Modern machine learning (ML) training workloads place substantial demands on both computational and communication resources. Consequently, accurate performance estimation has become increasingly critical for guiding system design decisions, such as the selection of parallelization strategies, cluster configurations, and hardware provisioning. Existing simulation-based performance estimation requires reimplementing the ML framework in a simulator, which demands significant manual effort and is hard to maintain as ML frameworks evolve rapidly. This paper introduces Phantora, a hybrid GPU cluster simulator designed for performance estimation of ML training workloads. Phantora executes unmodified ML frameworks as is within a distributed, containerized environment. Each container emulates the behavior of a GPU server in a large-scale cluster, while Phantora intercepts and simulates GPU- and communication-related operations to provide high-fidelity performance estimation. We call this approach hybrid simulation of ML systems, in contrast to traditional methods that simulate static workloads. The primary advantage of hybrid simulation is that it allows direct reuse of ML framework source code in simulation, avoiding the need for reimplementation. Our evaluation shows that Phantora provides accuracy comparable to static workload simulation while supporting three state-of-the-art LLM training frameworks out-of-the-box. In addition, Phantora operates on a single GPU, eliminating the need for the resource-intensive trace collection and workload extraction steps required by traditional trace-based simulators. Phantora is open-sourced at https://github.com/QDelta/Phantora.

Performance monitoring is essential for safe clinical deployment of image classification models. However, because ground-truth labels are typically unavailable in the target dataset, direct assessment of real-world model performance is infeasible. State-of-the-art performance estimation methods address this by leveraging confidence scores to estimate the target accuracy. Despite being a promising direction, the established methods mainly estimate the model's accuracy and are rarely evaluated in a clinical domain, where strong class imbalances and dataset shifts are common. Our contributions are twofold: First, we introduce generalisations of existing performance prediction methods that directly estimate the full confusion matrix. Then, we benchmark their performance on chest x-ray data in real-world distribution shifts as well as simulated covariate and prevalence shifts. The proposed confusion matrix estimation methods reliably predicted clinically relevant counting metrics on medical images under distribution shifts. However, our simulated shift scenarios exposed important failure modes of current performance estimation techniques, calling for a better understanding of real-world deployment contexts when implementing these performance monitoring techniques for postmarket surveillance of medical AI models.

We study the problem of monitoring machine learning models under gradual distribution shifts, where circumstances change slowly over time, often leading to unnoticed yet significant declines in accuracy. To address this, we propose Incremental Uncertainty-aware Performance Monitoring (IUPM), a novel label-free method that estimates performance changes by modeling gradual shifts using optimal transport. In addition, IUPM quantifies the uncertainty in the performance prediction and introduces an active labeling procedure to restore a reliable estimate under a limited labeling budget. Our experiments show that IUPM outperforms existing performance estimation baselines in various gradual shift scenarios and that its uncertainty awareness guides label acquisition more effectively compared to other strategies.

However, the success of ICL varies depending on the task and context, leading to heterogeneous service quality. Directly estimating the performance of LLM services at each invocation can be laborious, especially requiring abundant labeled data or internal information within the LLM. This paper introduces a novel method to estimate the performance of LLM services across different tasks and contexts, which can be "plug-and-play" utilizing only a few unlabeled samples like ICL. Our findings suggest that the negative log-likelihood and perplexity derived from LLM service invocation can function as effective and significant features. Based on these features, we utilize four distinct meta-models to estimate the performance of LLM services. Our proposed method is compared against unlabeled estimation baselines across multiple LLM services and tasks. And it is experimentally applied to two scenarios, demonstrating its effectiveness in the selection and further optimization of LLM services.

Knowing if a model will generalize to data 'in the wild' is crucial for safe deployment. To this end, we study model disagreement notions that consider the full predictive distribution - specifically disagreement based on Hellinger distance, Jensen-Shannon and Kullback-Leibler divergence. We find that divergence-based scores provide better test error estimates and detection rates on out-of-distribution data compared to their top-1 counterparts. Experiments involve standard vision and foundation models.

