arXiv subjects

Ivoline C. Ngong

Publications and source records attributed to Ivoline C. Ngong.

10 recordsLinked to original sources

arXiv2026

Differentially Private Multimodal In-Context Learning

Vision-language models are increasingly applied to sensitive domains such as medical imaging and personal photographs, yet existing differentially private methods for in-context learning are limited to few-shot, text-only settings because privacy cost scales with the number of tokens processed. We present Differentially Private Multimodal Task Vectors (DP-MTV), the first framework enabling many-shot multimodal in-context learning with formal $(\varepsilon, δ)$-differential privacy by aggregating hundreds of demonstrations into compact task vectors in activation space. DP-MTV partitions private data into disjoint chunks, applies per-layer clipping to bound sensitivity, and adds calibrated noise to the aggregate, requiring only a single noise addition that enables unlimited inference queries. We evaluate on eight benchmarks across three VLM architectures, supporting deployment with or without auxiliary data. At $\varepsilon=1.0$, DP-MTV achieves 50% on VizWiz compared to 55% non-private and 35% zero-shot, preserving most of the gain from in-context learning under meaningful privacy constraints.

cs.AI↗

arXiv2026

AgentSCOPE: Evaluating Contextual Privacy Across Agentic Workflows

Agentic systems are increasingly acting on users' behalf, accessing calendars, email, and personal files to complete everyday tasks. Privacy evaluation for these systems has focused on the input and output boundaries, but each task involves several intermediate information flows, from agent queries to tool responses, that are not currently evaluated. We argue that every boundary in an agentic pipeline is a site of potential privacy violation and must be assessed independently. To support this, we introduce the Privacy Flow Graph, a Contextual Integrity-grounded framework that decomposes agentic execution into a sequence of information flows, each annotated with the five CI parameters, and traces violations to their point of origin. We present AgentSCOPE, a benchmark of 62 multi-tool scenarios across eight regulatory domains with ground truth at every pipeline stage. Our evaluation across seven state-of-the-art LLMs show that privacy violations in the pipeline occur in over 80% of scenarios, even when final outputs appear clean (24%), with most violations arising at the tool-response stage where APIs return sensitive data indiscriminately. These results indicate that output-level evaluation alone substantially underestimates the privacy risk of agentic systems.

cs.CR↗

arXiv2025

Distributed HDMM: Scalable, Distributed, Accurate, and Differentially Private Query Workloads without a Trusted Curator

We present the Distributed High-Dimensional Matrix Mechanism (Distributed HDMM), a protocol for answering workloads of linear queries on distributed data that provides the accuracy of central-model HDMM without a trusted curator. Distributed HDMM leverages a secure aggregation protocol to evaluate HDMM on distributed data, and is secure in the context of a malicious aggregator and malicious clients (assuming an honest majority). Our preliminary empirical evaluation shows that Distributed HDMM can run on realistic datasets and workloads with thousands of clients in less than one minute.

cs.CR↗

arXiv2025

SoK: Usability Studies in Differential Privacy

Differential Privacy (DP) has emerged as a pivotal approach for safeguarding individual privacy in data analysis, yet its practical adoption is often hindered by challenges in the implementation and communication of DP. This paper presents a comprehensive systematization of existing research studies around the usability of DP, synthesizing insights from studies on both the practical use of DP tools and strategies for conveying DP parameters that determine privacy protection levels, such as epsilon($\varepsilon$). By reviewing and analyzing these studies, we identify core usability challenges, best practices, and critical gaps in current DP tools that affect adoption across diverse user groups, including developers, data analysts, and non-technical stakeholders. Our analysis highlights actionable insights and pathways for future research that emphasizes user-centered design and clear communication, fostering the development of more accessible DP tools that meet practical needs and support broader adoption.

cs.HC↗

arXiv2024

Differentially Private Learning Needs Better Model Initialization and Self-Distillation

