Contagion Tensor Measures LLM Output Coupling in Multi-Agent Systems.
Summary
This paper introduces the Contagion Tensor, a framework to quantify how LLM output distributions couple across modalities, agents, and time. It derives the Coupling Amplification Factor (CAF) metric, enabling quantitative auditing of claims about multi-agent system interactions, revealing significant image-condition super-linear effects in GPT-4o-mini.
Why it matters
For professionals designing, deploying, or auditing multi-agent LLM systems, the Contagion Tensor provides a critical tool to quantitatively understand and measure how different components and modalities interact. This enables more robust system design, better debugging, and verifiable claims about system behavior, especially in complex multi-modal scenarios.
How to implement this in your domain
- 1Adopt the Contagion Tensor framework to analyze interaction dynamics in your multi-agent LLM applications.
- 2Utilize the Coupling Amplification Factor (CAF) to quantitatively measure output-distribution coupling.
- 3Implement the transferable ablation protocol to distinguish genuine coupling from system design artifacts.
- 4Integrate these measurement tools into your LLM system development and testing pipelines.
- 5Use the insights gained to refine agent interaction strategies and improve multi-modal integration.
Who benefits
Key takeaways
- The Contagion Tensor quantifies LLM output coupling in multi-agent systems.
- CAF metrics provide unitless, baseline-referenced measurements of coupling.
- Ablation protocols help distinguish genuine coupling from design artifacts.
- GPT-4o-mini shows significant super-linear coupling with real vision input.
Original post by Zewen Liu
"arXiv:2606.28839v1 Announce Type: new Abstract: We introduce the Contagion Tensor, a measurement framework for quantifying how large language model (LLM) output distributions couple across modalities, agents, and time steps. From the tensor we derive the Coupling Amplification Fa…"
View on XOriginally posted by Zewen Liu on X · view source
Want to go deeper?
Turn these trends into skills with Learnijoy's hands-on AI & tech courses.
Explore coursesMore in AI Research
BaRA Improves LoRA Fine-Tuning with Adaptive Rank Allocation
Researchers introduce BaRA, a Bayesian Adaptive Rank Allocation framework for parameter-efficient fine-tuning, which dynamically adjusts adaptation capacity based on context. This method enhances predictive performance, robustness, and uncertainty calibration compared to standard LoRA and other Bayesian LoRA variants.
New Preconditioner Improves Deep Network Training Stability and Performance
Researchers introduce Dead-Direction Conditioners (DDC), a novel preconditioning method that leverages gauge-equivariant optimization to prevent deep network training from drifting along symmetry orbits. This technique improves model stability, reduces overfitting, and enhances performance in language and vision models.
SMDA Traces Training Data Influence on LLM Behavioral Policies
Researchers introduce Symbolic Mechanistic Data Attribution (SMDA), a framework that attributes specific training examples to the interpretable symbolic policies governing an LLM's high-level behavior. SMDA offers a fine-grained diagnostic tool to understand how training data shapes model decisions, revealing safety gaps and unintended influences.