We present a new high-probability PAC-Bayes oracle bound for unbounded losses. This result can be understood as a PAC-Bayes version of the Chernoff bound. The proof technique relies on uniformly bounding the tail of certain random variable based on the Cram\'er transform of the loss. We highlight two applications of our main result. First, we show that our bound solves the open problem of optimizing the free parameter on many PAC-Bayes bounds. Finally, we show that our approach allows working with flexible assumptions on the loss function, resulting in novel bounds that generalize previous ones and can be minimized to obtain Gibbs-like posteriors.

The current approach to fetal anomaly screening is based on biometric measurements derived from individually selected ultrasound images. In this paper, we introduce a paradigm shift that attains human-level performance in biometric measurement by aggregating automatically extracted biometrics from every frame across an entire scan, with no need for operator intervention. We use a convolutional neural network to classify each frame of an ultrasound video recording. We then measure fetal biometrics in every frame where appropriate anatomy is visible. We use a Bayesian method to estimate the true value of each biometric from a large number of measurements and probabilistically reject outliers. We performed a retrospective experiment on 1457 recordings (comprising 48 million frames) of 20-week ultrasound scans, estimated fetal biometrics in those scans and compared our estimates to the measurements sonographers took during the scan. Our method achieves human-level performance in estimating fetal biometrics and estimates well-calibrated credible intervals in which the true biometric value is expected to lie.

AI alignment aims to make AI systems behave in line with human intentions and values. As AI systems grow more capable, so do risks from misalignment. To provide a comprehensive and up-to-date overview of the alignment field, in this survey, we delve into the core concepts, methodology, and practice of alignment. First, we identify four principles as the key objectives of AI alignment: Robustness, Interpretability, Controllability, and Ethicality (RICE). Guided by these four principles, we outline the landscape of current alignment research and decompose them into two key components: forward alignment and backward alignment. The former aims to make AI systems aligned via alignment training, while the latter aims to gain evidence about the systems' alignment and govern them appropriately to avoid exacerbating misalignment risks. On forward alignment, we discuss techniques for learning from feedback and learning under distribution shift. On backward alignment, we discuss assurance techniques and governance practices. We also release and continually update the website (www.alignmentsurvey.com) which features tutorials, collections of papers, blog posts, and other resources.

We present a Transfer Causal Learning (TCL) framework when target and source domains share the same covariate/feature spaces, aiming to improve causal effect estimation accuracy in limited data. Limited data is very common in medical applications, where some rare medical conditions, such as sepsis, are of interest. Our proposed method, named \texttt{$\ell_1$-TCL}, incorporates $\ell_1$ regularized TL for nuisance models (e.g., propensity score model); the TL estimator of the nuisance parameters is plugged into downstream average causal/treatment effect estimators (e.g., inverse probability weighted estimator). We establish non-asymptotic recovery guarantees for the \texttt{$\ell_1$-TCL} with generalized linear model (GLM) under the sparsity assumption in the high-dimensional setting, and demonstrate the empirical benefits of \texttt{$\ell_1$-TCL} through extensive numerical simulation for GLM and recent neural network nuisance models. Our method is subsequently extended to real data and generates meaningful insights consistent with medical literature, a case where all baseline methods fail.

How do language models "think"? This paper formulates a probabilistic cognitive model called the bounded pragmatic speaker, which can characterize the operation of different variations of language models. Specifically, we demonstrate that large language models fine-tuned with reinforcement learning from human feedback (Ouyang et al., 2022) embody a model of thought that conceptually resembles a fast-and-slow model (Kahneman, 2011), which psychologists have attributed to humans. We discuss the limitations of reinforcement learning from human feedback as a fast-and-slow model of thought and propose avenues for expanding this framework. In essence, our research highlights the value of adopting a cognitive probabilistic modeling approach to gain insights into the comprehension, evaluation, and advancement of language models.

