SKA Time-Invariance Explorer

Fix the characteristic time T = η · K = 0.5 and run 6 different (η, K) pairs automatically. All entropy and cosine curves collapse onto the same trajectory — revealing the intrinsic timescale of the architecture [256, 128, 64, 10] on MNIST.

Architecture (fixed): [256, 128, 64, 10]

Characteristic time (fixed): T = η · K = 0.5

1 100
0 99

The 6 configurations

η K
0.0200 25
0.0100 50
0.0050 100
0.0033 150
0.0025 200
0.0010 500

Reference Paper

Abstract This paper aims to extend the Structured Knowledge Accumulation (SKA) framework recently proposed by mahi. We introduce two core concepts: the Tensor Net function and the characteristic time property of neural learning. First, we reinterpret the learning rate as a time step in a continuous system. This transforms neural learning from discrete optimization into continuous-time evolution. We show that learning dynamics remain consistent when the product of learning rate and iteration steps stays constant. This reveals a time-invariant behavior and identifies an intrinsic timescale of the network. Second, we define the Tensor Net function as a measure that captures the relationship between decision probabilities, entropy gradients, and knowledge change. Additionally, we define its zero-crossing as the equilibrium state between decision probabilities and entropy gradients. We show that the convergence of entropy and knowledge flow provides a natural stopping condition, replacing arbitrary thresholds with an information-theoretic criterion. We also establish that SKA dynamics satisfy a variational principle based on the Euler-Lagrange equation. These findings extend SKA into a continuous and self-organizing learning model. The framework links computational learning with physical systems that evolve by natural laws. By understanding learning as a time-based process, we open new directions for building efficient, robust, and biologically-inspired AI systems.


SKA Explorer Suite


About this App

Six (η, K) pairs all share the same characteristic time T = η · K = 0.5, the intrinsic timescale of the architecture [256, 128, 64, 10]. Each configuration is run independently and plotted as a function of the step index K. The trajectory shapes remain identical across all configurations while the amplitude scales with η — demonstrating that T is the true timescale of learning, not η or K individually. The characteristic time is the necessary time exposure of the sample to the learning system to complete. T = 0.5 is the characteristic time of the architecture [256, 128, 64, 10] on MNIST.