Federated and Collaborative Learning Boot Camp 2026
2026-01-26..2026-01-30
Berkeley, United States
as the organisers put it
The boot camp is intended to acquaint program participants with the key themes of the program. It will consist of five days of tutorial presentations. If you require special accommodation, please contact our access coordinator at simonsevents@berkeley.edu with as much advance notice as possible.
who it's for
federated learning researchers, machine learning engineers, privacy engineers, security engineers, platform engineers, research scientists, data scientists
topics
federated optimization algorithms; federated optimization for LLMs; secure aggregation SMPC; AMD SEV-SNP TEEs; differential privacy mechanisms; DP-SGD and amplification; binary tree mechanism for privacy; communication compression techniques; natural quantization and Rand-1; client heterogeneity algorithms; incentive mechanisms for data sharing; production frameworks NVFlare and Flower
| title | author | length |
|---|---|---|
| Toward Provably Private Federated Learning | Dan Ramage | 01h 01m |
Production systems claim formal user/device-level privacy by shipping DP-FTRL models for Gboard and running a large-scale secure aggregation SMPC stack… | ||
| Opening Remarks | Dan Ramage, Kamalika Chaudhuri, Om Thakkar, Micah Sheller, Adria Gascon, Chhavi Yadav | 21m |
Program states a semester-long Simons initiative with a bootcamp this week plus workshops in February March and April on heterogeneous sources, trust and… | ||
| Short Talks | Ziyue Xu, Sean Augenstein, Sewon Min, Kunal Talwar, Patrick Foley | 01h 08m |
Open-source NVFlare is production-ready since 2021 and runs the 20-hospital COVID oxygen predictor, supports on‑prem IP protection, and surfaces in 50–60+… | ||
| Tutorial: Federated Optimization, Part I | Peter Richtarik | 01h 12m |
Overview of concrete results: the lecture compares cost models (notably the communication-to-server-only cost model), analyzes DCGD with unbiased compressors… | ||
| Tutorial: Federated Optimization, Part II | Peter Richtarik | 01h 13m |
DIANA removes the interpolation assumption by compressing gradient differences, yields near‑linear convergence without shared optimizers, and can be… | ||
| Tutorial: Federated Optimization, Part III | Peter Richtarik | 01h 05m |
ProxSkip connects compression plus EF21 to acceleration through local training and links to proximal-point lineage (DANE, Axfeldt, double stabilization) while… | ||
| FedOpt for LLMs | Keith Rush | 01h 00m |
Adaptive federated optimization applied to LLM pretraining uses JAX/DrJAX primitives and a non-blocking fragment sync to shard models (e.g. ~100 fragments) and… | ||
| Tutorial: Privacy, Part I | Adam M. Smith, Lydia Zakynthinou | 01h 17m |
Covers rigorous limits and concrete attacks: models repeatedly emit training data (GPT-2 and diffusion memorization, census reconstructions), a classic result… | ||
| Tutorial: Privacy, Part II | Adam M. Smith, Lydia Zakynthinou | 01h 14m |
Defines differential privacy as an output-distribution bound parametrized by epsilon/delta, proves noise calibrated to global sensitivity (e.g., mean of… | ||
| Tutorial: Privacy, Part III | Adam M. Smith, Lydia Zakynthinou | 01h 14m |
DP-SGD suffers from realistic sampling gaps and amplification limits, so the lecture reframes iterative training as continual observation and private release… | ||
| Tutorial: Security | Adria Gascon | 01h 08m |
Covers precise definitions (simulation/ideal world, DP-style leakage, covert/abort variants) and a running Two‑Server Model example: computing a DP histogram… | ||
| Tutorial: Incentives for Collaborative Learning and Data Sharing, Part I | Han Shao, Sai Praneeth Karimireddy | 45m |
Presents a formal model where K agents choose contribution levels (Mi) and utilities equal local model quality minus linear data cost, proves Nash equilibria… | ||
| Tutorial: Incentives for Collaborative Learning and Data Sharing, Part Ib | Han Shao, Sai Praneeth Karimireddy | 01h 00m |
Shows that naive collaboration can be K times inefficient: a double-star network has an optimal effort of 2 but any stable outcome costs Theta(K). Uses… | ||
| Talk by Micah Sheller (Flower/ML Commons) | Micah Sheller | 27m |
Cross-silo federated learning runs real experiments across 71 hospitals and yields concrete outcomes: a simple ReSCUNet tumor segmentation model beat… | ||
| Talk by Adria Gascon (Google) | Adria Gascon | 35m |
Presents SNPeek, a practical evaluation of AMD SEV SNP showing nested page tables leak 4 KB granularity memory-access patterns, ciphertext visibility reveals… | ||
| Talk by Chhavi Yadav (CMU) | Chhavi Yadav | 17m |
Study interviews 21 interviewees from banks, hospitals, platform teams and researchers and finds a clear misalignment: papers rarely address the top… | ||
| Talk by Kamalika Chaudhuri (Meta) | Kamalika Chaudhuri | 36m |
Built a controllable web-agent testbed and AgentDAM benchmark that runs multi-step UI tasks on synthetic user data and memory profiles, measures two… | ||
| Tutorial: Incentives for Collaborative Learning and Data Sharing, Part II | Han Shao, Sai Praneeth Karimireddy | 01h 27m |
Covers practical definitions and limits of data valuation: leave-one-out needs N+1 retrainings and fails on redundancy; Shapley value solves redundancy but… | ||
| Talk by Om Thakkar (OpenAI) | Om Thakkar | 43m |
Audits target ASR memorization by inserting synthetic audio canaries and measuring exposure versus a 20,000-sample holdout using a 600M parameter model… | ||
| nothing matches | ||