Decision-focused (DF) model-based reinforcement learning has recently been introduced as a powerful algorithm that can focus on learning the MDP dynamics that are most relevant for obtaining high returns. While this approach increases the agent's performance by directly optimizing the reward, it does so by learning less accurate dynamics from a maximum likelihood perspective. We demonstrate that when the reward function is defined by preferences over multiple objectives, the DF model may be sensitive to changes in the objective preferences.In this work, we develop the robust decision-focused (RDF) algorithm, which leverages the non-identifiability of DF solutions to learn models that maximize expected returns while simultaneously learning models that transfer to changes in the preference over multiple objectives. We demonstrate the effectiveness of RDF on two synthetic domains and two healthcare simulators, showing that it significantly improves the robustness of DF model learning to changes in the reward function without compromising training-time return.

Bayesian neural networks (BNNs) are a principled approach to modeling predictive uncertainties in deep learning, which are important in safety-critical applications. Since exact Bayesian inference over the weights in a BNN is intractable, various approximate inference methods exist, among which sampling methods such as Hamiltonian Monte Carlo (HMC) are often considered the gold standard. While HMC provides high-quality samples, it lacks interpretable summary statistics because its sample mean and variance is meaningless in neural networks due to permutation symmetry. In this paper, we first show that the role of permutations can be meaningfully quantified by a number of transpositions metric. We then show that the recently proposed rebasin method [1] allows us to summarize HMC samples into a compact representation that provides a meaningful explicit uncertainty estimate for each weight in a neural network, thus unifying sampling methods with variational inference. We show that this compact representation allows us to compare trained BNNs directly in weight space across sampling methods and variational inference, and to efficiently prune neural networks trained without explicit Bayesian frameworks by exploiting uncertainty estimates from HMC.

Uncertainty quantification is a critical aspect of machine learning models, providing important insights into the reliability of predictions and aiding the decision-making process in real-world applications. This paper proposes a novel way to use variance-based measures to quantify uncertainty on the basis of second-order distributions in classification problems. A distinctive feature of the measures is the ability to reason about uncertainties on a class-based level, which is useful in situations where nuanced decision-making is required. Recalling some properties from the literature, we highlight that the variance-based measures satisfy important (axiomatic) properties. In addition to this axiomatic approach, we present empirical results showing the measures to be effective and competitive to commonly used entropy-based measures.

Many modern statistical analysis and machine learning applications require training models on sensitive user data. Differential privacy provides a formal guarantee that individual-level information about users does not leak. In this framework, randomized algorithms inject calibrated noise into the confidential data, resulting in privacy-protected datasets or queries. However, restricting access to only privatized data during statistical analysis makes it computationally challenging to make valid inferences on the parameters underlying the confidential data. In this work, we propose simulation-based inference methods from privacy-protected datasets. In addition to sequential Monte Carlo approximate Bayesian computation, we use neural conditional density estimators as a flexible family of distributions to approximate the posterior distribution of model parameters given the observed private query results. We illustrate our methods on discrete time-series data under an infectious disease model and with ordinary linear regression models. Illustrating the privacy-utility trade-off, our experiments and analysis demonstrate the necessity and feasibility of designing valid statistical inference procedures to correct for biases introduced by the privacy-protection mechanisms.

We propose the Plaid Atoms Model (PAM), a novel Bayesian nonparametric model for grouped data. Founded on an idea of `atom skipping', PAM is part of a well-established category of models that generate dependent random distributions and clusters across multiple groups. Atom skipping referrs to stochastically assigning 0 weights to atoms in an infinite mixture. Deploying atom skipping across groups, PAM produces a dependent clustering pattern with overlapping and non-overlapping clusters across groups. As a result, interpretable posterior inference is possible such as reporting the posterior probability of a cluster being exclusive to a single group or shared among a subset of groups. We discuss the theoretical properties of the proposed and related models. Minor extensions of the proposed model for multivariate or count data are presented. Simulation studies and applications using real-world datasets illustrate the performance of the new models with comparison to existing models.