Papers

We propose a method for non-parametric conditional distribution estimation based on partitioning covariate-sorted observations into contiguous bins and using the within-bin empirical CDF as the predictive distribution. Bin boundaries are chosen to minimise the total leave-one-out Continuous Ranked Probability Score (LOO-CRPS), which admits a closed-form cost function with $O(n^2 \log n)$ precomputation and $O(n^2)$ storage; the globally optimal $K$-partition is recovered by a dynamic programme in $O(n^2 K)$ time. Minimisation of within-sample LOO-CRPS turns out to be inappropriate for selecting $K$ as it results in in-sample optimism. We instead select $K$ by $K$-fold cross-validation of test CRPS, which yields a U-shaped criterion with a well-defined minimum. Having selected $K^*$ and fitted the full-data partition, we form two complementary predictive objects: the Venn prediction band and a conformal prediction set based on CRPS as the nonconformity score, which carries a finite-sample marginal coverage guarantee at any prescribed level $\varepsilon$. The conformal prediction is transductive and data-efficient, as all observations are used for both partitioning and p-value calculation, with no need to reserve a hold-out set. On real benchmarks against split-conformal competitors (Gaussian split conformal, CQR, CQR-QRF, and conformalized isotonic distributional regression), the method produces substantially narrower prediction intervals while maintaining near-nominal coverage.

StreamSampling$.$jl is a Julia library designed to provide general and efficient methods for sampling from data streams in a single pass, even when the total number of items is unknown. In this paper, we describe the capabilities of the library and its advantages over traditional sampling procedures, such as maintaining a small, constant memory footprint and avoiding the need to fully materialize the stream in memory. Furthermore, we provide empirical benchmarks comparing online sampling methods against standard approaches, demonstrating performance and memory improvements.

Stochastic epidemic models can estimate infection and removal rates, and derived quantities such as the basic reproductive number ($R_0$), when both infection and removal times are observed. In practice, however, removal times are often available while infection times are not, and existing methods that rely only on removal times can become unstable or biased. We study inference for stochastic SIR/SEIR models in a partial–observation setting. We develop imputation–based estimators that use a small calibration sample of fully observed infectious periods, derive closed–form expressions for the pairwise exposure terms they require, and use a studentized parametric bootstrap for bias correction and uncertainty quantification. In simulations, removal time–only methods performed poorly in moderate to large $R_0$ scenarios, while observing even tens of complete infectious periods substantially improved the estimation of the infection rate. A reanalysis of the 1861 Hagelloch measles outbreak under simulated missingness recovered stable qualitative differences in transmission between school classes. Based on our results, we advocate for the targeted collection of a modest number of complete infectious periods as a means of improving surveillance in the early stages of an epidemic.

Evaluating treatment effect heterogeneity across patient subgroups is a fundamental aspect of clinical trial analysis. These analyses have inherent limitations due to small sample sizes and the substantial number of subgroups investigated. There is a tendency to focus on extreme estimates, which may reflect random variation rather than true effects, potentially leading to spurious clinical conclusions. Statisticians in regulatory agencies and pharmaceutical companies have begun considering shrinkage methods grounded in Bayesian theory. These methods incorporate priors on treatment effect heterogeneity, which shrink subgroup estimates towards the overall treatment effect. Various shrinkage estimators have been proposed, yet it remains unclear which perform best. This work provides a unified presentation and software implementation of shrinkage methods. It also provides simulation comparisons of one-way and global shrinkage methods for two simulation set-ups. One-way models fit a separate shrinkage model for each subgrouping variable while global models include all subgroup indicators. Both can derive standardized subgroup-specific treatment effects. Across all simulation scenarios, shrinkage methods outperformed the standard subgroup estimator in terms of mean squared error. They were also more efficient in identifying a non-efficacious subgroup. Global shrinkage models tended to have smaller mean squared error and less dependence on hyperprior parameters than one-way models, but also exhibited slightly larger bias and worse frequentist coverage of credible intervals. For both models, hyperprior choices anchored in trial assumptions about the anticipated overall treatment effect size performed well. We conclude that some shrinkage is preferable to none and advocate routine inclusion of shrunken estimates in clinical forest plots to facilitate robust decision-making.

