Trust in Decentralized Systems 2026
2026-03-16..2026-03-20
-- online --
who it's for
privacy researchers, federated learning engineers, machine learning engineers, security engineers, cryptography engineers, platform engineers, product managers
topics
federated diffusion model training; local differential privacy (LDP) guarantees; entropic optimal transport for GANs; BFV (RNS) homomorphic encryption deployments; secure aggregation with VDAFs; zero-knowledge PCD in aggregation; secret-sharing for encrypted backups; differentially private SGD clipping (GeoClip); privacy amplification by sampling; side-channel mitigations for SEV-SNP
| title | author | length |
|---|---|---|
| Personalized Federated Training of Diffusion Models with Privacy Guarantees | Kumar Kshitij Patel | 44m |
Federated pipeline trains local diffusion models up to a chosen t0 to produce clipped, noised uploads and trains a global model on those noisy aggregates… | ||
| Talk by Mahdi Haghifam (Toyota Technological Institute at Chicago) | Mahdi Haghifam | 36m |
Shows that differentially private gradient descent incurs a linear R penalty for geometric median but a new private algorithm localizes to the data so error… | ||
| Training generative models from locally privatized data via entropic optimal transport | Ayfer Ozgur | 40m |
Claim: entropic optimal transport lets a GAN trained only on locally privatized samples recover the raw data distribution rather than the noise-convolved one… | ||
| Talk by Suhas Diggavi (UCLA) | Suhas Diggavi | 30m |
Frame personalization as hierarchical empirical Bayes where clients sample parameters from an unknown population and plug‑in estimators yield James–Stein–style… | ||
| Hardening Confidential Federated Computations against Side-Channel Attacks | Albert Cheu | 25m |
Identifies two concrete side channels against confidential federated analytics running on AMD SEV‑SNP VMs: host-observable page faults from dynamic hash tables… | ||
| AgentCrypt: Advancing Privacy and (Secure) Computation in AI Agent Collaboration | Antigoni Polychroniadou | 33m |
Claims a practical framework called AgentCrypt that enforces 100% privacy guarantee for tagged data by replacing OBO tokens with ID-only auth plus… | ||
| Apple's Deployment of Homomorphic Encryption at Scale | Rehan Rishi | 30m |
Deploys homomorphic encryption plus differential privacy and Oblivious HTTP to power real features: private Live Caller ID (iOS 18), photo landmark tagging and… | ||
| Splitting Secrets for Encrypted Backups | Emma Dauterman, Graeme Connell, Vivian Fang, Allison Li, Raluca Ada Popa, Deevashwer Rathee, Rolfe Schmidt | 25m |
Proposes two deployable designs that split backup keys so one compromised component cannot leak many users: a three-cloud enclave design that uses OPRF-based… | ||
| Distributed Aggregation Protocol: A Standard for Privacy-Preserving Aggregation in MPC | Tim Geoghegan | 30m |
Distributed Aggregation Protocol standardizes running Prio-based VDAFs over the network: two non-colluding aggregators accept secret-shared client inputs… | ||
| Secure Aggregation with Lightweight Committees | Phillipp Schoppmann | 31m |
Secure aggregation that keeps clients one‑shot and avoids shipping huge ciphertexts uses a short RLWE-based key encrypted under an additively homomorphic… | ||
| Private aggregation at scale with decentralized trust | Yiping Ma | 32m |
Designs enable private aggregation with a single untrusted server by shifting trust to the client population: Flamingo uses two-round semi-honest / three-round… | ||
| Towards making private telemetry as ubiquitous as TLS | Ryan Lehmkuhl | 31m |
Proposes Heli, a heavy‑light private aggregation design that makes the nonprofit “light” helper’s cost essentially constant by using a key‑homomorphic PRF and… | ||
| WIP title: Building Practical Privacy-Preserving Inference Systems | Wenting Zheng | 57m |
Bolt implements privacy-preserving two‑party inference for BERT using MPC plus HE and cuts the extreme communication of prior work (Iron used 280 GB and 3.5… | ||
| Efficient Robust Private Aggregation for DP | Kunal Talwar | 34m |
Two-server Prio model for secure aggregation, with concrete constructions: PINE: sqrt(D) additional communication, about 3% overhead at D=1,000,000, gives… | ||
| Dimension-free Private Mean Estimation for Anisotropic Distributions | Lydia Zakynthinou | 21m |
Provides an (ε,δ)-DP estimator that achieves Euclidean error ≤α with sample complexity n ≳ tr(Σ)/α^2 + tr(Σ^{1/2})/(αε), and proves that bound is optimal up to… | ||
| From Theory to Practice: Advances in Real-World Federated Learning | Holger Roth | 31m |
