Machine unlearning (MU) aims to remove the influence of particular data points from the learnable parameters of a trained machine learning model. This is a crucial capability in light of data privacy requirements, trustworthiness, and safety in deployed models. MU is particularly challenging for deep neural networks (DNNs), such as convolutional nets or vision transformers, as such DNNs tend to memorize a notable portion of their training dataset. Nevertheless, the community lacks a rigorous and multifaceted study that looks into the success of MU methods for DNNs. In this paper, we investigate 18 state-of-the-art MU methods across various benchmark datasets and models, with each evaluation conducted over 10 different initializations, a comprehensive evaluation involving MU over 100K models. We show that, with the proper hyperparameters, Masked Small Gradients (MSG) and Convolution Transpose (CT), consistently perform better in terms of model accuracy and run-time efficiency across different models, datasets, and initializations, assessed by population-based membership inference attacks (MIA) and per-sample unlearning likelihood ratio attacks (U-LiRA). Furthermore, our benchmark highlights the fact that comparing a MU method only with commonly used baselines, such as Gradient Ascent (GA) or Successive Random Relabeling (SRL), is inadequate, and we need better baselines like Negative Gradient Plus (NG+) with proper hyperparameter selection.

Despite significant advances in quality and complexity of the generations in text-to-image models, prompting does not always lead to the desired outputs. Controlling model behaviour by directly steering intermediate model activations has emerged as a viable alternative allowing to reach concepts in latent space that may otherwise remain inaccessible by prompt. In this work, we introduce a set of experiments to deepen our understanding of concept reachability. We design a training data setup with three key obstacles: scarcity of concepts, underspecification of concepts in the captions, and data biases with tied concepts. Our results show: (i) concept reachability in latent space exhibits a distinct phase transition, with only a small number of samples being sufficient to enable reachability, (ii) where in the latent space the intervention is performed critically impacts reachability, showing that certain concepts are reachable only at certain stages of transformation, and (iii) while prompting ability rapidly diminishes with a decrease in quality of the dataset, concepts often remain reliably reachable through steering. Model providers can leverage this to bypass costly retraining and dataset curation and instead innovate with user-facing control mechanisms.

Adversaries have been targeting unique identifiers to launch typo-squatting, mobile app squatting and even voice squatting attacks. Anecdotal evidence suggest that online social networks (OSNs) are also plagued with accounts that use similar usernames. This can be confusing to users but can also be exploited by adversaries. However, to date no study characterizes this problem on OSNs. In this work, we define the username squatting problem and design the first multi-faceted measurement study to characterize it on X. We develop a username generation tool (UsernameCrazy) to help us analyze hundreds of thousands of username variants derived from celebrity accounts. Our study reveals that thousands of squatted usernames have been suspended by X, while tens of thousands that still exist on the network are likely bots. Out of these, a large number share similar profile pictures and profile names to the original account signalling impersonation attempts. We found that squatted accounts are being mentioned by mistake in tweets hundreds of thousands of times and are even being prioritized in searches by the network's search recommendation algorithm exacerbating the negative impact squatted accounts can have in OSNs. We use our insights and take the first step to address this issue by designing a framework (SQUAD) that combines UsernameCrazy with a new classifier to efficiently detect suspicious squatted accounts. Our evaluation of SQUAD's prototype implementation shows that it can achieve 94% F1-score when trained on a small dataset.

We introduce an analytical framework to quantify the changes in a machine learning algorithm's output distribution following the inclusion of a few data points in its training set, a notion we define as leave-one-out distinguishability (LOOD). This is key to measuring data **memorization** and information **leakage** as well as the **influence** of training data points in machine learning. We illustrate how our method broadens and refines existing empirical measures of memorization and privacy risks associated with training data. We use Gaussian processes to model the randomness of machine learning algorithms, and validate LOOD with extensive empirical analysis of leakage using membership inference attacks. Our analytical framework enables us to investigate the causes of leakage and where the leakage is high. For example, we analyze the influence of activation functions, on data memorization. Additionally, our method allows us to identify queries that disclose the most information about the training data in the leave-one-out setting. We illustrate how optimal queries can be used for accurate **reconstruction** of training data.

It is known that deep neural networks, trained for the classification of non-sensitive target attributes, can reveal sensitive attributes of their input data through internal representations extracted by the classifier. We take a step forward and show that deep classifiers can be trained to secretly encode a sensitive attribute of their input data into the classifier's outputs for the target attribute, at inference time. Our proposed attack works even if users have a full white-box view of the classifier, can keep all internal representations hidden, and only release the classifier's estimations for the target attribute. We introduce an information-theoretical formulation for such attacks and present efficient empirical implementations for training honest-but-curious (HBC) classifiers: classifiers that can be accurate in predicting their target attribute, but can also exploit their outputs to secretly encode a sensitive attribute. Our work highlights a vulnerability that can be exploited by malicious machine learning service providers to attack their user's privacy in several seemingly safe scenarios; such as encrypted inferences, computations at the edge, or private knowledge distillation. Experimental results on several attributes in two face-image datasets show that a semi-trusted server can train classifiers that are not only perfectly honest but also accurately curious. We conclude by showing the difficulties in distinguishing between standard and HBC classifiers, discussing challenges in defending against this vulnerability of deep classifiers, and enumerating related open directions for future studies.