Ensuring that the outputs of neural networks satisfy specific constraints is crucial for applying neural networks to real-life decision-making problems. In this paper, we consider making a batch of neural network outputs satisfy bounded and general linear constraints. We first reformulate the neural network output projection problem as an entropy-regularized linear programming problem. We show that such a problem can be equivalently transformed into an unconstrained convex optimization problem with Lipschitz continuous gradient according to the duality theorem. Then, based on an accelerated gradient descent algorithm with numerical performance enhancement, we present our architecture, GLinSAT, to solve the problem. To the best of our knowledge, this is the first general linear satisfiability layer in which all the operations are differentiable and matrix-factorization-free. Despite the fact that we can explicitly perform backpropagation based on automatic differentiation mechanism, we also provide an alternative approach in GLinSAT to calculate the derivatives based on implicit differentiation of the optimality condition. Experimental results on constrained traveling salesman problems, partial graph matching with outliers, predictive portfolio allocation and power system unit commitment demonstrate the advantages of GLinSAT over existing satisfiability layers. Our implementation is available at \url{https://github.com/HunterTracer/GLinSAT}.

Genetic screens mediated via CRISPR-Cas9 combined with high-content readouts have emerged as powerful tools for biological discovery. However, computational analyses of these screens come with additional challenges beyond those found with standard scRNA-seq analyses. For example, perturbation-induced variations of interest may be subtle and masked by other dominant source of variation shared with controls, and variable guide efficiency results in some cells not undergoing genetic perturbation despite expressing a guide RNA. While a number of methods have been developed to address the former problem by explicitly disentangling perturbation-induced variations from those shared with controls, less attention has been paid to the latter problem of noisy perturbation labels. To address this issue, here we propose ContrastiveVI+, a generative modeling framework that both disentangles perturbation-induced from non-perturbation-related variations while also inferring whether cells truly underwent genomic edits. Applied to three large-scale Perturb-seq datasets, we find that ContrastiveVI+ better recovers known perturbation-induced variations compared to previous methods while successfully identifying cells that escaped the functional consequences of guide RNA expression. An open-source implementation of our model is available at https://github.

Model stealing, where a learner tries to recover an unknown model via carefully chosen queries, is a critical problem in machine learning, as it threatens the security of proprietary models and the privacy of data they are trained on. In recent years, there has been particular interest in stealing large language models (LLMs). In this paper, we aim to build a theoretical understanding of stealing language models by studying a simple and mathematically tractable setting. We study model stealing for Hidden Markov Models (HMMs), and more generally low-rank language models. We assume that the learner works in the conditional query model, introduced by Kakade, Krishnamurthy, Mahajan and Zhang. Our main result is an efficient algorithm in the conditional query model, for learning any low-rank distribution. In other words, our algorithm succeeds at stealing any language model whose output distribution is low-rank. This improves upon the previous result by Kakade, Krishnamurthy, Mahajan and Zhang, which also requires the unknown distribution to have high "fidelity", a property that holds only in restricted cases. There are two key insights behind our algorithm: First, we represent the conditional distributions at each timestep by constructing barycentric spanners among a collection of vectors of exponentially large dimension. Second, for sampling from our representation, we iteratively solve a sequence of convex optimization problems that involve projection in relative entropy to prevent compounding of errors over the length of the sequence. This is an interesting example where, at least theoretically, allowing a machine learning model to solve more complex problems at inference time can lead to drastic improvements in its performance.

Machine learning is increasingly used to select which individuals receive limited-resource interventions in domains such as human services, education, development, and more. However, it is often not apparent what the right quantity is for models to predict. In particular, policymakers rarely have access to data from a randomized controlled trial (RCT) that would enable accurate estimates of treatment effects -- which individuals would benefit more from the intervention. Observational data is more likely to be available, creating a substantial risk of bias in treatment effect estimates. Practitioners instead commonly use a technique termed "risk-based targeting" where the model is just used to predict each individual's status quo outcome (an easier, non-causal task). Those with higher predicted risk are offered treatment. There is currently almost no empirical evidence to inform which choices lead to the most effect machine learning-informed targeting strategies in social domains. In this work, we use data from 5 real-world RCTs in a variety of domains to empirically assess such choices. We find that risk-based targeting is almost always inferior to targeting based on even biased estimates of treatment effects. Moreover, these results hold even when the policymaker has strong normative preferences for assisting higher-risk individuals. Our results imply that, despite the widespread use of risk prediction models in applied settings, practitioners may be better off incorporating even weak evidence about heterogeneous causal effects to inform targeting.

