Search for: All records

The ability to make targeted updates to models, whether for unlearning, debiasing, model editing, or safety alignment, is central to AI safety. While these interventions aim to modify specific knowledge (e.g., removing virology content), their effects often propagate to related but unintended areas (e.g., allergies). Due to lack of standardized tools, existing evaluations typically compare performance on targeted versus unrelated general tasks, overlooking this broader collateral impact called the "ripple effect". We introduce RippleBench, a benchmark for systematically measuring how interventions affect semantically related knowledge. Using RippleBench, built on top of a Wikipedia-RAG pipeline for generating multiple-choice questions, we evaluate eight state-of-the-art unlearning methods. We find that all methods exhibit non-trivial accuracy drops on topics increasingly distant from the unlearned knowledge, each with distinct propagation profiles. We release our codebase for on-the-fly ripple evaluation as well as RippleBench-WMDP-Bio, a dataset derived from WMDP biology, containing 9,888 unique topics and 49,247 questions. 

Differential Privacy (DP) is a mathematical definition that enshrines a formal guarantee that the output of a query does not depend greatly on any individual in the dataset. DP does not formalize a notion of "background information" and does not provide a guarantee about how much an output can be identifying to someone who has background information about an individual. In this paper, we argue that privately fine-tuning a pre-trained machine learning model on a private dataset using differential privacy does not always yield meaningful notions of privacy. Simply offering differential privacy guarantees in terms of (ε, δ) is insufficient to ensure human notions privacy, when the original training data is correlated with the fine-tuning dataset. We emphasize that, alongside differential privacy assurances, it is essential to report measures of dataset similarity and model attackability (for which model-size can be a proxy). This is a work in progress; this work is primarily a position piece, arguing for how DP should be used in practice, and what future research needs to be conducted in order to better answer those questions. 

Evaluating machine unlearning methods remains technically challenging, with recent benchmarks requiring complex setups and significant engineering overhead. We introduce a unified and extensible benchmarking suite that simplifies the evaluation of unlearning algorithms using the KLoM (KL divergence of Margins) metric. Our framework provides precomputed model ensembles, oracle outputs, and streamlined infrastructure for running evaluations out of the box. By standardizing setup and metrics, it enables reproducible, scalable, and fair comparison across unlearning methods. We aim for this benchmark to serve as a practical foundation for accelerating research and promoting best practices in machine unlearning. Our code and data are publicly available. 

Machine unlearning---efficiently removing the effect of a small "forget set" of training data on a pre-trained machine learning model---has recently attracted significant research interest. Despite this interest, however, recent work shows that existing machine unlearning techniques do not hold up to thorough evaluation in non-convex settings. In this work, we introduce a new machine unlearning technique that exhibits strong empirical performance even in such challenging settings. Our starting point is the perspective that the goal of unlearning is to produce a model whose outputs are statistically indistinguishable from those of a model re-trained on all but the forget set. This perspective naturally suggests a reduction from the unlearning problem to that of *data attribution, where the goal is to predict the effect of changing the training set on a model's outputs. Thus motivated, we propose the following meta-algorithm, which we call Datamodel Matching (DMM): given a trained model, we (a) use data attribution to predict the output of the model if it were re-trained on all but the forget set points; then (b) fine-tune the pre-trained model to match these predicted outputs. In a simple convex setting, we show how this approach provably outperforms a variety of iterative unlearning algorithms. Empirically, we use a combination of existing evaluations and a new metric based on the KL-divergence to show that even in non-convex settings, DMM achieves strong unlearning performance relative to existing algorithms. An added benefit of DMM is that it is a meta-algorithm, in the sense that future advances in data attribution translate directly into better unlearning algorithms, pointing to a clear direction for future progress in unlearning.