differential privacy
PulseAugur coverage of differential privacy — every cluster mentioning differential privacy across labs, papers, and developer communities, ranked by signal.
5 天有情绪数据
-
New theory offers differential privacy guarantees for sampling
Researchers have developed a perturbation theory for spherical Hellinger-Kantorovich (SHK) gradient flows, allowing for precise comparison of flows based on differing potentials. This theory provides dimension-free boun…
-
New ICA method offers privacy-preserving ML without performance loss
Researchers have introduced Informationally Compressive Anonymization (ICA) and the VEIL architecture as a novel approach to privacy-preserving machine learning. This method uses an encoder within a trusted environment …
-
New audit protocol tackles privacy risks in multi-tenant RAG systems
Researchers have identified a privacy vulnerability in multi-tenant Retrieval-Augmented Generation (RAG) systems, specifically concerning account collusion. While these services typically guarantee differential privacy …
-
Research Paper Analyzes Privacy's Impact on CVaR Learning
A new research paper explores how differential privacy impacts the learning of Conditional Value at Risk (CVaR). The study reveals that privacy mechanisms alter the effective sample size for CVaR calculations, with the …
-
Personalized differential privacy budgets show limited gains
Researchers have identified significant limitations in personalized differential privacy budgets, particularly for mean estimation tasks. Their findings indicate that the primary factor for utility is not full personali…
-
New framework models AI audits with strategic developer responses
Researchers have developed a new framework for designing regulatory audits of AI systems that accounts for strategic responses from developers. The proposed method models the interaction as a bilevel Stackelberg game, w…
-
New Quadratic Objective Perturbation method enhances differential privacy for ML
Researchers have introduced Quadratic Objective Perturbation (QOP) as a novel method for differential privacy in machine learning. Unlike Linear Objective Perturbation (LOP), which requires bounded gradients, QOP uses a…
-
Researchers explore differential privacy for bandit problems and multi-agent learning
Two new research papers explore the application of differential privacy in bandit problems. The first paper introduces an algorithm for extensive-form bandit problems that achieves local differential privacy with a regr…
-
Researchers explore privacy-utility trade-offs in Graph Convolutional Networks
Researchers have developed a theoretical framework to understand differential privacy in Graph Convolutional Networks (GCNs) by examining subsampling stability. The study derives upper bounds on misclassification rates,…
-
Researchers explore text decomposition and budget distribution for private text obfuscation
Researchers have explored methods for differentially private text obfuscation, focusing on how to distribute privacy budgets across text segments. The study systematically evaluated different text decomposition techniqu…
-
New research explores differential privacy's impact on text style and recommendation accuracy
Two new research papers explore advancements in differential privacy. One paper demonstrates that differentially-private text rewriting, while preserving semantic content, significantly alters the stylistic and communic…
-
New research advances federated learning for privacy and heterogeneity
Researchers are developing new methods to improve federated learning, a technique that allows models to train on decentralized data without compromising privacy. Several papers introduce novel algorithms for handling da…
-
ESPADA framework speeds up robot imitation learning by 2x
Researchers have developed ESPADA, a new framework designed to accelerate robot manipulation tasks by intelligently downsampling demonstration data. ESPADA utilizes a VLM-LLM pipeline to identify and preserve critical p…
-
Federated Learning advances balance privacy, utility, and fairness
Researchers are exploring advanced techniques to enhance privacy in Federated Learning (FL), a method where models train on decentralized data. One study compares Differential Privacy (DP) and Homomorphic Encryption (HE…
-
LLMs improve privacy-utility trade-off for Dutch clinical note de-identification
Researchers have conducted a comparative study on methods for de-identifying Dutch clinical notes to protect patient privacy while allowing for data reuse. The study evaluated traditional methods like differential priva…
-
Apple details privacy-preserving AI research and differential privacy for Apple Intelligence
Apple is advancing research in privacy-preserving machine learning and AI, hosting a workshop to discuss techniques like federated learning and differential privacy. The company is applying these methods to its upcoming…