High-dimensional biomedical studies require models that are simultaneously accurate, sparse, and interpretable, yet exact best subset selection for generalized linear models is computationally intractable. We develop a scalable method that combines a continuous Boolean relaxation of the subset problem with a Frank–Wolfe algorithm driven by envelope gradients. The resulting method, which we refer to as COMBSS-GLM, is simple to implement, requires one penalized generalized linear model fit per iteration, and produces sparse models along a model-size path. Theoretically, we identify a curvature-based parameter regime in which the relaxed objective is concave in the selection weights, implying that global minimizers occur at binary corners. Empirically, in logistic and multinomial simulations across low- and high-dimensional correlated settings, the proposed method consistently improves variable-selection quality relative to established penalised likelihood competitors while maintaining strong predictive performance. In biomedical applications, it recovers established loci in a binary-outcome rice genome-wide association study and achieves perfect multiclass test accuracy on the Khan SRBCT cancer dataset using a small subset of genes. Open-source implementations are available in R at https://github.com/benoit-liquet/COMBSS-GLM-R and in Python at https://github.com/saratmoka/COMBSS-GLM-Python.

This paper develops a unified framework for measuring concentration in weighted systems embedded in networks of interactions. While traditional indices such as the Herfindahl-Hirschman Index capture dispersion in weights, they neglect the topology of relationships among the elements receiving those weights. To address this limitation, we introduce a family of topology-aware concentration indices that jointly account for weight distributions and network structure. At the core of the framework lies a baseline Network Concentration Index (NCI), defined as a normalized quadratic form that measures the fraction of potential weighted interconnection realized along observed network links. Building on this foundation, we construct a flexible class of extensions that modify either the interaction structure or the normalization benchmark, including weighted, density-adjusted, null-model, degree-constrained, transformed-data, and multi-layer variants. This family of indices preserves key properties such as normalization, invariance, and interpretability, while allowing concentration to be evaluated across different dimensions of dependence, including intensity, higher-order interactions, and extreme events. Theoretical results characterize the indices and establish their relationship with classical concentration and network measures. Empirical and simulation evidence demonstrate that systems with identical weight distributions may exhibit markedly different levels of structural concentration depending on network topology, highlighting the additional information captured by the proposed framework. The approach is broadly applicable to economic, financial, and complex systems in which weighted elements interact through networks.

Governments routinely adjust capacity in rationed programs such as university fields, medical training and public housing, where admitting one individual displaces others and triggers chains of reallocation. We show that in such settings, the standard multi-treatment two-stage least squares (2SLS) coefficient identifies exactly the total societal effect of a marginal expansion, including all downstream reallocations. The result is an algebraic identity: under instrument relevance and a single alignment condition, satisfied in centralized admissions systems, the 2SLS coefficient equals the general-equilibrium shadow value of relaxing a capacity constraint, while the single-instrument Wald ratio captures only the direct effect. Their difference recovers the full equilibrium adjustment without additional structure. Monotonicity is not required. The identity extends beyond queue-based allocation to any fixed-supply setting, including competitive markets with price instruments. We apply the framework to two policy questions in Swedish university admissions, where marginal students are allocated across fields through a centralized lottery mechanism. First, revisiting the debate on whether economics and business education erodes prosocial values, we find that the direct effect of expanding business on charitable giving is precisely zero, but expanding the less competitive fields that business students are displaced from has large prosocial effects. Second, analyzing gender-targeted STEM policies, we find that admitting four women to competitive STEM generates one additional male STEM degree through downstream vacancies. Both are general-equilibrium effects invisible to single-instrument methods.

We study identifiability of finite mixtures of Dirichlet distributions on the interior of the simplex. We first prove a shift identity showing that every Dirichlet density can be written as a mixture of $J$ shifted Dirichlet densities, where $J-1$ is the dimension of the simplex support, which yields non-identifiability on the full parameter space. We then show that identifiability is recovered on a fixed-total parameter slice and on restricted box-type regions. On the full parameter space, we prove that any nontrivial linear relation among Dirichlet kernels must involve at least $J$ coefficients sharing a common sign, and deduce that mixtures with fewer than $J$ atoms are identifiable. We further report direct non-identifiability implications for unrestricted finite mixtures of generalized Dirichlet, Dirichlet-multinomial, fixed-topic-matrix latent Dirichlet allocation, Beta-Liouville, and inverted Beta-Liouville models.

