The SKA Explorer Suite — Nine Interactive Demos, Built and Served in the AI Agent Lab
One of the clearest ways to understand Structured Knowledge Accumulation (SKA) is to stop reading and start turning the dials. That's what the SKA Explorer suite is for: nine interactive demos that visualize the SKA forward-learning algorithm on MNIST — entropy dynamics, knowledge flow, the Lagrangian, entropy gradients — as live, adjustable plots rather than static equations.
The whole suite ships with the AI Agent Lab. Each demo is a small app you build in code-server (the lab's VS Code) and serve live from the lab itself — no separate deployment, no extra infrastructure. It's the run tier of read → run → learn: read the papers, then touch the theory.
The nine explorers
- SKA Explorer — interactive visualization of the SKA forward-learning algorithm on MNIST. Adjust the architecture, the number of steps K, and the learning budget τ to explore the entropy dynamics.
- Single Digit Explorer — explore the 4D entropy-state trajectory for a single digit across different architectures.
- SKA Entropy State Explorer — compare the normalized entropy-convergence state across architectures; each run is archived and compared.
- SKA Knowledge Flow Explorer — visualize the knowledge flow per layer across the forward-learning steps, and its trajectory in knowledge space.
- SKA Tensor Net Explorer — visualize the Tensor Net per layer; the zero-crossing marks the transition from unstructured to structured knowledge accumulation.
- SKA Lagrangian Explorer — visualize the Lagrangian trajectory of each layer in 3D: knowledge magnitude, knowledge flow, and Lagrangian value.
- SKA Entropy Gradient Explorer — compare the entropy and cosine-alignment trajectories with and without the ΔD term in the SKA gradient. Same architecture, same data, same weights — only the gradient formulation differs.
- SKA Per-Class Entropy Explorer — run SKA independently for each digit class and overlay the entropy trajectories. Each digit has its own model and weights, so its trajectory is a pure fingerprint of that digit.
- SKA Time-Invariance Explorer — fix the characteristic time T = η · K = 0.5 and run six different (η, K) pairs automatically. All the entropy and cosine curves collapse onto the same trajectory — revealing the intrinsic timescale of the algorithm.
Built and served in the lab
Each explorer is a Gradio app authored in code-server and launched straight from the environment — the same lab a subscriber gets. So the suite is also a demonstration of the lab itself: a place where SKA isn't just described, but built, run, and served. Read the trilogy, open an explorer, and watch the entropy fall.