Flavio du Pin Calmon

Since being proposed in 2006, differential privacy has become a standard method for quantifying certain risks in publishing or sharing analyses of sensitive data. At its heart, differential privacy measures risk in terms of the differences between probability distributions, which is a central topic in information theory. A differentially private algorithm is a channel between the underlying data and the output of the analysis. Seen in this way, the guarantees made by differential privacy can be understood in terms of properties of this channel. In this article we examine a few of the key connections between information theory and the formulation/application of differential privacy, giving an “operational significance” for relevant information measures.

This study focuses on characterizing the channel of a reconfigurable intelligent surface (RIS)-assisted wireless system operating over Nakagami-$m$ fading channels. Although numerous studies have proposed approximate solutions for the RIS channel statistics (namely, the probability density function (PDF) and the cumulative distribution function (CDF)), only a limited number of studies have tackled this problem by resorting to exact approaches. Regrettably, as the number of RIS elements increases, these approaches lead to solutions that are computationally expensive or entail a high mathematical intricacy. This has prompted the analytical development undertaken in this work. Our emphasis is on an exact approach. Specifically, we furnish handy, tractable formulas for the PDF and the CDF of the investigated RIS channel. The expressions introduced in this study stand out as new contributions to the literature and are, to the best of our knowledge, among the most tractable and efficient exact solutions currently available. Numerical simulations demonstrate that our proposed PDF and CDF expressions offer superior efficiency compared to state-of-the-art solutions. The results also highlight that our approach is particularly attractive in scenarios where approximate solutions fall short, such as (i) a small number of RIS elements and (ii) pronounced fading imbalance, i.e., when the shape parameters of the Nakagami$m$ distributions associated with each communication link differ considerably. Furthermore, we conducted a performance assessment analysis for the considered RIS-assisted wireless communication system by deriving exact and asymptotic expressions for the key performance indicators, namely the outage probability (OP), average bit error rate (ABER), and ergodic capacity (EC).

Large-language models (LLMs) are now able to produce text that is, in many cases, seemingly indistinguishable from human-generated content. This has fueled the development of watermarks that imprint a ``signal'' in LLM-generated text with minimal perturbation of an LLM's output. This paper provides an analysis of text watermarking in a one-shot setting. Through the lens of hypothesis testing with side information, we formulate and analyze the fundamental trade-off between watermark detection power and distortion in generated textual quality. We argue that a key component in watermark design is generating a coupling between the side information shared with the watermark detector and a random partition of the LLM vocabulary. Our analysis identifies the optimal coupling and randomization strategy under the worst-case LLM next-token distribution that satisfies a min-entropy constraint. We provide a closed-form expression of the resulting detection rate under the proposed scheme and quantify the cost in a max-min sense. Finally, we provide an array of numerical results, comparing the proposed scheme with the theoretical optimum and existing schemes, in both synthetic data and LLM watermarking. Our code is available at https://github.com/Carol-Long/CC_Watermark

The aim of this paper is threefold: 1) we introduce a unified statistical characterization for the instantaneous power of a plethora of Gaussian-based fading models; 2) we propose a general framework for the power sum of either independent and identically distributed (i.i.d.) or independent and non-identically distributed (i.non-i.d.) fading distributions; and 3) we provide a unified performance assessment of wireless communications systems in the presence of small-scale fading channels. To accomplish these three goals, we first show that the instantaneous power of a broad Gaussian class of fading models is governed by a mixture of gamma (MG) distribution. Later, we propose novel, tractable, and efficient solutions for the exact sum statistics of i.i.d. and i.non-i.d. MG variates. Lastly, we show that two key performance indicators—namely the average bit-error rate and the outage probability—of a maximal-ratio combining (MRC) diversity receiver subject to a wide variety of Gaussian-based fading channels can be expressed, in a unified fashion, as a weighted sum of the performance indicators of a single-branch wireless system subject to independent Nakagami- m fading channels. Extensive numerical simulations reveal the outstanding improvement in computational efficiency and mathematical tractability of our derived expressions when compared with state-of-the-art solutions.

Machine learning (ML) is widely used to moderate online content. Despite its scalability relative to human moderation, the use of ML introduces unique challenges to content moderation. One such challenge is predictive multiplicity: multiple competing models for content classification may perform equally well on average, yet assign conflicting predictions to the same content. This multiplicity can result from seemingly innocuous choices made during training, which do not meaningfully change the accuracy of the ML model, but can nevertheless change what the model gets wrong. We experimentally demonstrate how content moderation tools can arbitrarily classify samples as “toxic,” leading to arbitrary restrictions on speech. We use the principles set by the International Covenant on Civil and Political Rights (ICCPR), namely freedom of expression, non-discrimination, and procedural justice to interpret the effects of these findings in terms of Human Rights. We analyze (i) the extent of predictive multiplicity among popular state-of-the-art LLMs used for detecting “toxic” content; (ii) the disparate impact of this arbitrariness across social groups; and (iii) the magnitude of model multiplicity on content that is unanimously recognized as toxic by human annotators. Our findings indicate that the up-scaled algorithmic moderation risks legitimizing an “algorithmic leviathan”, where an algorithm disproportionately manages human rights. To mitigate such risks, our study underscores the need to identify and increase the transparency of arbitrariness in content moderation applications. Our findings have implications to content moderation and intermediary liability laws being discussed and passed in many countries, such as the Digital Services Act in the European Union, the Online Safety Act in the United Kingdom, and the recent TSE resolutions in Brazil.

Machine learning (ML) models used in prediction and classification tasks may display performance disparities across population groups determined by sensitive attributes (e.g., race, sex, age). We consider the problem of evaluating the performance of a fixed ML model across population groups defined by multiple sensitive attributes (e.g., race and sex and age). Here, the sample complexity for estimating the worst-case performance gap across groups (e.g., the largest difference in error rates) increases exponentially with the number of group-denoting sensitive attributes. To address this issue, we propose an approach to test for performance disparities based on Conditional Value-at-Risk (CVaR). By allowing a small probabilistic slack on the groups over which a model has approximately equal performance, we show that the sample complexity required for discovering performance violations is reduced exponentially to be at most upper bounded by the square root of the number of groups. As a byproduct of our analysis, when the groups are weighted by a specific prior distribution, we show that Rényi entropy of order 2/3 of the prior distribution captures the sample complexity of the proposed CVaR test algorithm. Finally, we also show that there exists a non-i.i.d. data collection strategy that results in a sample complexity independent of the number of groups.

Machine learning (ML) models can underperform on certain population groups due to choices made during model development and bias inherent in the data. We categorize sources of discrimination in the ML pipeline into two classes: aleatoric discrimination, which is inherent in the data distribution, and epistemic discrimination, which is due to decisions made during model development. We quantify aleatoric discrimination by determining the performance limits of a model under fairness constraints, assuming perfect knowledge of the data distribution. We demonstrate how to characterize aleatoric discrimination by applying Blackwell's results on comparing statistical experiments. We then quantify epistemic discrimination as the gap between a model's accuracy when fairness constraints are applied and the limit posed by aleatoric discrimination. We apply this approach to benchmark existing fairness interventions and investigate fairness risks in data with missing values. Our results indicate that state-of-the-art fairness interventions are effective at removing epistemic discrimination on standard (overused) tabular datasets. However, when data has missing values, there is still significant room for improvement in handling aleatoric discrimination.