship between the brain and AI
- Teachers who want to use it as teaching material
- Those who want to experience the differences and continuity between neurons and perceptrons