papers/hanzo-federated-agents/sections/99-reproducibility-details.tex

\section{Reproducibility Details}

\label{app:reproducibility}

\begin{itemize}[leftmargin=1.5em]
    \item \textbf{Environment:} Python 3.11, LanceDB 0.3.0, sentence-transformers
          1.2.0 (all-MiniLM-L6-v2, $d=384$; we use $d=768$ in the paper via
          all-mpnet-base-v2).
    \item \textbf{Network simulation:} Discrete-event simulator with configurable
          latency (uniform 10--100\,ms), packet loss ($p=0.01$), and node failures.
    \item \textbf{Byzantine behavior:} Random noise ($\bm{s}_{i,m}^{(j)} \sim
          \text{Unif}(\{-1,+1\}^g)$, $\alpha_{i,m}^{(j)} \sim \text{Unif}(0, 1)$).
    \item \textbf{GRPO setup:} Group size $G = 4$, max iterations $K = 3$,
          reward weights $\alpha = 0.7$, $\beta = 0.3$.
    \item \textbf{Seeds:} Fixed seeds 42, 123, 456 for three independent runs;
          all tables report mean $\pm$ std ($\leq 0.4\%$ in all cases).
    \item \textbf{Hardware:} Simulated on a single 16-core x86 machine;
          no GPU required for simulation.
\end{itemize}

% ============================================================================
% REFERENCES
% ============================================================================
\bibliographystyle{plainnat}
\begin{thebibliography}{40}

\bibitem[Baker et~al.(2020)]{baker2020emergent}
Baker, B., Kanitscheider, I., Marber, T., Wu, Y., Powell, G., McGrew, B.,
  and Mordatch, I.
\newblock Emergent tool use from multi-agent autocurricula.
\newblock \emph{ICLR}, 2020.

\bibitem[Bernstein et~al.(2018)]{bernstein2018signsgd}
Bernstein, J., Wang, Y.-X., Azizzadenesheli, K., and Anandkumar, A.
\newblock signSGD: Compressed optimisation for non-convex problems.
\newblock \emph{ICML}, 2018.

\bibitem[Bittensor(2023)]{bittensor2023}
Bittensor.
\newblock Bittensor: A peer-to-peer intelligence market.
\newblock Whitepaper, 2023.

\bibitem[Blanchard et~al.(2017)]{blanchard2017machine}
Blanchard, P., El~Mhamdi, E.~M., Guerraoui, R., and Stainer, J.
\newblock Machine learning with adversaries: Byzantine tolerant gradient
  descent.
\newblock \emph{NeurIPS}, 2017.

\bibitem[Bonawitz et~al.(2017)]{bonawitz2017practical}
Bonawitz, K., Ivanov, V., Kreuter, B., Marcedone, A., McMahan, H.~B.,
  Patel, S., Ramage, D., Segal, A., and Seth, K.
\newblock Practical secure aggregation for privacy-preserving machine learning.
\newblock \emph{ACM CCS}, 2017.

\bibitem[Chen et~al.(2018)]{chen2018draco}
Chen, L., Wang, H., Charles, Z., and Papailiopoulos, D.
\newblock DRACO: Byzantine-resilient distributed training via redundant
  gradients.
\newblock \emph{ICML}, 2018.

\bibitem[Christianos et~al.(2021)]{christianos2021scaling}
Christianos, F., Papoudakis, G., Rahman, M.~A., and Albrecht, S.~V.
\newblock Scaling multi-agent reinforcement learning with selective parameter
  sharing.
\newblock \emph{ICML}, 2021.

\bibitem[Conneau et~al.(2020)]{conneau2020unsupervised}
Conneau, A., Rinott, R., Lample, G., Williams, A., Bowman, S., Schwenk, H.,
  and Stoyanov, V.
\newblock XNLI: Evaluating cross-lingual sentence representations.
\newblock \emph{EMNLP}, 2020.