Learning causal relations from observational data is a fundamental problem with wide-ranging applications across many fields. Constraint-based methods infer the underlying causal structure by performing conditional independence tests. However, existing algorithms such as the prominent PC algorithm need to perform a large number of independence tests, which in the worst case is exponential in the maximum degree of the causal graph. Despite extensive research, it remains unclear if there exist algorithms with better complexity without additional assumptions. Here, we establish an algorithm that achieves a better complexity of $p^{\mathcal{O}(s)}$ tests, where $p$ is the number of nodes in the graph and $s$ denotes the maximum undirected clique size of the underlying essential graph. Complementing this result, we prove that any constraint-based algorithm must perform at least $2^{\Omega(s)}$ conditional independence tests, establishing that our proposed algorithm achieves exponent-optimality up to a logarithmic factor in terms of the number of conditional independence tests needed. Finally, we validate our theoretical findings through simulations, on semi-synthetic gene-expression data, and real-world data, demonstrating the efficiency of our algorithm compared to existing methods in terms of number of conditional independence tests needed.

Bayesian inference is a powerful tool for parameter estimation and uncertainty quantification in dynamical systems. However, for nonlinear oscillator networks such as Kuramoto models, widely used to study synchronization phenomena in physics, biology, and engineering, inference is often computationally prohibitive due to high-dimensional state spaces and intractable likelihood functions. We present an amortized Bayesian inference approach that learns a neural approximation of the posterior from simulated phase dynamics, enabling fast, scalable inference without repeated sampling or optimization. Applied to synthetic Kuramoto networks, the method shows promising results in approximating posterior distributions and capturing uncertainty, with computational savings compared to traditional Bayesian techniques. These findings suggest that amortized inference is a practical and flexible framework for uncertainty-aware analysis of oscillator networks.

This work develops a multivariate extension of the Fixed Rank Kriging (FRK) framework for spatial prediction in settings where multiple spatial processes may provide complementary information. The goal is to preserve the computational efficiency, the ability to operate without assuming stationarity over the domain, and the spatial support flexibility of FRK, while incorporating cross-process dependence. To this end, we employ a multiresolution coregionalization structure for the latent spatial effects, in which spatial basis functions are combined with Gaussian Markov Random Field coefficients. An estimation procedure based on the expectation-maximization algorithm is developed, designed to exploit the multiresolution latent structure. Through simulation studies, we examine when the proposed joint modeling is beneficial. We consider cases in which one process is observed more sparsely or is entirely unobserved in a subregion and find that the multivariate formulation is able to borrow information from the more densely observed process, producing coherent and accurate predictions even where direct observations are limited or absent. Finally, the model is applied to the analysis of PM10 concentrations in Northern Italy, illustrating its applicability in a real environmental context.

Recommendation systems (RSs) are increasingly used to guide job seekers on online platforms, yet the algorithms currently deployed are typically optimized for predictive objectives such as clicks, applications, or hires, rather than job seekers' welfare. We develop a job-search model with an application stage in which the value of a vacancy depends on two dimensions: the utility it delivers to the worker and the probability that an application succeeds. The model implies that welfare-optimal RSs rank vacancies by an expected-surplus index combining both, and shows why rankings based solely on utility, hiring probabilities, or observed application behavior are generically suboptimal, an instance of the inversion problem between behavior and welfare. We test these predictions and quantify their practical importance through two randomized field experiments conducted with the French public employment service. The first experiment, comparing existing algorithms and their combinations, provides behavioral evidence that both dimensions shape application decisions. Guided by the model and these results, the second experiment extends the comparison to an RS designed to approximate the welfare-optimal ranking. The experiments generate exogenous variation in the vacancies shown to job seekers, allowing us to estimate the model, validate its behavioral predictions, and construct a welfare metric. Algorithms informed by the model-implied optimal ranking substantially outperform existing approaches and perform close to the welfare-optimal benchmark. Our results show that embedding predictive tools within a simple job-search framework and combining it with experimental evidence yields recommendation rules with substantial welfare gains in practice.

We study the approximation of operators acting on probability measures on a product space with prescribed marginal. Let $I$ be a label space endowed with a reference measure $\lambda$, and define $\cal M_\lambda$ as the set of probability measures on $I\times \mathbb{R}^d$ with first marginal $\lambda$. By disintegration, elements of $\cal M_\lambda$ correspond to families of labeled conditional distributions. Operators defined on this constrained measure space arise naturally in mean-field control problems with heterogeneous, non-exchangeable agents. Our main theoretical result establishes a universal approximation theorem for continuous operators on $\cal M_\lambda$. The proof combines cylindrical approximations of probability measures with DeepONet-type branch-trunk neural architecture, yielding finite-dimensional representations of such operators. We further introduce a sampling strategy for generating training measures in $\cal M_\lambda$, enabling practical learning of such conditional mean-field operators. We apply the method to the numerical resolution of mean-field control problems with heterogeneous interactions, thereby extending previous neural approaches developed for homogeneous (exchangeable) systems. Numerical experiments illustrate the accuracy and computational effectiveness of the proposed framework.

