papers/zoo-federated-wildlife/sections/99-conclusion.tex

\section{Conclusion}

\label{sec:conclusion}

We have presented ZooFed, a federated learning framework for cross-institutional wildlife monitoring that addresses modality heterogeneity through sensor-agnostic feature adapters, provides conservation-calibrated differential privacy with species-specific budgets, and handles extreme non-IID distributions through clustered multi-task aggregation.

Evaluation on 18 sites with 4.2 million sensor events demonstrates 89.4\% detection accuracy (within 2.1\% of centralized training), 31.2\% improvement over local-only models for rare species, and formal privacy guarantees with $\epsilon \leq 0.5$ for critically endangered species.

ZooFed enables a new model for conservation collaboration: institutions contribute to collective intelligence without surrendering data sovereignty.
The framework is released as open infrastructure on the Zoo Network, with the aspiration that federated learning becomes the default paradigm for multi-site ecological research.

% ============================================================
% References
% ============================================================
\begin{thebibliography}{30}

\bibitem[Abadi et~al.(2016)]{abadi2016deep}
Abadi, M., Chu, A., Goodfellow, I., et~al. (2016).
\newblock Deep learning with differential privacy.
\newblock {\em Proceedings of CCS}, pages 308--318.

\bibitem[Ahumada et~al.(2020)]{ahumada2020wildlife}
Ahumada, J.~A., Fegraus, E., Birch, T., et~al. (2020).
\newblock Wildlife Insights: A platform to maximize the potential of camera trap and other passive sensor wildlife data.
\newblock {\em Environmental Conservation}, 47(1):1--6.

\bibitem[Alistarh et~al.(2017)]{alistarh2017qsgd}
Alistarh, D., Grubic, D., Li, J., Tomioka, R., and Vojnovic, M. (2017).
\newblock QSGD: Communication-efficient SGD via gradient quantization and encoding.
\newblock {\em NeurIPS}, 30.

\bibitem[Beery et~al.(2019)]{beery2019efficient}
Beery, S., Morris, D., and Yang, S. (2019).
\newblock Efficient pipeline for camera trap image review.
\newblock {\em arXiv preprint arXiv:1907.06772}.

\bibitem[Beery et~al.(2021)]{beery2021iwildcam}
Beery, S., Cole, E., Parker, J., et~al. (2021).
\newblock The iWildCam 2021 competition dataset.
\newblock {\em arXiv preprint arXiv:2105.03494}.

\bibitem[Deiner et~al.(2017)]{deiner2017environmental}
Deiner, K., Bik, H.~M., M\"{a}chler, E., et~al. (2017).
\newblock Environmental DNA metabarcoding: Transforming how we survey animal and plant communities.
\newblock {\em Molecular Ecology}, 26(21):5872--5895.

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

\bibitem[Fritz et~al.(2019)]{fritz2019citizen}
Fritz, S., See, L., Carlson, T., et~al. (2019).
\newblock Citizen science and the United Nations Sustainable Development Goals.
\newblock {\em Nature Sustainability}, 2(10):922--930.

\bibitem[Geyer et~al.(2017)]{geyer2017differentially}
Geyer, R.~C., Klein, T., and Nabi, M. (2017).
\newblock Differentially private federated learning: A client level perspective.
\newblock {\em arXiv preprint arXiv:1712.07557}.

\bibitem[Ghosh et~al.(2020)]{ghosh2020efficient}
Ghosh, A., Chung, J., Yin, D., and Ramchandran, K. (2020).
\newblock An efficient framework for clustered federated learning.
\newblock {\em NeurIPS}, 33:19586--19597.

\bibitem[Gibb et~al.(2019)]{gibb2019emerging}
Gibb, R., Browning, E., Glover-Kapfer, P., and Jones, K.~E. (2019).
\newblock Emerging opportunities and challenges for passive acoustics in ecological assessment and monitoring.
\newblock {\em Methods in Ecology and Evolution}, 10(2):169--185.

\bibitem[Kahl et~al.(2021)]{kahl2021birdnet}
Kahl, S., Wood, C.~M., Eibl, M., and Klinck, H. (2021).
\newblock BirdNET: A deep learning solution for avian diversity monitoring.
\newblock {\em Ecological Informatics}, 61:101236.

\bibitem[Kairouz et~al.(2021)]{kairouz2021advances}
Kairouz, P., McMahan, H.~B., Avent, B., et~al. (2021).
\newblock Advances and open problems in federated learning.
\newblock {\em Foundations and Trends in Machine Learning}, 14(1-2):1--210.