Open-source NVIDIA FLARE provides a production-ready federated learning SDK with new APIs (collaboration API, model streaming, native tensor transfer) and… | ||
| Distributed Models for Private Analysis of Graph Data | Adam M. Smith | 35m |
Defines a local node-private model where each node randomizes its adjacency list noninteractively and an untrusted aggregator sums reports. Algorithms clip… | ||
| Talk by Audra McMillan (Apple Inc) | Audra McMillan | 29m |
Defines local pan-privacy as protection against a continual on‑device intruder who sees every internal state and the server output, proves an… | ||
| Stargazing into AI Usage with Differential Privacy | Pritish Kamath | 33m |
Urania implements end-to-end differential privacy for AI usage logs by rewriting chats with an LLM, embedding the summaries, running DP k‑means clustering, and… | ||
| GeoClip: Geometry-Aware Clipping for Differentially Private SGD | Lalitha Sankar | 26m |
GeoClip transforms per-sample gradients into a learned basis (running noisy mean and covariance, eigendecomposition) and clips plus adds isotropic Gaussian… | ||
| Virtual talk by Jonathan Ullman (Northeastern) | Jonathan Ullman | 39m |
Shows that for a noisy Gaussian mean setting membership inference can be information-theoretically possible yet demand an auxiliary sample size Omega(d) that… | ||
| Synthetic Data as an Enabler for Learning from Decentralized Private Data | Guilia Fanti | 48m |
Centralized training on differentially private synthetic data using the Private Evolution pipeline (nearest‑neighbor histograms with DP noise, foundation-model… | ||
| Private Insights into AI Use | Peter Kairouz | 32m |
Describes a practical break of Anthropic's CLIO via the CLIOPATRA attack that seeds dozens of poisoned Claude chats to bypass PII scrubbing, clustering and LLM… | ||
| Virtual talk by Janardhan Kulkarni | Janardhan Kulkarni | 32m |
Claim: privacy becomes a scaling problem as agents act across many workflows. Approach: train LLMs with reinforcement learning with chain-of-thought using a… | ||
| Talk by Amer Sinha (Google) | Amer Sinha | 34m |
A one billion parameter Gemma variant is trained from scratch with DP-SGD, yielding epsilon < 2 sequence-level DP on 1,024-token units packed from web sources… | ||
| Is the New 2016, but Make It Privacy: On Federated Memory, Contextual Privacy, and... | Niloofar Mireshghallah | 37m |
Persistent agents like those in OpenClaw deployment collect session memories and cause contextual leaks; a new benchmark CIMemories finds up to 69%… | ||
| When Unstoppable Frontier Agents Meet Immovable Attack Vectors | Eugene Bagdasarian | 56m |
Argues that deployable fixes must pair language models that decide appropriate data and actions with system primitives that enforce contextual policies… | ||
| Pairwise Network Differential Privacy | Edwige Cyffers | 35m |
Shows that a standard gossip algorithm that alternates a gradient step with a gossip step enables a gradient-reconstruction attack that recovers exact images… | ||
| "Having Confidence in My Confidence Intervals": How Data Users Engage with Privacy-Protected... | Jayshree Sarathy, Harold Triedman, Priyanka Nanayakkara, Rachel Cummings, Gabriel Kaptchuk, Sean Kross, Elissa M. Redmiles | 27m |
Compares Wikipedia pageview releases produced with rounding vs differential privacy, builds dataset documentation refined with five privacy-communication… | ||
| Privately Estimating Black-Box Statistics | Thomas Steinke | 30m |
Provides a concrete scheme that privately estimates arbitrary black‑box functions by evaluating F on overlapping subsets and aggregating with the shifted… | ||
| Chasing the Constants and Its Implication in Private Learning | Jalaj Upadhyay | 31m |
Claims a constructive upper bound and a matching lower bound for the factorization norm of the lower‑triangular all‑ones matrix that drives private continual… | ||
| Heterogeneity and Privacy in Modern Learning | Maryam Aliakbarpour | 33m |
Survey of concrete models and provable separations. Proves DP meta‑learning ≡ DP multitask but shows joint differential privacy is strictly stronger than the… | ||
| Privacy amplification by random allocation | Vitaly Feldman | 35m |
Analyzes k‑out‑t random allocation sampling and provides an algorithm that computes the privacy loss distribution (PLD) for that scheme efficiently, with… | ||
| Privacy Amplification by Sampling in Practice | Arun Ganesh | 43m |
Demonstrates how truncated Poisson sampling lets JAX-style JIT pipelines keep fixed batch sizes by padding or truncating Poisson-drawn batches, trading rare… | ||
| nothing matches | ||