Sampling and automatic differentiation are both ubiquitous in modern machine learning. At its intersection, differentiating through a sampling operation, with respect to the parameters of the sampling process, is a problem that is both challenging and broadly applicable. We introduce a general framework and a new algorithm for first-order optimization of parameterized stochastic diffusions, performing jointly, in a single loop, optimization and sampling steps. This approach is inspired by recent advances in bilevel optimization and automatic implicit differentiation, leveraging the point of view of sampling as optimization over the space of probability distributions. We provide theoretical and experimental results showcasing the performance of our method.

Aligning language models with human preferences is crucial for reducing errors and biases in these models. Alignment techniques, such as reinforcement learning from human feedback (RLHF), are typically cast as optimizing a tradeoff between human preference rewards and a proximity regularization term that encourages staying close to the unaligned model. Selecting an appropriate level of regularization is critical: insufficient regularization can lead to reduced model capabilities due to reward hacking, whereas excessive regularization hinders alignment. Traditional methods for finding the optimal regularization level require retraining multiple models with varying regularization strengths. This process, however, is resource-intensive, especially for large models. To address this challenge, we propose decoding-time realignment (DeRa), a simple method to explore and evaluate different regularization strengths in aligned models without retraining. DeRa enables control over the degree of alignment, allowing users to smoothly transition between unaligned and aligned models. It also enhances the efficiency of hyperparameter tuning by enabling the identification of effective regularization strengths using a validation dataset.

We introduce a differentiable clustering method based on stochastic perturbations of minimum-weight spanning forests. This allows us to include clustering in end-to-end trainable pipelines, with efficient gradients. We show that our method performs well even in difficult settings, such as data sets with high noise and challenging geometries. We also formulate an ad hoc loss to efficiently learn from partial clustering data using this operation. We demonstrate its performance on several data sets for supervised and semi-supervised tasks.

Automatic differentiation (autodiff) has revolutionized machine learning. Itallows to express complex computations by composing elementary ones in creativeways and removes the burden of computing their derivatives by hand. Morerecently, differentiation of optimization problem solutions has attractedwidespread attention with applications such as optimization layers, and inbi-level problems such as hyper-parameter optimization and meta-learning.However, so far, implicit differentiation remained difficult to use forpractitioners, as it often required case-by-case tedious mathematicalderivations and implementations. In this paper, we proposeautomatic implicit differentiation, an efficientand modular approach for implicit differentiation of optimization problems. Inour approach, the user defines directly in Python a function $F$ capturing theoptimality conditions of the problem to be differentiated. Once this is done, weleverage autodiff of $F$ and the implicit function theorem to automaticallydifferentiate the optimization problem. Our approach thus combines the benefitsof implicit differentiation and autodiff. It is efficient as it can be added ontop of any state-of-the-art solver and modular as the optimality conditionspecification is decoupled from the implicit differentiation mechanism. We showthat seemingly simple principles allow to recover many existing implicitdifferentiation methods and create new ones easily. We demonstrate the ease offormulating and solving bi-level optimization problems using our framework. Wealso showcase an application to the sensitivity analysis of molecular dynamics.

Energy-based models, a.k.a. energy networks, perform inference by optimizing an energy function, typically parametrized by a neural network. This allows one to capture potentially complex relationships between inputs andoutputs.To learn the parameters of the energy function, the solution to thatoptimization problem is typically fed into a loss function.The key challenge for training energy networks lies in computing loss gradients,as this typically requires argmin/argmax differentiation.In this paper, building upon a generalized notion of conjugate function,which replaces the usual bilinear pairing with a general energy function,we propose generalized Fenchel-Young losses, a natural loss construction forlearning energy networks. Our losses enjoy many desirable properties and theirgradients can be computed efficiently without argmin/argmax differentiation.We also prove the calibration of their excess risk in the case of linear-concaveenergies. We demonstrate our losses on multilabel classification and imitation learning tasks.

Most graph kernels are an instance of the class of R-Convolution kernels, which measure the similarity of objects by comparing their substructures. Despite their empirical success, most graph kernels use a naive aggregation of the final set of substructures, usually a sum or average, thereby potentially discarding valuable information about the distribution of individual components. Furthermore, only a limited instance of these approaches can be extended to continuously attributed graphs. We propose a novel method that relies on the Wasserstein distance between the node feature vector distributions of two graphs, which allows to find subtler differences in data sets by considering graphs as high-dimensional objects, rather than simple means. We further propose a Weisfeiler--Lehman inspired embedding scheme for graphs with continuous node attributes and weighted edges, enhance it with the computed Wasserstein distance, and thus improve the state-of-the-art prediction performance on several graph classification tasks.

