Analysis of Privacy Leakage in Federated Large Language Models

With the rapid adoption of Federated Learning (FL) as the training and tuning protocol for applications utilizing Large Language Models (LLMs), recent research highlights the need for significant modifications to FL to accommodate the large-scale of LLMs. While substantial adjustments to the protocol have been introduced as a response, comprehensive privacy analysis for the adapted FL protocol is currently lacking. To address this gap, our work delves into an extensive examination of the privacy analysis of FL when used for training LLMs, both from theoretical and practical perspectives. In particular, we design two active membership inference attacks with guaranteed theoretical success rates to assess the privacy leakages of various adapted FL configurations. Our theoretical findings are translated into practical attacks, revealing substantial privacy vulnerabilities in popular LLMs, including BERT, RoBERTa, DistilBERT, and OpenAI's GPTs, across multiple real-world language datasets. Additionally, we conduct thorough experiments to evaluate the privacy leakage of these models when data is protected by state-of-the-art differential privacy (DP) mechanisms.

On the Limit of Explaining Black-box Temporal Graph Neural Networks

Temporal Graph Neural Network (TGNN) has been receiving a lot of attention recently due to its capability in modeling time-evolving graph-related tasks. Similar to Graph Neural Networks, it is also non-trivial to interpret predictions made by a TGNN due to its black-box nature. A major approach tackling this problems in GNNs is by analyzing the model' responses on some perturbations of the model's inputs, called perturbation-based explanation methods. While these methods are convenient and flexible since they do not need internal access to the model, does this lack of internal access prevent them from revealing some important information of the predictions? Motivated by that question, this work studies the limit of some classes of perturbation-based explanation methods. Particularly, by constructing some specific instances of TGNNs, we show (i) node-perturbation cannot reliably identify the paths carrying out the prediction, (ii) edge-perturbation is not reliable in determining all nodes contributing to the prediction and (iii) perturbing both nodes and edges does not reliably help us identify the graph's components carrying out the temporal aggregation in TGNNs.

EMaP: Explainable AI with Manifold-based Perturbations

In the last few years, many explanation methods based on the perturbations of input data have been introduced to improve our understanding of decisions made by black-box models. The goal of this work is to introduce a novel perturbation scheme so that more faithful and robust explanations can be obtained. Our study focuses on the impact of perturbing directions on the data topology. We show that perturbing along the orthogonal directions of the input manifold better preserves the data topology, both in the worst-case analysis of the discrete Gromov-Hausdorff distance and in the average-case analysis via persistent homology. From those results, we introduce EMaP algorithm, realizing the orthogonal perturbation scheme. Our experiments show that EMaP not only improves the explainers' performance but also helps them overcome a recently-developed attack against perturbation-based methods.

NeuCEPT: Locally Discover Neural Networks' Mechanism via Critical Neurons Identification with Precision Guarantee

Despite recent studies on understanding deep neural networks (DNNs), there exists numerous questions on how DNNs generate their predictions. Especially, given similar predictions on different input samples, are the underlying mechanisms generating those predictions the same? In this work, we propose NeuCEPT, a method to locally discover critical neurons that play a major role in the model's predictions and identify model's mechanisms in generating those predictions. We first formulate a critical neurons identification problem as maximizing a sequence of mutual-information objectives and provide a theoretical framework to efficiently solve for critical neurons while keeping the precision under control. NeuCEPT next heuristically learns different model's mechanisms in an unsupervised manner. Our experimental results show that neurons identified by NeuCEPT not only have strong influence on the model's predictions but also hold meaningful information about model's mechanisms.

An Explainer for Temporal Graph Neural Networks

Temporal graph neural networks (TGNNs) have been widely used for modeling time-evolving graph-related tasks due to their ability to capture both graph topology dependency and non-linear temporal dynamic. The explanation of TGNNs is of vital importance for a transparent and trustworthy model. However, the complex topology structure and temporal dependency make explaining TGNN models very challenging. In this paper, we propose a novel explainer framework for TGNN models. Given a time series on a graph to be explained, the framework can identify dominant explanations in the form of a probabilistic graphical model in a time period. Case studies on the transportation domain demonstrate that the proposed approach can discover dynamic dependency structures in a road network for a time period.

Learning Interpretation with Explainable Knowledge Distillation

Knowledge Distillation (KD) has been considered as a key solution in model compression and acceleration in recent years. In KD, a small student model is generally trained from a large teacher model by minimizing the divergence between the probabilistic outputs of the two. However, as demonstrated in our experiments, existing KD methods might not transfer critical explainable knowledge of the teacher to the student, i.e. the explanations of predictions made by the two models are not consistent. In this paper, we propose a novel explainable knowledge distillation model, called XDistillation, through which both the performance the explanations' information are transferred from the teacher model to the student model. The XDistillation model leverages the idea of convolutional autoencoders to approximate the teacher explanations. Our experiments shows that models trained by XDistillation outperform those trained by conventional KD methods not only in term of predictive accuracy but also faithfulness to the teacher models.

PGM-Explainer: Probabilistic Graphical Model Explanations for Graph Neural Networks

In Graph Neural Networks (GNNs), the graph structure is incorporated into the learning of node representations. This complex structure makes explaining GNNs' predictions become much more challenging. In this paper, we propose PGM-Explainer, a Probabilistic Graphical Model (PGM) model-agnostic explainer for GNNs. Given a prediction to be explained, PGM-Explainer identifies crucial graph components and generates an explanation in form of a PGM approximating that prediction. Different from existing explainers for GNNs where the explanations are drawn from a set of linear functions of explained features, PGM-Explainer is able to demonstrate the dependencies of explained features in form of conditional probabilities. Our theoretical analysis shows that the PGM generated by PGM-Explainer includes the Markov-blanket of the target prediction, i.e. including all its statistical information. We also show that the explanation returned by PGM-Explainer contains the same set of independence statements in the perfect map. Our experiments on both synthetic and real-world datasets show that PGM-Explainer achieves better performance than existing explainers in many benchmark tasks.

Evaluating Explainers via Perturbation

Due to high complexity of many modern machine learning models such as deep convolutional networks, understanding the cause of model's prediction is critical. Many explainers have been designed to give us more insights on the decision of complex classifiers. However, there is no common ground on evaluating the quality of different classification methods. Motivated by the needs for comprehensive evaluation, we introduce the c-Eval metric and the corresponding framework to quantify the explainer's quality on feature-based explainers of machine learning image classifiers. Given a prediction and the corresponding explanation on that prediction, c-Eval is the minimum-power perturbation that successfully alters the prediction while keeping the explanation's features unchanged. We also provide theoretical analysis linking the proposed parameter with the portion of predicted object covered by the explanation. Using a heuristic approach, we introduce the c-Eval plot, which not only displays a strong connection between c-Eval and explainers' quality, but also serves as a low-complexity approach of assessing explainers. We finally conduct extensive experiments of explainers on three different datasets in order to support the adoption of c-Eval in evaluating explainers' performance.