\bibitem[Demers et~al.(1987)]{demers1987epidemic}
Demers, A., Greene, D., Hauser, C., Irish, W., Larson, J., Shenker, S.,
  Sturgis, H., Swinehart, D., and Terry, D.
\newblock Epidemic algorithms for replicated database maintenance.
\newblock \emph{PODC}, 1987.

\bibitem[Dwork \& Roth(2014)]{dwork2014algorithmic}
Dwork, C. and Roth, A.
\newblock The algorithmic foundations of differential privacy.
\newblock \emph{Foundations and Trends in Theoretical Computer Science},
  9(3--4):211--407, 2014.

\bibitem[Furlanello et~al.(2018)]{furlanello2018born}
Furlanello, T., Lipton, Z.~C., Tschannen, M., Itti, L., and Anandkumar, A.
\newblock Born again neural networks.
\newblock \emph{ICML}, 2018.

\bibitem[Gensyn(2023)]{gensyn2023}
Gensyn.
\newblock Gensyn: Protocol for decentralized ML computation.
\newblock Whitepaper, 2023.

\bibitem[Hanzo AI Research(2026a)]{hanzo2026aso}
Hanzo AI Research.
\newblock Active Semantic Optimization: Training-free adaptation via Bayesian
  Product-of-Experts decoding.
\newblock Technical report, Hanzo AI, 2026.

\bibitem[Hanzo AI Research(2026b)]{hanzo2026dso}
Hanzo AI Research.
\newblock Decentralized Semantic Optimization: Byzantine-robust prior
  aggregation for collective model adaptation.
\newblock Technical report, Hanzo AI (ZIP-001/400), 2026.

\bibitem[Hanzo AI Research(2026c)]{hanzo2026bitdelta}
Hanzo AI Research.
\newblock BitDelta: 1-bit quantization of experience embedding deltas for
  efficient fleet communication.
\newblock Technical report, Hanzo AI (ZIP-007), 2026.

\bibitem[Hanzo AI Research(2026d)]{hanzo2026poai}
Hanzo AI Research.
\newblock Proof of AI: Verifiable attestation for agent trajectory extraction.
\newblock Technical report, Hanzo AI (ZIP-002), 2026.

\bibitem[Hinton(2002)]{hinton2002training}
Hinton, G.~E.
\newblock Training products of experts by minimizing contrastive divergence.
\newblock \emph{Neural Computation}, 14(8):1771--1800, 2002.

\bibitem[Hinton et~al.(2015)]{hinton2015distilling}
Hinton, G., Vinyals, O., and Dean, J.
\newblock Distilling the knowledge in a neural network.
\newblock \emph{NeurIPS Workshop}, 2015.

\bibitem[Jimenez et~al.(2024)]{jimenez2024swebench}
Jim{\'e}nez, C.~E., Yang, J., Wettig, A., Yao, S., Pei, K., Press, O.,
  and Narasimhan, K.
\newblock SWE-bench: Can language models resolve real-world {GitHub} issues?
\newblock \emph{ICLR}, 2024.

\bibitem[Koloskova et~al.(2020)]{koloskova2020unified}
Koloskova, A., Stich, S.~U., and Jaggi, M.
\newblock Decentralized stochastic optimization and gossip algorithms with
  compressed communication.
\newblock \emph{ICML}, 2020.

\bibitem[Lamport et~al.(1982)]{lamport1982byzantine}
Lamport, L., Shostak, R., and Pease, M.
\newblock The Byzantine generals problem.
\newblock \emph{ACM TOPLAS}, 4(3):382--401, 1982.

\bibitem[LanceDB(2024)]{lancedb2024}
LanceDB.
\newblock LanceDB: Serverless vector database for AI applications.
\newblock \url{https://lancedb.com}, 2024.