SUMMARY: Combinatorial association mapping aims to assess the statistical association of higher-order interactions of genetic markers with a phenotype of interest. This article presents combinatorial association mapping (CASMAP), a software package that leverages recent advances in significant pattern mining to overcome the statistical and computational challenges that have hindered combinatorial association mapping. CASMAP can be used to perform region-based association studies and to detect higher-order epistatic interactions of genetic variants. Most importantly, unlike other existing significant pattern mining-based tools, CASMAP allows for the correction of categorical covariates such as age or gender, making it suitable for genome-wide association studies. AVAILABILITY AND IMPLEMENTATION: The R and Python packages can be downloaded from our GitHub repository http://github.com/BorgwardtLab/CASMAP. The R package is also available on CRAN. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Summary: Measuring the similarity of graphs is a fundamental step in the analysis of graph-structured data, which is omnipresent in computational biology. Graph kernels have been proposed as a powerful and efficient approach to this problem of graph comparison. Here we provide graphkernels, the first R and Python graph kernel libraries including baseline kernels such as label histogram based kernels, classic graph kernels such as random walk based kernels, and the state-of-the-art Weisfeiler-Lehman graph kernel. The core of all graph kernels is implemented in C ++ for efficiency. Using the kernel matrices computed by the package, we can easily perform tasks such as classification, regression and clustering on graph-structured samples. Availability and implementation: The R and Python packages including source code are available at https://CRAN.R-project.org/package=graphkernels and https://pypi.python.org/pypi/graphkernels. Contact: [email protected] or [email protected]. Supplementary information: Supplementary data are available online at Bioinformatics.

MOTIVATION: Genetic heterogeneity is the phenomenon that distinct genetic variants may give rise to the same phenotype. The recently introduced algorithm Fast Automatic Interval Search ( FAIS ) enables the genome-wide search of candidate regions for genetic heterogeneity in the form of any contiguous sequence of variants, and achieves high computational efficiency and statistical power. Although FAIS can test all possible genomic regions for association with a phenotype, a key limitation is its inability to correct for confounders such as gender or population structure, which may lead to numerous false-positive associations. RESULTS: We propose FastCMH , a method that overcomes this problem by properly accounting for categorical confounders, while still retaining statistical power and computational efficiency. Experiments comparing FastCMH with FAIS and multiple kinds of burden tests on simulated data, as well as on human and Arabidopsis samples, demonstrate that FastCMH can drastically reduce genomic inflation and discover associations that are missed by standard burden tests. AVAILABILITY AND IMPLEMENTATION: An R package fastcmh is available on CRAN and the source code can be found at: https://www.bsse.ethz.ch/mlcb/research/bioinformatics-and-computational-biology/fastcmh.html. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

In high-dimensional settings, where the number of features p is typically much larger than the number of samples n, methods which can systematically examine arbitrary combinations of features, a huge 2^p-dimensional space, have recently begun to be explored. However, none of the current methods is able to assess the association between feature combinations and a target variable while conditioning on a categorical covariate, in order to correct for potential confounding effects. We propose the Fast Automatic Conditional Search (FACS) algorithm, a significant discriminative itemset mining method which conditions on categorical covariates and only scales as O(k log k), where k is the number of states of the categorical covariate. Based on the Cochran-Mantel-Haenszel Test, FACS demonstrates superior speed and statistical power on simulated and real-world datasets compared to the state of the art, opening the door to numerous applications in biomedicine.

We present a novel algorithm for significant pattern mining, Westfall-Young light. The target patterns are statistically significantly enriched in one of two classes of objects. Our method corrects for multiple hypothesis testing and correlations between patterns via the Westfall-Young permutation procedure, which empirically estimates the null distribution of pattern frequencies in each class via permutations.

The problem of finding itemsets that are statistically significantly enriched in a class of transactions is complicated by the need to correct for multiple hypothesis testing. Pruning untestable hypotheses was recently proposed as a strategy for this task of significant itemset mining. It was shown to lead to greater statistical power, the discovery of more truly significant itemsets, than the standard Bonferroni correction on real-world datasets. An open question, however, is whether this strategy of excluding untestable hypotheses also leads to greater statistical power in subgraph mining, in which the number of hypotheses is much larger than in itemset mining. Here we answer this question by an empirical investigation on eight popular graph benchmark datasets. We propose a new efficient search strategy, which always returns the same solution as the state-of-the-art approach and is approximately two orders of magnitude faster. Moreover, we exploit the dependence between subgraphs by considering the effective number of tests and thereby further increase the statistical power.

MOTIVATION: Genetic heterogeneity, the fact that several sequence variants give rise to the same phenotype, is a phenomenon that is of the utmost interest in the analysis of complex phenotypes. Current approaches for finding regions in the genome that exhibit genetic heterogeneity suffer from at least one of two shortcomings: (i) they require the definition of an exact interval in the genome that is to be tested for genetic heterogeneity, potentially missing intervals of high relevance, or (ii) they suffer from an enormous multiple hypothesis testing problem due to the large number of potential candidate intervals being tested, which results in either many false positives or a lack of power to detect true intervals. RESULTS: Here, we present an approach that overcomes both problems: it allows one to automatically find all contiguous sequences of single nucleotide polymorphisms in the genome that are jointly associated with the phenotype. It also solves both the inherent computational efficiency problem and the statistical problem of multiple hypothesis testing, which are both caused by the huge number of candidate intervals. We demonstrate on Arabidopsis thaliana genome-wide association study data that our approach can discover regions that exhibit genetic heterogeneity and would be missed by single-locus mapping. CONCLUSIONS: Our novel approach can contribute to the genome-wide discovery of intervals that are involved in the genetic heterogeneity underlying complex phenotypes. AVAILABILITY AND IMPLEMENTATION: The code can be obtained at: http://www.bsse.ethz.ch/mlcb/research/bioinformatics-and-computational-biology/sis.html.