On Measuring Localization of Shortcuts in Deep Networks

Shortcuts, spurious rules that perform well during training but fail to generalize, present a major challenge to the reliability of deep networks (Geirhos et al., 2020). However, the impact of shortcuts on feature representations remains understudied, obstructing the design of principled shortcut-mitigation methods. To overcome this limitation, we investigate the layer-wise localization of shortcuts in deep models. Our novel experiment design quantifies the layer-wise contribution to accuracy degradation caused by a shortcut-inducing skew by counterfactual training on clean and skewed datasets. We employ our design to study shortcuts on CIFAR-10, Waterbirds, and CelebA datasets across VGG, ResNet, DeiT, and ConvNeXt architectures. We find that shortcut learning is not localized in specific layers but distributed throughout the network. Different network parts play different roles in this process: shallow layers predominantly encode spurious features, while deeper layers predominantly forget core features that are predictive on clean data. We also analyze the differences in localization and describe its principal axes of variation. Finally, our analysis of layer-wise shortcut-mitigation strategies suggests the hardness of designing general methods, supporting dataset- and architecture-specific approaches instead.

LARP: Learner-Agnostic Robust Data Prefiltering

The widespread availability of large public datasets is a key factor behind the recent successes of statistical inference and machine learning methods. However, these datasets often contain some low-quality or contaminated data, to which many learning procedures are sensitive. Therefore, the question of whether and how public datasets should be prefiltered to facilitate accurate downstream learning arises. On a technical level this requires the construction of principled data prefiltering methods which are learner-agnostic robust, in the sense of provably protecting a set of pre-specified downstream learners from corrupted data. In this work, we formalize the problem of Learner-Agnostic Robust data Prefiltering (LARP), which aims at finding prefiltering procedures that minimize a worst-case loss over a pre-specified set of learners. We first instantiate our framework in the context of scalar mean estimation with Huber estimators under the Huber data contamination model. We provide a hardness result on a specific problem instance and analyze several natural prefiltering procedures. Our theoretical results indicate that performing LARP on a heterogeneous set of learners leads to some loss in model performance compared to the alternative of prefiltering data for each learner/use-case individually. We explore the resulting utility loss and its dependence on the problem parameters via extensive experiments on real-world image and tabular data, observing statistically significant reduction in utility. Finally, we model the trade-off between the utility drop and the cost of repeated (learner-specific) prefiltering within a game-theoretic framework and showcase benefits of LARP for large datasets.

On the Impact of Performative Risk Minimization for Binary Random Variables

Performativity, the phenomenon where outcomes are influenced by predictions, is particularly prevalent in social contexts where individuals strategically respond to a deployed model. In order to preserve the high accuracy of machine learning models under distribution shifts caused by performativity, Perdomo et al. (2020) introduced the concept of performative risk minimization (PRM). While this framework ensures model accuracy, it overlooks the impact of the PRM on the underlying distributions and the predictions of the model. In this paper, we initiate the analysis of the impact of PRM, by studying performativity for a sequential performative risk minimization problem with binary random variables and linear performative shifts. We formulate two natural measures of impact. In the case of full information, where the distribution dynamics are known, we derive explicit formulas for the PRM solution and our impact measures. In the case of partial information, we provide performative-aware statistical estimators, as well as simulations. Our analysis contrasts PRM to alternatives that do not model data shift and indicates that PRM can have amplified side effects compared to such methods.

Incentivizing Truthful Collaboration in Heterogeneous Federated Learning

It is well-known that Federated Learning (FL) is vulnerable to manipulated updates from clients. In this work we study the impact of data heterogeneity on clients' incentives to manipulate their updates. We formulate a game in which clients may upscale their gradient updates in order to ``steer'' the server model to their advantage. We develop a payment rule that disincentivizes sending large gradient updates, and steers the clients towards truthfully reporting their gradients. We also derive explicit bounds on the clients' payments and the convergence rate of the global model, which allows us to study the trade-off between heterogeneity, payments and convergence.

COMPL-AI Framework: A Technical Interpretation and LLM Benchmarking Suite for the EU Artificial Intelligence Act

The EU's Artificial Intelligence Act (AI Act) is a significant step towards responsible AI development, but lacks clear technical interpretation, making it difficult to assess models' compliance. This work presents COMPL-AI, a comprehensive framework consisting of (i) the first technical interpretation of the EU AI Act, translating its broad regulatory requirements into measurable technical requirements, with the focus on large language models (LLMs), and (ii) an open-source Act-centered benchmarking suite, based on thorough surveying and implementation of state-of-the-art LLM benchmarks. By evaluating 12 prominent LLMs in the context of COMPL-AI, we reveal shortcomings in existing models and benchmarks, particularly in areas like robustness, safety, diversity, and fairness. This work highlights the need for a shift in focus towards these aspects, encouraging balanced development of LLMs and more comprehensive regulation-aligned benchmarks. Simultaneously, COMPL-AI for the first time demonstrates the possibilities and difficulties of bringing the Act's obligations to a more concrete, technical level. As such, our work can serve as a useful first step towards having actionable recommendations for model providers, and contributes to ongoing efforts of the EU to enable application of the Act, such as the drafting of the GPAI Code of Practice.

