Quantifying predictive uncertainty is essential for safe and trustworthy real-world AI deployment. Yet, fully nonparametric estimation of conditional distributions remains challenging for multivariate targets. We propose Tomographic Quantile Forests (TQF), a nonparametric, uncertainty-aware, tree-based regression model for multivariate targets. TQF learns conditional quantiles of directional projections $\mathbf{n}^{\top}\mathbf{y}$ as functions of the input $\mathbf{x}$ and the unit direction $\mathbf{n}$. At inference, it aggregates quantiles across many directions and reconstructs the multivariate conditional distribution by minimizing the sliced Wasserstein distance via an efficient alternating scheme with convex subproblems. Unlike classical directional-quantile approaches that typically produce only convex quantile regions and require training separate models for different directions, TQF covers all directions with a single model without imposing convexity restrictions. We evaluate TQF on synthetic and real-world datasets, and release the source code on GitHub.

We address network structure learning from zero-inflated count data by casting each node as a zero-inflated generalized linear model and optimizing a smooth, score-based objective under a directed acyclic graph constraint. Our Zero-Inflated Continuous Optimization (ZICO) approach uses node-wise likelihoods with canonical links and enforces acyclicity through a differentiable surrogate constraint combined with sparsity regularization. ZICO achieves superior performance with faster runtimes on simulated data. It also performs comparably to or better than common algorithms for reverse engineering gene regulatory networks. ZICO is fully vectorized and mini-batched, enabling learning on larger variable sets with practical runtimes in a wide range of domains.

This paper addresses the challenging computational problem of estimating intractable expectations over discrete domains. Existing approaches, including Monte Carlo and Russian Roulette estimators, are consistent but often require a large number of samples to achieve accurate results. We propose a novel estimator, \emph{BayesSum}, which is an extension of Bayesian quadrature to discrete domains. It is more sample efficient than alternatives due to its ability to make use of prior information about the integrand through a Gaussian process. We show this through theory, deriving a convergence rate significantly faster than Monte Carlo in a broad range of settings. We also demonstrate empirically that our proposed method does indeed require fewer samples on several synthetic settings as well as for parameter estimation for Conway-Maxwell-Poisson and Potts models.

Streamflow forecasting is crucial for water resource management and risk mitigation. While deep learning models have achieved strong predictive performance, they often overlook underlying physical processes, limiting interpretability and generalization. Recent causal learning approaches address these issues by integrating domain knowledge, yet they typically rely on fixed causal graphs that fail to adapt to data. We propose CauStream, a unified framework for causal spatiotemporal streamflow forecasting. CauSTream jointly learns (i) a runoff causal graph among meteorological forcings and (ii) a routing graph capturing dynamic dependencies across stations. We further establish identifiability conditions for these causal structures under a nonparametric setting. We evaluate CauSTream on three major U.S. river basins across three forecasting horizons. The model consistently outperforms prior state-of-the-art methods, with performance gaps widening at longer forecast windows, indicating stronger generalization to unseen conditions. Beyond forecasting, CauSTream also learns causal graphs that capture relationships among hydrological factors and stations. The inferred structures align closely with established domain knowledge, offering interpretable insights into watershed dynamics. CauSTream offers a principled foundation for causal spatiotemporal modeling, with the potential to extend to a wide range of scientific and environmental applications.

It is widely believed that complex machine learning models generally encode features through linear representations, but these features exist in superposition, making them challenging to recover. We study the following fundamental setting for learning features in superposition from black-box query access: we are given query access to a function \[ f(x)=\sum_{i=1}^n a_i\,σ_i(v_i^\top x), \] where each unit vector $v_i$ encodes a feature direction and $σ_i:\mathbb{R} \rightarrow \mathbb{R}$ is an arbitrary response function and our goal is to recover the $v_i$ and the function $f$. In learning-theoretic terms, superposition refers to the overcomplete regime, when the number of features is larger than the underlying dimension (i.e. $n > d$), which has proven especially challenging for typical algorithmic approaches. Our main result is an efficient query algorithm that, from noisy oracle access to $f$, identifies all feature directions whose responses are non-degenerate and reconstructs the function $f$. Crucially, our algorithm works in a significantly more general setting than all related prior results -- we allow for essentially arbitrary superpositions, only requiring that $v_i, v_j$ are not nearly identical for $i \neq j$, and general response functions $σ_i$. At a high level, our algorithm introduces an approach for searching in Fourier space by iteratively refining the search space to locate the hidden directions $v_i$.