Time-dependent reliability analysis of nonlinear dynamical systems under stochastic excitations is a critical yet computationally demanding task. Conventional approaches, such as Monte Carlo simulation, necessitate repeated evaluations of computationally expensive numerical solvers, leading to significant computational bottlenecks. To address this challenge, we propose CoNBONet, a neuroscience-inspired surrogate model that enables fast, energy-efficient, and uncertainty-aware reliability analysis, providing a scalable alternative to techniques such as Monte Carlo simulations. CoNBONet, short for Conformalized Neuroscience-inspired Bayesian Operator Network, leverages the expressive power of deep operator networks while integrating neuroscience-inspired neuron models to achieve fast, low-power inference. Unlike traditional surrogates such as Gaussian processes, polynomial chaos expansions, or support vector regression, that may face scalability challenges for high-dimensional, time-dependent reliability problems, CoNBONet offers fast and energy-efficient inference enabled by a neuroscience-inspired network architecture, calibrated uncertainty quantification with theoretical guarantees via split conformal prediction, and strong generalization capability through an operator-learning paradigm that maps input functions to system response trajectories. Validation of the proposed CoNBONet for various nonlinear dynamical systems demonstrates that CoNBONet preserves predictive fidelity, and achieves reliable coverage of failure probabilities, making it a powerful tool for robust and scalable reliability analysis in engineering design.

While asset-pricing models increasingly recognize that factor risk premia are subject to structural change, existing literature typically assumes that investors correctly account for such instability. This paper studies how investors instead learn under a misspecified model that underestimates structural breaks. We propose a minimal Bayesian framework in which this misspecification generates persistent prediction errors and pricing distortions, and we introduce an empirically tractable measure of mislearning intensity $(\Delta_t)$ based on predictive likelihood ratios. The empirical results yield three main findings. First, in benchmark factor systems, elevated mislearning does not forecast a deterministic short-run collapse in performance; instead, it is associated with stronger long-horizon returns and Sharpe ratios, consistent with an equilibrium premium for acute model uncertainty. Second, in a broader anomaly universe, this pricing relation does not generalize uniformly: mislearning is more strongly associated with future drawdowns, downside semivolatility, and other measures of instability, with substantial heterogeneity across anomaly families. Third, the cross-sectional relation between instability and mislearning is inherently conditional: while a monotonic link between break-proneness and average mislearning does not hold in the full cross-section, it re-emerges in low-friction (low-IVOL) environments where break-state severity is more comparable across assets.

In many classification problems, misclassification costs are highly asymmetric, while training labels are often corrupted due to measurement error, annotator variability, or adversarial noise. The Neyman-Pearson multiclass classification (NPMC) framework addresses such asymmetry by controlling class-specific errors, but existing methods assume that training labels are correctly observed. To our knowledge, no existing approach handles NPMC under label noise in the multiclass setting, and the only binary method requires prior knowledge of the noise mechanism. A fundamental difficulty is that, without structural assumptions, noisy-label models are non-identifiable: distinct combinations of class-conditional distributions and noise mechanisms can induce the same observed distribution, preventing recovery of the quantities required for error control. We show that the exponential tilting density ratio model restores identifiability, and leverage this structure to develop an empirical likelihood approach for NPMC with noisy labels. The proposed method jointly estimates clean-label class proportions, posterior probabilities, and the noise mechanism from noisy data, without requiring prior knowledge of the confusion matrix. An expectation-maximization algorithm enables efficient computation. The resulting estimators are root n consistent and asymptotically normal, and the induced classifiers satisfy Neyman-Pearson oracle inequalities in both binary and multiclass settings. Simulation and real-data experiments demonstrate near-oracle performance.