\bibitem[Li et~al.(2020)]{li2020federated}
Li, T., Sahu, A.~K., Zaheer, M., Sanjabi, M., Talwalkar, A., and
  Smith, V.
\newblock Federated optimization in heterogeneous networks.
\newblock \emph{MLSys}, 2020.

\bibitem[Liu et~al.(2024)]{liu2024bitdelta}
Liu, J., Kulikov, R., Fox, Z., Nikdan, A., Kim, Y., Papailiopoulos, D.,
  and De~Sa, C.
\newblock BitDelta: Your fine-tune may only be worth one bit.
\newblock \emph{arXiv preprint arXiv:2402.10193}, 2024.

\bibitem[McMahan et~al.(2017)]{mcmahan2017communication}
McMahan, H.~B., Moore, E., Ramage, D., Hampson, S., and
  Ag{\"u}era~y~Arcas, B.
\newblock Communication-efficient learning of deep networks from decentralized
  data.
\newblock \emph{AISTATS}, 2017.

\bibitem[McMahan et~al.(2018)]{mcmahan2018learning}
McMahan, H.~B., Ramage, D., Talwar, K., and Zhang, L.
\newblock Learning differentially private recurrent language models.
\newblock \emph{ICLR}, 2018.

\bibitem[El~Mhamdi et~al.(2018)]{mhamdi2018hidden}
El~Mhamdi, E.~M., Guerraoui, R., and Rouault, S.
\newblock The hidden vulnerability of distributed learning in Byzantium.
\newblock \emph{ICML}, 2018.

\bibitem[Minsker(2015)]{minsker2015geometric}
Minsker, S.
\newblock Geometric median and robust estimation in Banach spaces.
\newblock \emph{Bernoulli}, 21(4):2308--2335, 2015.

\bibitem[Pease et~al.(1980)]{pease1980reaching}
Pease, M., Shostak, R., and Lamport, L.
\newblock Reaching agreement in the presence of faults.
\newblock \emph{Journal of the ACM}, 27(2):228--234, 1980.

\bibitem[Seide et~al.(2014)]{seide20141bit}
Seide, F., Fu, H., Droppo, J., Li, G., and Yu, D.
\newblock 1-bit stochastic gradient descent and its application to
  data-parallel distributed training of speech DNNs.
\newblock \emph{Interspeech}, 2014.

\bibitem[Shao et~al.(2024)]{shao2024deepseekmath}
Shao, Z., Wang, P., Zhu, Q., Xu, R., Song, J., Bi, X., Zhang, H., Zhang, M.,
  Li, Y.~K., Wu, Y., and Guo, D.
\newblock DeepSeekMath: Pushing the limits of mathematical reasoning in open
  language models.
\newblock \emph{arXiv preprint arXiv:2402.03300}, 2024.

\bibitem[Stich et~al.(2018)]{stich2018sparsified}
Stich, S.~U., Cordonnier, J.-B., and Jaggi, M.
\newblock Sparsified SGD with memory.
\newblock \emph{NeurIPS}, 2018.

\bibitem[Warnat-Herresthal et~al.(2021)]{warnat2021swarm}
Warnat-Herresthal, S., Schultze, H., Shastry, K.~L., et~al.
\newblock Swarm learning for decentralized and confidential clinical machine
  learning.
\newblock \emph{Nature}, 594:265--270, 2021.

\bibitem[Yin et~al.(2018)]{yin2018byzantine}
Yin, D., Chen, Y., Kannan, R., and Bartlett, P.
\newblock Byzantine-robust distributed learning: Towards optimal statistical
  rates.
\newblock \emph{ICML}, 2018.

\bibitem[Zhang et~al.(2024)]{zhang2024fedlora}
Zhang, J., Hua, Y., Wang, H., Song, T., Xue, Z., Ma, R., and Guan, H.
\newblock FedLoRA: When personalized federated learning meets low-rank
  adaptation.
\newblock \emph{arXiv preprint arXiv:2307.06768}, 2024.

\end{thebibliography}