One strategy to scale up ML-driven science is to increase wet lab experiments' information density. We present a method based on a neural extension of compressed sensing to function space. We measure the activity of multiple different molecules simultaneously, rather than individually. Then, we deconvolute the molecule-activity map during model training. Co-design of wet lab experiments and learning algorithms provably leads to orders-of-magnitude gains in information density. We demonstrate on antibodies and cell therapies.

The double machine learning (DML) method combines the predictive power of machine learning with statistical estimation to conduct inference about the structural parameter of interest. This paper presents the R package `xtdml`, which implements DML methods for partially linear panel regression models with low-dimensional fixed effects, high-dimensional confounding variables, proposed by Clarke and Polselli (2025). The package provides functionalities to: (a) learn nuisance functions with machine learning algorithms from the `mlr3` ecosystem, (b) handle unobserved individual heterogeneity choosing among first-difference transformation, within-group transformation, and correlated random effects, (c) transform the covariates with min-max normalization and polynomial expansion to improve learning performance. We showcase the use of `xtdml` with both simulated and real longitudinal data.

Dimension reduction methods are a fundamental class of techniques in data analysis, which aim to find a lower-dimensional representation of higher-dimensional data while preserving as much of the original information as possible. These methods are extensively used in practice, including in exploratory data analyses to visualize data--arguably, one of the first and most vital steps in any data analysis (Ray et al., 2021). Notably, in genomics, dimension reduction methods are ubiquitously applied to visualize high-dimensional single-cell RNA sequencing data in two dimensions (Becht et al., 2019). Beyond visualization, dimension reduction methods are also frequently employed to mitigate the curse of dimensionality (Bellman, 1957), engineer new features to improve downstream tasks like prediction (e.g., Massy, 1965), and enable scientific discovery in unsupervised learning settings (Chang et al., 2025). For example, many researchers have used dimension reduction in conjunction with clustering to discover new cell types and cell states (Wu et al., 2021), new cancer subtypes (Northcott et al., 2017), and other substantively-meaningful structure in a variety of domains (Bergen et al., 2019; Traven et al., 2017). Given the widespread use and need for dimension reduction methods, numerous dimension reduction techniques have been developed. Popular techniques include but are not limited to principal component analysis (PCA) (Pearson, 1901; Hotelling, 1933), multidimensional scaling (MDS) (Torgerson, 1952; Kruskal, 1964a), Isomap (Tenenbaum et al., 2000), locally linear embedding (LLE) (Roweis and Saul, 2000), t-distributed stochastic neighbor embedding (t-SNE) (van der 1

Partial Differential Equations (PDEs) are central to modeling complex systems across physical, biological, and engineering domains, yet traditional numerical methods often struggle with high-dimensional or complex problems. Physics-Informed Neural Networks (PINNs) have emerged as an efficient alternative by embedding physics-based constraints into deep learning frameworks, but they face challenges in achieving high accuracy and handling complex boundary conditions. In this work, we extend the Time-Evolving Natural Gradient (TENG) framework to address Dirichlet boundary conditions, integrating natural gradient optimization with numerical time-stepping schemes, including Euler and Heun methods, to ensure both stability and accuracy. By incorporating boundary condition penalty terms into the loss function, the proposed approach enables precise enforcement of Dirichlet constraints. Experiments on the heat equation demonstrate the superior accuracy of the Heun method due to its second-order corrections and the computational efficiency of the Euler method for simpler scenarios. This work establishes a foundation for extending the framework to Neumann and mixed boundary conditions, as well as broader classes of PDEs, advancing the applicability of neural network-based solvers for real-world problems.

Data is a critical asset for training large language models (LLMs), alongside compute resources and skilled workers. While some training data is publicly available, substantial investment is required to generate proprietary datasets, such as human preference annotations or to curate new ones from existing sources. As larger datasets generally yield better model performance, two natural questions arise. First, how can data owners make informed decisions about curation strategies and data sources investment? Second, how can multiple data owners collaboratively pool their resources to train superior models while fairly distributing the benefits? This problem, data valuation, which is not specific to large language models, has been addressed by the machine learning community through the lens of cooperative game theory, with the Shapley value being the prevalent solution concept. However, computing Shapley values is notoriously expensive for data valuation, typically requiring numerous model retrainings, which can become prohibitive for large machine learning models. In this work, we demonstrate that this computational challenge is dramatically simplified for LLMs trained with Direct Preference Optimization (DPO). We show how the specific mathematical structure of DPO enables scalable Shapley value computation. We believe this observation unlocks many applications at the intersection of data valuation and large language models.