Multi-Agent LLMs
Last edited: March 3, 2026Background
- originally, multi-agent team pre-assigns roles
- LLMs are heterogeneous, but they are treated homogeneously
- problem decomposition is hard
Eval
synergy
- weak synergy: team >= average member
- strong synergy: team >= best member
Human teams reliably achieve strong synergy IFF when expert identity is given (e.g., the teams easily know who is the expert).
Dataset
NASA moon survival / lost at seay
Rank 15 items by importance
Student body president
Different people are given different information + shared info. Hidden-profile (shared info + unique info must be paired to reveal the right one.)
o
Last edited: March 3, 2026SU-EE364A MAR102026
Last edited: March 3, 2026convex-concave problems
Heuristic method for solving a specific type of non-convex problem. Solves a small sequence of convex problems.
difference-of-convex function
For:
\begin{equation} h\qty(x) = f\qty(x) - g\qty(x) \end{equation}
for convex \(f\qty(x)\) and \(g\qty(x)\).
some examples
- a convex quadratic, except the \(P\) in \(\qty(\frac{1}{2}) x^{T}Px\) is not PSD; we express this in terms of \(P = P_{\text{psd}} - P_{\text{nsd}}\). And thus we can get: \(\qty(\frac{1}{2})x^{T}P_{\text{psd}} x - \qty(\frac{1}{2})x^{T}P_{\text{nsd}}x\)
majorization
Taylor approximation:
a
Last edited: March 3, 2026Convex Optimization Index
Last edited: March 3, 2026EE364A.stanford.edu