Gaussian Processes face two primary challenges: constructing models for large datasets and selecting the optimal model. This master's thesis tackles these challenges in the low-dimensional case. We examine recent convergence results to identify models with optimal convergence rates and pinpoint essential parameters. Utilizing this model, we propose a Samplet-based approach to efficiently construct and train the Gaussian Processes, reducing the cubic computational complexity to a log-linear scale. This method facilitates optimal regression while maintaining efficient performance.

We present BALDUR, a novel Bayesian algorithm designed to deal with multi-modal datasets and small sample sizes in high-dimensional settings while providing explainable solutions. To do so, the proposed model combines within a common latent space the different data views to extract the relevant information to solve the classification task and prune out the irrelevant/redundant features/data views. Furthermore, to provide generalizable solutions in small sample size scenarios, BALDUR efficiently integrates dual kernels over the views with a small sample-to-feature ratio. Finally, its linear nature ensures the explainability of the model outcomes, allowing its use for biomarker identification. This model was tested over two different neurodegeneration datasets, outperforming the state-of-the-art models and detecting features aligned with markers already described in the scientific literature.

Recent rapid advancements of machine learning have greatly enhanced the accuracy of prediction models, but most models remain "black boxes", making prediction error diagnosis challenging, especially with outliers. This lack of transparency hinders trust and reliability in industrial applications. Heuristic attribution methods, while helpful, often fail to capture true causal relationships, leading to inaccurate error attributions. Various root-cause analysis methods have been developed using Shapley values, yet they typically require predefined causal graphs, limiting their applicability for prediction errors in machine learning models. To address these limitations, we introduce the Causal-Discovery-based Root-Cause Analysis (CD-RCA) method that estimates causal relationships between the prediction error and the explanatory variables, without needing a pre-defined causal graph. By simulating synthetic error data, CD-RCA can identify variable contributions to outliers in prediction errors by Shapley values. Extensive simulations show CD-RCA outperforms current heuristic attribution methods, and a sensitivity analysis reveals new patterns where Shapley values may misattribute errors, paving the way for more accurate error attribution methods.

Federated Learning (FL) employs a training approach to address scenarios where users' data cannot be shared across clients. Achieving fairness in FL is imperative since training data in FL is inherently geographically distributed among diverse user groups. Existing research on fairness predominantly assumes access to the entire training data, making direct transfer to FL challenging. However, the limited existing research on fairness in FL does not effectively address two key challenges, i.e., (CH1) Current methods fail to deal with the inconsistency between fair optimization results obtained with surrogate functions and fair classification results. (CH2) Directly aggregating local fair models does not always yield a globally fair model due to non Identical and Independent data Distributions (non-IID) among clients. To address these challenges, we propose a Wasserstein Fair Federated Learning framework, namely WassFFed. To tackle CH1, we ensure that the outputs of local models, rather than the loss calculated with surrogate functions or classification results with a threshold, remain independent of various user groups. To resolve CH2, we employ a Wasserstein barycenter calculation of all local models' outputs for each user group, bringing local model outputs closer to the global output distribution to ensure consistency between the global model and local models. We conduct extensive experiments on three real-world datasets, demonstrating that WassFFed outperforms existing approaches in striking a balance between accuracy and fairness.

Breast cancer detection is still an open research field, despite a tremendous effort devoted to work in this area. Effect size is a statistical concept that measures the strength of the relationship between two variables on a numeric scale. Feature selection is widely used to reduce the dimensionality of data by selecting only a subset of predictor variables to improve a learning model. In this work, an algorithm and experimental results demonstrate the feasibility of developing a statistical featureselector-based learning tool capable of reducing the data dimensionality using parametric effect size measures from features extracted from cell nuclei images. The SVM classifier with a linear kernel as a learning tool achieved an accuracy of over 90%. These excellent results suggest that the effect size is within the standards of the feature-selector methods. Keywords: Effect Size Cohen's d Standardized Mean Difference Feature selection Breast Cancer

We consider the approximation of functions by 2-layer neural networks with a small number of hidden weights based on the squared loss and small datasets. Due to the highly non-convex energy landscape, gradient-based training often suffers from local minima. As a remedy, we initialize the hidden weights with samples from a learned proposal distribution, which we parameterize as a deep generative model. To train this model, we exploit the fact that with fixed hidden weights, the optimal output weights solve a linear equation. After learning the generative model, we refine the sampled weights with a gradient-based post-processing in the latent space. Here, we also include a regularization scheme to counteract potential noise. Finally, we demonstrate the effectiveness of our approach by numerical examples.