[01] Case Study
Scalable Choreography: SC2025
How do you make complex metadata management systems comprehensible to an audience of 18,000 supercomputing experts? "Sinwave" bridged the gap between embodied expression and high-performance computing (HPC) by using dance as a rigorous data visualization tool.
[02] Quick Facts
[03] Introduction
Bringing Dance to Super Computing
In collaboration with Arcitecta, we brought dance to SC2025 not as entertainment, but as a rigorous data visualization tool. The core premise was a metaphor for distributed computing: Distributed computing is like choreography.
Multiple processors (dancers) work simultaneously toward a unified goal, each following different algorithmic paths to achieve the same result.
[04] The Approach
Generative Systems & Patterns
We proved that interesting choreography could be created generatively using systems and patterns rather than traditional composition methods. The movement emerged from mathematical rules.
The "Sinwave" Algorithm
Each of the 5 dancers represented a data node, assigned a unique mathematical function:
- Dancer 1: Standard Sine Wave (A · sin(ωt))
- Dancer 2: Phase-Shifted Wave (offset by 90°)
- Dancer 3: Composite Wave (Sum of Dancer 1 + Dancer 2)
- Dancer 4 & 5: Variable amplitudes and phase shifts
From Math to Movement
The dancers were given a framework to generate 12 unique poses. These poses were mapped onto their assigned sine wave trajectories. The dancers' vertical positions on stage were dictated by their equation's value at that specific moment.
Complex, interlocking patterns emerged naturally. The "choreography" was simply the visualization of the math.
[05] Scalability
From 5 to Infinite
A key discovery was the inherent scalability of this generative approach. While we stopped at five dancers for the physical performance, the system is designed to apply to a scalable amount of dancers.
Because the choreography is rule-based rather than fixed, we could theoretically input an "infinite" number of dancers, each with a unique variable in the equation, and the system would generate a coherent, non-colliding group choreography.
Whether you have 5 data points or 5 million, the choreographic rules hold, creating a "Murmuration" effect that turns abstract data into a comprehensible organic flow.
[06] Process
The "Murmuration" Pivot
We initially tried to create a "visual murmuration" using feedback loops on video footage, tracking dancers by shirt color. However, shirt colors blended with skin tones, creating messy visuals.
The Solution: We pivoted to AI Pose Estimation (MediaPipe/PoseNet). Instead of relying on raw video pixels, we trained an AI to recognize discrete poses and map the skeleton joints in 3D space. This allowed us to strip away the "messiness" and reveal the pure mathematical skeleton underneath—the "bone structure" of the data.
The Human Element in Algorithms
Interestingly, the "errors" in the system were as valuable as the precision. Unlike digital processors, human dancers take up physical space. When five sine waves converged, dancers had to "negotiate" their proximity, subtly adjusting to avoid collision. This introduced a layer of organic "noise" or "latency" into the system, making the visualization more relatable.
[07] Conclusion
"Sinwave" demonstrated that dance is a viable medium for high-level data visualization. By applying generative systems and patterns, we transformed the abstract concept of parallel processing into an intuitive, scalable visual experience. The project showed that with the right rules, you can scale choreography indefinitely, turning the "black box" of algorithms into a visible, human performance.