Interactive Demo

Explore Integer Gate Logic (IGL) through an interactive mathematical laboratory.

Adjust node weights, bias, and input values to investigate how the patented non-monotonic activation function produces Boolean logic, graded nonlinear responses, and fixed-point integer computations. Experiment with AND, OR, XOR, XNOR, NAND, and NOR configurations; explore the activation curve and three-dimensional response surface; and observe how discrete parameter changes affect computational behavior. (also view the Demonstration Video linked below).

MLiglon • Interactive research demonstrator

Integer Gate Logic Laboratory

Experiment with exact Boolean gates, graded logic, non-monotonic activation, and scaled-integer inference—all calculated live in your browser.

2 · Patented five-region activation

zero → rise → fall → rise → plateau
00011011Live (X₁,X₂)

f(S) = 0 for S ≤ 0; S for 0 < S ≤ P; 2P − S for P < S ≤ 2P; S − 2P for 2P < S ≤ 3P; P for S > 3P.

3 · Live truth table

Exact XOR
X₁X₂SRegionY/PResponse
Truth-table rows are evaluated at binary inputs regardless of the live-input sliders. Click a row to set X₁ and X₂.

4 · Arithmetic trace

Region II · rising
Detailed expression and evaluation appear here.

5 · 3D response surface

X₁, X₂ ∈ [0,1]

The colored surface shows Y/P as both inputs vary continuously. The white dot marks the selected live input. In fixed-point mode, the surface reflects input and parameter quantization.

6 · Illustrative local candidate search

Not full IGL training

Explore six nearby changes (±step in W₁, W₂, B), score the four Boolean examples against a target, and accept the best improvement. No derivatives.

Mean squared error: · tested candidates/step: 6

No candidate steps run.

This one-node coordinate search is an educational illustration of discrete candidate evaluation—not a claim to reproduce the complete patented chain-isolation algorithm or benchmark training performance.
Reference and technical notesU.S. Patents 12,242,946 B1 · 12,488,222 B1
[1] This demonstrator uses the normalized five-region activation f(S;P): zero, rising, falling, rising, plateau; the discontinuities shown in some stylized infographics are not part of this continuous function. [2] “Exact gate” means all four normalized outputs match the Boolean truth table to numerical tolerance. “Graded” means they do not. [3] Fixed INT uses integer weights and inputs at scale P, computes Sₚ = round((W₁X₁ₚ + W₂X₂ₚ)/P) + B, and evaluates an integer piecewise activation with breakpoints P, 2P, 3P. [4] The 3D plot represents a two-input node, not a trained multilevel IGL network. [5] This educational simulation makes no hardware latency, power, or training-speed claims.