Simplicity Bias of Two-Layer Networks beyond Linearly Separable Data

Simplicity bias, the propensity of deep models to over-rely on simple features, has been identified as a potential reason for limited out-of-distribution generalization of neural networks (Shah et al., 2020). Despite the important implications, this phenomenon has been theoretically confirmed and characterized only under strong dataset assumptions, such as linear separability (Lyu et al., 2021). In this work, we characterize simplicity bias for general datasets in the context of two-layer neural networks initialized with small weights and trained with gradient flow. Specifically, we prove that in the early training phases, network features cluster around a few directions that do not depend on the size of the hidden layer. Furthermore, for datasets with an XOR-like pattern, we precisely identify the learned features and demonstrate that simplicity bias intensifies during later training stages. These results indicate that features learned in the middle stages of training may be more useful for OOD transfer. We support this hypothesis with experiments on image data.

Provable Mutual Benefits from Federated Learning in Privacy-Sensitive Domains

Cross-silo federated learning (FL) allows data owners to train accurate machine learning models by benefiting from each others private datasets. Unfortunately, the model accuracy benefits of collaboration are often undermined by privacy defenses. Therefore, to incentivize client participation in privacy-sensitive domains, a FL protocol should strike a delicate balance between privacy guarantees and end-model accuracy. In this paper, we study the question of when and how a server could design a FL protocol provably beneficial for all participants. First, we provide necessary and sufficient conditions for the existence of mutually beneficial protocols in the context of mean estimation and convex stochastic optimization. We also derive protocols that maximize the total clients' utility, given symmetric privacy preferences. Finally, we design protocols maximizing end-model accuracy and demonstrate their benefits in synthetic experiments.

Strategic Data Sharing between Competitors

Collaborative learning techniques have significantly advanced in recent years, enabling private model training across multiple organizations. Despite this opportunity, firms face a dilemma when considering data sharing with competitors -- while collaboration can improve a company's machine learning model, it may also benefit competitors and hence reduce profits. In this work, we introduce a general framework for analyzing this data-sharing trade-off. The framework consists of three components, representing the firms' production decisions, the effect of additional data on model quality, and the data-sharing negotiation process, respectively. We then study an instantiation of the framework, based on a conventional market model from economic theory, to identify key factors that affect collaboration incentives. Our findings indicate a profound impact of market conditions on the data-sharing incentives. In particular, we find that reduced competition, in terms of the similarities between the firms' products, and harder learning tasks foster collaboration.

Incentivizing Honesty among Competitors in Collaborative Learning and Optimization

Collaborative learning techniques have the potential to enable training machine learning models that are superior to models trained on a single entity's data. However, in many cases, potential participants in such collaborative schemes are competitors on a downstream task, such as firms that each aim to attract customers by providing the best recommendations. This can incentivize dishonest updates that damage other participants' models, potentially undermining the benefits of collaboration. In this work, we formulate a game that models such interactions and study two learning tasks within this framework: single-round mean estimation and multi-round SGD on strongly-convex objectives. For a natural class of player actions, we show that rational clients are incentivized to strongly manipulate their updates, preventing learning. We then propose mechanisms that incentivize honest communication and ensure learning quality comparable to full cooperation. Lastly, we empirically demonstrate the effectiveness of our incentive scheme on a standard non-convex federated learning benchmark. Our work shows that explicitly modeling the incentives and actions of dishonest clients, rather than assuming them malicious, can enable strong robustness guarantees for collaborative learning.

Human-Guided Fair Classification for Natural Language Processing

Text classifiers have promising applications in high-stake tasks such as resume screening and content moderation. These classifiers must be fair and avoid discriminatory decisions by being invariant to perturbations of sensitive attributes such as gender or ethnicity. However, there is a gap between human intuition about these perturbations and the formal similarity specifications capturing them. While existing research has started to address this gap, current methods are based on hardcoded word replacements, resulting in specifications with limited expressivity or ones that fail to fully align with human intuition (e.g., in cases of asymmetric counterfactuals). This work proposes novel methods for bridging this gap by discovering expressive and intuitive individual fairness specifications. We show how to leverage unsupervised style transfer and GPT-3's zero-shot capabilities to automatically generate expressive candidate pairs of semantically similar sentences that differ along sensitive attributes. We then validate the generated pairs via an extensive crowdsourcing study, which confirms that a lot of these pairs align with human intuition about fairness in the context of toxicity classification. Finally, we show how limited amounts of human feedback can be leveraged to learn a similarity specification that can be used to train downstream fairness-aware models.