\bibitem[Kang et~al.(2020)]{kang2020incentive}
Kang, J., Xiong, Z., Niyato, D., et~al. (2020).
\newblock Incentive mechanism for reliable federated learning: A joint optimization approach.
\newblock {\em IEEE Internet of Things Journal}, 7(12):12221--12236.

\bibitem[Kays et~al.(2015)]{kays2015terrestrial}
Kays, R., Crofoot, M.~C., Jetz, W., and Wikelski, M. (2015).
\newblock Terrestrial animal tracking as an eye on life and planet.
\newblock {\em Science}, 348(6240):aaa2478.

\bibitem[Li et~al.(2020)]{li2020federated}
Li, T., Sahu, A.~K., Zaheer, M., et~al. (2020).
\newblock Federated optimization in heterogeneous networks.
\newblock {\em Proceedings of MLSys}, 2:429--450.

\bibitem[Lin et~al.(2020)]{lin2020ensemble}
Lin, T., Kong, L., Stich, S.~U., and Jaggi, M. (2020).
\newblock Ensemble distillation for robust model fusion in federated learning.
\newblock {\em NeurIPS}, 33:2351--2363.

\bibitem[Lindenmayer et~al.(2017)]{lindenmayer2017do}
Lindenmayer, D. and Scheele, B. (2017).
\newblock Do not publish.
\newblock {\em Science}, 356(6340):800--801.

\bibitem[McMahan et~al.(2017)]{mcmahan2017communication}
McMahan, B., Moore, E., Ramage, D., et~al. (2017).
\newblock Communication-efficient learning of deep networks from decentralized data.
\newblock {\em Proceedings of AISTATS}, pages 1273--1282.

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

\bibitem[Norouzzadeh et~al.(2018)]{norouzzadeh2018automatically}
Norouzzadeh, M.~S., Nguyen, A., Kosmala, M., et~al. (2018).
\newblock Automatically identifying, counting, and describing wild animals in camera-trap images.
\newblock {\em Proceedings of the National Academy of Sciences}, 115(25):E5716--E5725.

\bibitem[Patterson et~al.(2017)]{patterson2017statistical}
Patterson, T.~A., Parton, A., Langrock, R., et~al. (2017).
\newblock Statistical modelling of individual animal movement.
\newblock {\em Journal of the Royal Statistical Society: Series B}, 79(4):1055--1069.

\bibitem[Smith et~al.(2017)]{smith2017federated}
Smith, V., Chiang, C.~K., Sanjabi, M., and Talwalkar, A.~S. (2017).
\newblock Federated multi-task learning.
\newblock {\em NeurIPS}, 30:4424--4434.

\bibitem[Stich et~al.(2018)]{stich2018sparsified}
Stich, S.~U., Cordonnier, J.~B., and Jaggi, M. (2018).
\newblock Sparsified SGD with memory.
\newblock {\em NeurIPS}, 31:4452--4463.

\bibitem[Stowell(2019)]{stowell2019automatic}
Stowell, D. (2019).
\newblock Automatic acoustic identification of individuals in multiple species.
\newblock {\em Journal of the Royal Society Interface}, 16(153):20180940.

\bibitem[Tenopir et~al.(2011)]{tenopir2011data}
Tenopir, C., Allard, S., Douglass, K., et~al. (2011).
\newblock Data sharing by scientists: Practices and perceptions.
\newblock {\em PLOS ONE}, 6(6):e21101.

\bibitem[Tuia et~al.(2022)]{tuia2022perspectives}
Tuia, D., Kellenberger, B., Beery, S., et~al. (2022).
\newblock Perspectives in machine learning for wildlife conservation.
\newblock {\em Nature Communications}, 13:792.

\bibitem[Tulloch et~al.(2018)]{tulloch2018decision}
Tulloch, A.~I.~T., Auerbach, N., and Avery-Gomm, S. (2018).
\newblock A decision tree for assessing the risks and benefits of publishing biodiversity data.
\newblock {\em Nature Ecology \& Evolution}, 2(8):1209--1217.

\bibitem[Verma et~al.(2022)]{verma2022towards}
Verma, V., Lamb, A., and Beckham, C. (2022).
\newblock Towards multi-modal wildlife monitoring.
\newblock {\em Ecological Informatics}, 68:101552.

\bibitem[Antje Worring, Zoo Labs Foundation Research(2025)]{zoo2025dso}
Antje Worring, Zoo Labs Foundation Research (2025).
\newblock Decentralized Semantic Optimization: Collaborative knowledge curation on distributed networks.
\newblock {\em Zoo Improvement Proposal ZIP-001}.

\end{thebibliography}