Compliance monitoring in rule-governed domains (tax administration, clinical protocol adherence, environmental regulation) faces three structural obstacles that standard machine learning does not simultaneously address: the absence of labeled outcomes at deployment, strategically missing observations where non-compliant entities selectively withhold evidence, and a regulatory environment that changes faster than any supervised model can be retrained. We introduce Rule-State Inference (RSI), a Bayesian framework that reverses the usual paradigm. Rather than learning rules from data, RSI treats an authoritative, formalized rule set as structured Bayesian priors and infers the latent compliance state of a population through mean-field variational inference with exact coordinate-ascent updates. The central modeling object is a joint latent state per regulatory period: a global compliance-culture factor eta and per-rule components for activation, population compliance level, and parametric drift. RSI delivers three formal guarantees: O(n_k + K) regulatory adaptability per rule update; Bernstein-von Mises consistency for the identifiable continuous components; and monotone ELBO convergence at every iteration. We instantiate RSI on the Togolese fiscal system on a benchmark of 2,000 synthetic enterprises grounded in official regulatory law; full numerical validation is forthcoming. The framework is designed for direct extension to Sequential RSI, a state-space formulation where the posterior from one regulatory period becomes the prior for the next, yielding an exact Kalman filter for compliance-trajectory tracking and entity-level Bayesian scoring.

In many learning systems, such as activity recognition systems, as new data collection methods continue to emerge in various dynamic environmental applications, the attributes of instances accumulate incrementally, with data being stored in gradually expanding feature spaces. How to design theoretically guaranteed algorithms to effectively cluster this special type of data stream, commonly referred to as activity recognition, remains unexplored. Compared to traditional scenarios, we will face at least two fundamental questions in this feature incremental scenario. (i) How to design preliminary and effective algorithms to address the feature incremental clustering problem? (ii) How to analyze the generalization bounds for the proposed algorithms and under what conditions do these algorithms provide a strong generalization guarantee? To address these problems, by tailoring the most common clustering algorithm, i.e., $k$-means, as an example, we propose four types of Feature Incremental Clustering (FIC) algorithms corresponding to different situations of data access: Feature Tailoring (FT), Data Reconstruction (DR), Data Adaptation (DA), and Model Reuse (MR), abbreviated as FIC-FT, FIC-DR, FIC-DA, and FIC-MR. Subsequently, we offer a detailed analysis of the generalization error bounds for these four algorithms and highlight the critical factors influencing these bounds, such as the amounts of training data, the complexity of the hypothesis space, the quality of pre-trained models, and the discrepancy of the reconstruction feature distribution. The numerical experiments show the effectiveness of the proposed algorithms, particularly in their application to activity recognition clustering tasks.

Vector quantile regression (VQR) is an optimal transport (OT)-based framework that extends linear quantile regression to vector-valued response variables and can be formulated as an OT problem with a mean-independence constraint. In this paper, we study Sinkhorn-type algorithms for VQR with entropic regularization, building on the authors' previous work on its duality theory. We first study a direct adaptation of the classical Sinkhorn iteration based on solving the full Schrödinger-type system characterizing the dual potentials, which requires solving an implicit functional equation at each iteration. Since the implicit equation does not admit a closed-form solution in general, we consider a modified algorithm that replaces the implicit update with a gradient ascent step, resulting in a computational scheme that does not involve any subroutine. The latter algorithm can be viewed as an implementable version of the former, idealized scheme that serves as a theoretical benchmark. For both algorithms, and for general compactly supported marginals, we establish linear convergence in both the dual objective value and the iterates. A key innovation in our analysis is the derivation of explicit quantitative bounds on the dual potentials and Sinkhorn iterates.

Ordinary differential equation (ODE) models are widely used to describe systems in many areas of science. To ensure these models provide accurate and interpretable representations of real-world dynamics, it is often necessary to infer parameters from data, which involves specifying the form of the ODE system as well as a statistical model describing the observational process. A popular and convenient choice for the error model is a Gaussian distribution with constant variance. However, the choice may not be realistic in many systems, since the variance of the observational error may vary over time or have some dependence on the system state (heteroscedastic), reflecting changes in measurement conditions, environmental fluctuations, or intrinsic system variability. Misspecification of the error model can lead to substantial inaccuracies of the posterior estimates of the ODE model parameters and predictions. More elaborate parametric error models could be specified, but this would increase computational cost because additional parameters would need to be estimated within the MCMC procedure and may still be misspecified. In this work we propose a two-step semi-parametric framework for Bayesian parameter estimation of ODE model parameters when there exists heteroscedasticity in the error process. The first step applies a heteroscedastic Gaussian process to estimate the time-dependent error, and the second step performs Bayesian inference for the ODE model parameters using the estimated time-dependent error estimated from step one in the likelihood function. Through a simulation study and two real-world applications, we demonstrate that the proposed approach yields more reliable posterior inference and predictive uncertainty compared to the standard homoscedastic models. Although our focus is on heteroscedasticity, the framework could be applied to handle more complex error processes.