Differentially private SGD (DPSGD) enables privacy-preserving training of language models, but often reduces utility, diversity, and linguistic quality. We introduce DPRefine, a three-phase method that initializes a model using data synthesis from a small pre-trained LM with rigorous filtering, applies DP finetuning on private data, and performs self-distillation to refine outputs. This approach significantly outperforms vanilla DPSGD, with AlpacaEval preferring DPRefine's generations in 78.4% of cases across all datasets. Our analysis reveals that DPRefine reduces linguistic errors in generated text by 84.0%, mitigating grammar and spelling errors, commonly associated with DPSGD. It also reduces inconsistencies of non-private models, such as hallucinated details and misattributed quotes. We find that small models like GPT-2 can be effective for initialization and distillation, highlighting their potential in enabling scalable and efficient deployment of privacy-preserving language.

cs.LG↗

arXiv2024

Evaluating the Usability of Differential Privacy Tools with Data Practitioners

Differential privacy (DP) has become the gold standard in privacy-preserving data analytics, but implementing it in real-world datasets and systems remains challenging. Recently developed DP tools aim to make DP implementation easier, but limited research has investigated these DP tools' usability. Through a usability study with 24 US data practitioners with varying prior DP knowledge, we evaluated the usability of four Python-based open-source DP tools: DiffPrivLib, Tumult Analytics, PipelineDP, and OpenDP. Our results suggest that using DP tools in this study may help DP novices better understand DP; that Application Programming Interface (API) design and documentation are vital for successful DP implementation; and that user satisfaction correlates with how well participants completed study tasks with these DP tools. We provide evidence-based recommendations to improve DP tools' usability to broaden DP adoption.

cs.HC↗

arXiv2024

OLYMPIA: A Simulation Framework for Evaluating the Concrete Scalability of Secure Aggregation Protocols

Recent secure aggregation protocols enable privacy-preserving federated learning for high-dimensional models among thousands or even millions of participants. Due to the scale of these use cases, however, end-to-end empirical evaluation of these protocols is impossible. We present OLYMPIA, a framework for empirical evaluation of secure protocols via simulation. OLYMPIA. provides an embedded domain-specific language for defining protocols, and a simulation framework for evaluating their performance. We implement several recent secure aggregation protocols using OLYMPIA, and perform the first empirical comparison of their end-to-end running times. We release OLYMPIA as open source.

cs.CR↗

arXiv2022

Backpropagation Clipping for Deep Learning with Differential Privacy

We present backpropagation clipping, a novel variant of differentially private stochastic gradient descent (DP-SGD) for privacy-preserving deep learning. Our approach clips each trainable layer's inputs (during the forward pass) and its upstream gradients (during the backward pass) to ensure bounded global sensitivity for the layer's gradient; this combination replaces the gradient clipping step in existing DP-SGD variants. Our approach is simple to implement in existing deep learning frameworks. The results of our empirical evaluation demonstrate that backpropagation clipping provides higher accuracy at lower values for the privacy parameter $ε$ compared to previous work. We achieve 98.7% accuracy for MNIST with $ε= 0.07$ and 74% accuracy for CIFAR-10 with $ε= 3.64$.

cs.LG↗

arXiv2022

Prediction Sensitivity: Continual Audit of Counterfactual Fairness in Deployed Classifiers

As AI-based systems increasingly impact many areas of our lives, auditing these systems for fairness is an increasingly high-stakes problem. Traditional group fairness metrics can miss discrimination against individuals and are difficult to apply after deployment. Counterfactual fairness describes an individualized notion of fairness but is even more challenging to evaluate after deployment. We present prediction sensitivity, an approach for continual audit of counterfactual fairness in deployed classifiers. Prediction sensitivity helps answer the question: would this prediction have been different, if this individual had belonged to a different demographic group -- for every prediction made by the deployed model. Prediction sensitivity can leverage correlations between protected status and other features and does not require protected status information at prediction time. Our empirical results demonstrate that prediction sensitivity is effective for detecting violations of counterfactual fairness.

cs.LG↗

arXiv2020

Towards Auditability for Fairness in Deep Learning

Group fairness metrics can detect when a deep learning model behaves differently for advantaged and disadvantaged groups, but even models that score well on these metrics can make blatantly unfair predictions. We present smooth prediction sensitivity, an efficiently computed measure of individual fairness for deep learning models that is inspired by ideas from interpretability in deep learning. smooth prediction sensitivity allows individual predictions to be audited for fairness. We present preliminary experimental results suggesting that smooth prediction sensitivity can help distinguish between fair and unfair predictions, and that it may be helpful in detecting blatantly unfair predictions from "group-fair" models.

cs.LG↗