Neural Information Processing Systems http://nips.cc/

Inthis paper,werely onabroader viewofpropercompositelosses and arecent construct from information geometry,sourcefunctions, whose fitting alleviates constraints faced by canonical links.

A decision-theoretic characterization of perfect calibration is that an agent seeking to minimize a proper loss in expectation cannot improve their outcome by post-processing a perfectly calibrated predictor. Hu and Wu (FOCS'24) use this to define an approximate calibration measure called calibration decision loss ($\mathsf{CDL}$), which measures the maximal improvement achievable by any post-processing over any proper loss. Unfortunately, $\mathsf{CDL}$ turns out to be intractable to even weakly approximate in the offline setting, given black-box access to the predictions and labels. We suggest circumventing this by restricting attention to structured families of post-processing functions $K$. We define the calibration decision loss relative to $K$, denoted $\mathsf{CDL}_K$ where we consider all proper losses but restrict post-processings to a structured family $K$. We develop a comprehensive theory of when $\mathsf{CDL}_K$ is information-theoretically and computationally tractable, and use it to prove both upper and lower bounds for natural classes $K$. In addition to introducing new definitions and algorithmic techniques to the theory of calibration for decision making, our results give rigorous guarantees for some widely used recalibration procedures in machine learning.

We consider the problem of constructing probabilistic predictions that lead to accurate decisions when employed by downstream users to inform actions. For a single decision maker, designing an optimal predictor is equivalent to minimizing a proper loss function corresponding to the negative utility of that individual. For multiple decision makers, our problem can be viewed as a variant of omniprediction in which the goal is to design a single predictor that simultaneously minimizes multiple losses. Existing algorithms for achieving omniprediction broadly fall into two categories: 1) boosting methods that optimize other auxiliary targets such as multicalibration and obtain omniprediction as a corollary, and 2) adversarial two-player game based approaches that estimate and respond to the ``worst-case" loss in an online fashion. We give lower bounds demonstrating that multicalibration is a strictly more difficult problem than omniprediction and thus the former approach must incur suboptimal sample complexity. For the latter approach, we discuss how these ideas can be used to obtain a sample-efficient algorithm through an online-to-batch conversion. This conversion has the downside of returning a complex, randomized predictor. We improve on this method by designing a more direct, unrandomized algorithm that exploits structural elements of the set of proper losses.

Calibrated predictions can be reliably interpreted as probabilities. An important step towards achieving better calibration is to design an appropriate calibration measure to meaningfully assess the miscalibration level of a predictor. A recent line of work initiated by Haghtalab et al. [2024] studies the design of truthful calibration measures: a truthful measure is minimized when a predictor outputs the true probabilities, whereas a non-truthful measure incentivizes the predictor to lie so as to appear more calibrated. All previous calibration measures were non-truthful until Hartline et al. [2025] introduced the first perfectly truthful calibration measures for binary prediction tasks in the batch setting. We introduce a perfectly truthful calibration measure for multi-class prediction tasks, generalizing the work of Hartline et al. [2025] beyond binary prediction. We study common methods of extending calibration measures from binary to multi-class prediction and identify ones that do or do not preserve truthfulness. In addition to truthfulness, we mathematically prove and empirically verify that our calibration measure exhibits superior robustness: it robustly preserves the ordering between dominant and dominated predictors, regardless of the choice of hyperparameters (bin sizes). This result addresses the non-robustness issue of binned ECE, which has been observed repeatedly in prior work.

Learning from label proportions (LLP) is a weakly supervised classification problem where data points are grouped into bags, and the label proportions within each bag are observed instead of the instance-level labels. The task is to learn a classifier to predict the labels of future individual instances.

In this paper, we rely on a broader view of proper composite losses and a recent construct from information geometry, source functions, whose fitting alleviates constraints faced by canonical links.

The connection is established for all of Savage's DPO framework to generalize its functional parts (Alfano et al., 2025; Azar et al., 2024; Chen et al., The latter involves elements from Doignon-Falmagne's stochastic choice These many design elements lead to a generalization making the most of the connection since we encompass all of properness on Savage's side (regardless of optional properties like symmetry, We also encompass all of the modelling's power on Krantz, Luce, Suppes and Notably, our level of generalization is able to support "for free" important This is an important task because DPO was designed with the objective to simplify RLHF and getting "above" DPO is mandatory to improve results by getting more freedom on reward shapes, trajectories and preference behaviours (Gupta et al., 2025), all of which needs to be done while One perhaps unexpected pitfall comes from the RLHF/DPO inherited "gold To preserve readability, all proofs are given in an appendix. We adopt many definitions from Rafailov et al. (2023).