Knowledge distillation has emerged as a powerful technique for compressing large language models (LLMs) into efficient, deployable architectures while preserving their advanced capabilities. Recent advances in low-rank knowledge distillation, particularly methods like Low-Rank Clone (LRC), have demonstrated remarkable empirical success, achieving comparable performance to full-parameter distillation with significantly reduced training data and computational overhead. However, the theoretical foundations underlying these methods remain poorly understood. In this paper, we establish a rigorous theoretical framework for low-rank knowledge distillation in language models. We prove that under mild assumptions, low-rank projection preserves the optimization dynamics, yielding explicit convergence rates of $O(1/\sqrt{T})$. We derive generalization bounds that characterize the fundamental trade-off between model compression and generalization capability, showing that the generalization error scales with the rank parameter as $O(r(m+n)/\sqrt{n})$. Furthermore, we provide an information-theoretic analysis of the activation cloning mechanism, revealing its role in maximizing the mutual information between the teacher's and student's intermediate representations. Our theoretical results offer principled guidelines for rank selection, mathematically suggesting an optimal rank $r^* = O(\sqrt{n})$ where $n$ is the sample size. Experimental validation on standard language modeling benchmarks confirms our theoretical predictions, demonstrating that the empirical convergence, rank scaling, and generalization behaviors align closely with our bounds.

After the seminal Benjamini–Hochberg (BH) procedure for controlling the false discovery rate (FDR) was proposed, dozens of papers have attempted to improve its power by adapting to the unknown proportion of nulls. We observe that most null-proportion estimates are simply compound e-values in disguise, and thus most adaptive FDR procedures can be interpreted as instances of the e-weighted BH (ep-BH) procedure of Ignatiadis, Wang, and Ramdas [2024], i.e., the BH procedure weighted by compound e-values. This lens helps us show that most existing procedures are inadmissible, and we provide uniform improvements to them. While the improvements are small in practice, they still come for free (without additional assumptions), and help unify the literature. We also use our "leave-one-out ep-BH method" to design a new method with finite-sample FDR control for the simultaneous t-test setting.

Extreme economic outcomes are not shaped by tails alone. They are also shaped by unequal access to opportunities. This paper develops a theory of heterogeneous extremes by taking the distribution of opportunity access as the object of study. In a mixed Poisson search setting, normalized maxima admit a Laplace mixture representation that yields order comparisons and a clean benchmark against the homogeneous economy. The main contribution is geometric: a canonical coupling turns differences in heterogeneity into optimal transport bounds for the whole induced law of extremes, the full schedule of top quantiles, and structured counterfactual paths between economies. The paper also derives a second order expansion that separates classical extreme value approximation error from heterogeneity effects. As a complementary normative exercise, it studies an entropy regularized design problem for reallocating opportunities under a mean constraint. A stylized labor market network application interprets heterogeneity as unequal access to job opportunities and shows how the framework can be used for tail counterfactuals and robustness analysis of top wage distributions.

The widespread use of AI and ML models in sensitive areas raises significant concerns about fairness. While the research community has introduced various methods for bias mitigation in binary classification tasks, the issue remains under-explored in multi-class classification settings. To address this limitation, in this paper, we first formulate the problem of fair learning in multi-class classification as a multi-objective problem between effectiveness (i.e., prediction correctness) and multiple linear fairness constraints. Next, we propose a Generalised Exponentiated Gradient (GEG) algorithm to solve this task. GEG is an in-processing algorithm that enhances fairness in binary and multi-class classification settings under multiple fairness definitions. We conduct an extensive empirical evaluation of GEG against six baselines across seven multi-class and three binary datasets, using four widely adopted effectiveness metrics and three fairness definitions.