Materials discovery driven by statistical property models is an iterative decision process, during which an initial data collection is extended with new data proposed by a model-informed acquisition function--with the goal to maximize a certain "reward" over time, such as the maximum property value discovered so far. While the materials science community achieved much progress in developing property models that predict well on average with respect to the training distribution, this form of in-distribution performance measurement is not directly coupled with the discovery reward. This is because an iterative discovery process has a shifting reward distribution that is over-proportionally determined by the model performance for exceptional materials. We demonstrate this problem using the example of bulk modulus maximization among double perovskite oxides. We find that the in-distribution predictive performance suggests random forests as superior to Gaussian process regression, while the results are inverse in terms of the discovery rewards. We argue that the lack of proper performance estimation methods from pre-computed data collections is a fundamental problem for improving data-driven materials discovery, and we propose a novel such estimator that, in contrast to na\"ive reward estimation, successfully predicts Gaussian processes with the "expected improvement" acquisition function as the best out of four options in our demonstrational study for double perovskites. Importantly, it does so without requiring the over thousand ab initio computations that were needed to confirm this prediction.

Performance estimation under covariate shift is a crucial component of safe AI model deployment, especially for sensitive use-cases. Recently, several solutions were proposed to tackle this problem, most leveraging model predictions or softmax confidence to derive accuracy estimates. However, under dataset shifts, confidence scores may become ill-calibrated if samples are too far from the training distribution. In this work, we show that taking into account distances of test samples to their expected training distribution can significantly improve performance estimation under covariate shift. Precisely, we introduce a "distance-check" to flag samples that lie too far from the expected distribution, to avoid relying on their untrustworthy model outputs in the accuracy estimation step. We demonstrate the effectiveness of this method on 13 image classification tasks, across a wide-range of natural and synthetic distribution shifts and hundreds of models, with a median relative MAE improvement of 27% over the best baseline across all tasks, and SOTA performance on 10 out of 13 tasks. Our code is publicly available at https://github.com/melanibe/distance_matters_performance_estimation.

Testing deep learning-based systems is crucial but challenging due to the required time and labor for labeling collected raw data. To alleviate the labeling effort, multiple test selection methods have been proposed where only a subset of test data needs to be labeled while satisfying testing requirements. However, we observe that such methods with reported promising results are only evaluated under simple scenarios, e.g., testing on original test data. This brings a question to us: are they always reliable? In this paper, we explore when and to what extent test selection methods fail for testing. Specifically, first, we identify potential pitfalls of 11 selection methods from top-tier venues based on their construction. Second, we conduct a study on five datasets with two model architectures per dataset to empirically confirm the existence of these pitfalls. Furthermore, we demonstrate how pitfalls can break the reliability of these methods. Concretely, methods for fault detection suffer from test data that are: 1) correctly classified but uncertain, or 2) misclassified but confident. Remarkably, the test relative coverage achieved by such methods drops by up to 86.85%. On the other hand, methods for performance estimation are sensitive to the choice of intermediate-layer output. The effectiveness of such methods can be even worse than random selection when using an inappropriate layer.

Many applications utilize sensors in mobile devices and machine learning to provide novel services. However, various factors such as different users, devices, and environments impact the performance of such applications, thus making the domain shift (i.e., distributional shift between the training domain and the target domain) a critical issue in mobile sensing. Despite attempts in domain adaptation to solve this challenging problem, their performance is unreliable due to the complex interplay among diverse factors. In principle, the performance uncertainty can be identified and redeemed by performance validation with ground-truth labels. However, it is infeasible for every user to collect high-quality, sufficient labeled data. To address the issue, we present DAPPER (Domain AdaPtation Performance EstimatoR) that estimates the adaptation performance in a target domain with only unlabeled target data. Our key idea is to approximate the model performance based on the mutual information between the model inputs and corresponding outputs. Our evaluation with four real-world sensing datasets compared against six baselines shows that on average, DAPPER outperforms the state-of-the-art baseline by 39.8% in estimation accuracy. Moreover, our on-device experiment shows that DAPPER achieves up to 396X less computation overhead compared with the baselines.

Cross-validation (CV) is one of the main tools for performance estimation and parameter tuning in machine learning. The general recipe for computing CV estimate is to run a learning algorithm separately for each CV fold, a computationally expensive process. In this paper, we propose a new approach to reduce the computational burden of CV-based performance estimation. As opposed to all previous attempts, which are specific to a particular learning model or problem domain, we propose a general method applicable to a large class of incremental learning algorithms, which are uniquely fitted to big data problems. In particular, our method applies to a wide range of supervised and unsupervised learning tasks with different performance criteria, as long as the base learning algorithm is incremental. We show that the running time of the algorithm scales logarithmically, rather than linearly, in the number of CV folds. Furthermore, the algorithm has favorable properties for parallel and distributed implementation. Experiments with state-of-the-art incremental learning algorithms confirm the practicality of the proposed method.