I built TECS to explore a simple question:
What if data routing was dynamic, cyclic, and resource-constrained instead of static?
Most modern systems (including attention mechanisms) assume soft, continuous routing. TECS takes a different approach — elements physically “move” across slots over time, creating a discrete dynamic system:
[math]p_e(t+1) = (p_e(t) + v_e) \bmod N[/math]
and aggregation happens locally:
[math]Y_s(t) = F_s({\varphi_e(t) \mid p_e(t)=s})[/math]
This leads to interesting properties:
natural resource competition
emergent patterns and periodicity
hybrid behavior between simulation and learning systems
Originally, this started as a theoretical model of discrete dynamics, but it evolved into a working engine combining:
simulation (agent-like movement)
neural routing (Gumbel-Softmax)
and system-level constraints
I’m especially interested in feedback on:
potential applications (AI / distributed systems / simulations)
theoretical connections (dynamics, computation)
and ideas for extending it further
Happy to answer any questions 👇