We study Constrained Online Convex Optimization with Memory (COCO-M), where both the loss and the constraints depend on a finite window of past decisions made by the learner. This setting extends the previously studied unconstrained online optimization with memory framework and captures practical problems such as the control of constrained dynamical systems and scheduling with reconfiguration budgets. For this problem, we propose the first algorithms that achieve sublinear regret and sublinear cumulative constraint violation under time-varying constraints, both with and without predictions of future loss and constraint functions. Without predictions, we introduce an adaptive penalty approach that guarantees sublinear regret and constraint violation. When short-horizon and potentially unreliable predictions are available, we reinterpret the problem as online learning with delayed feedback and design an optimistic algorithm whose performance improves as prediction accuracy improves, while remaining robust when predictions are inaccurate. Our results bridge the gap between classical constrained online convex optimization and memory-dependent settings, and provide a versatile learning toolbox with diverse applications.

Current experimental design techniques for dynamical systems often only incorporate measurement noise, while dynamical systems also involve process noise. To construct experimental designs we need to quantify their information content. The Fisher information matrix is a popular tool to do so. Calculating the Fisher information matrix for linear dynamical systems with both process and measurement noise involves estimating the uncertain dynamical states using a Kalman filter. The Fisher information matrix, however, depends on the true but unknown model parameters. In this paper we combine two methods to solve this issue and develop a robust experimental design methodology. First, Bayesian experimental design averages the Fisher information matrix over a prior distribution of possible model parameter values. Second, adaptive experimental design allows for this information to be updated as measurements are being gathered. This updated information is then used to adapt the remainder of the design.

The coordinate-exchange algorithm is commonly used to construct optimal experimental designs. Every execution of the coordinate-exchange algorithm produces a new, seemingly random, order of the selected design points. In this short communication, we study the order of the design points produced by the algorithm and conclude that certain orders appear much more often than others. As a result, an explicit randomization step of the design points is required before conducting an experiment using a design produced by a coordinate-exchange algorithm.

We introduce Generalized Discrete Diffusion from Snapshots (GDDS), a unified framework for discrete diffusion modeling that supports arbitrary noising processes over large discrete state spaces. Our formulation encompasses all existing discrete diffusion approaches, while allowing significantly greater flexibility in the choice of corruption dynamics. The forward noising process relies on uniformization and enables fast arbitrary corruption. For the reverse process, we derive a simple evidence lower bound (ELBO) based on snapshot latents, instead of the entire noising path, that allows efficient training of standard generative modeling architectures with clear probabilistic interpretation. Our experiments on large-vocabulary discrete generation tasks suggest that the proposed framework outperforms existing discrete diffusion methods in terms of training efficiency and generation quality, and beats autoregressive models for the first time at this scale. We provide the code along with a blog post on the project page : \href{https://oussamazekri.fr/gdds}{https://oussamazekri.fr/gdds}.

We propose closed-form conditional diffusion models for data assimilation. Diffusion models use data to learn the score function (defined as the gradient of the log-probability density of a data distribution), allowing them to generate new samples from the data distribution by reversing a noise injection process. While it is common to train neural networks to approximate the score function, we leverage the analytical tractability of the score function to assimilate the states of a system with measurements. To enable the efficient evaluation of the score function, we use kernel density estimation to model the joint distribution of the states and their corresponding measurements. The proposed approach also inherits the capability of conditional diffusion models of operating in black-box settings, i.e., the proposed data assimilation approach can accommodate systems and measurement processes without their explicit knowledge. The ability to accommodate black-box systems combined with the superior capabilities of diffusion models in approximating complex, non-Gaussian probability distributions means that the proposed approach offers advantages over many widely used filtering methods. We evaluate the proposed method on nonlinear data assimilation problems based on the Lorenz-63 and Lorenz-96 systems of moderate dimensionality and nonlinear measurement models. Results show the proposed approach outperforms the widely used ensemble Kalman and particle filters when small to moderate ensemble sizes are used.

We propose a landmark-constrained algorithm, LA-VDM (Landmark Accelerated Vector Diffusion Maps), to accelerate the Vector Diffusion Maps (VDM) framework built upon the Graph Connection Laplacian (GCL), which captures pairwise connection relationships within complex datasets. LA-VDM introduces a novel two-stage normalization that effectively address nonuniform sampling densities in both the data and the landmark sets. Under a manifold model with the frame bundle structure, we show that we can accurately recover the parallel transport with landmark-constrained diffusion from a point cloud, and hence asymptotically LA-VDM converges to the connection Laplacian. The performance and accuracy of LA-VDM are demonstrated through experiments on simulated datasets and an application to nonlocal image denoising.