Self-training often falls short under distribution shifts due to an increased discrepancy between prediction confidence and actual accuracy. This typically necessitates computationally demanding methods such as neighborhood or ensemble-based label corrections. Drawing inspiration from insights on early learning regularization, we develop a principled method to improve self-training under distribution shifts based on temporal consistency. Specifically, we build an uncertainty-aware temporal ensemble with a simple relative thresholding. Then, this ensemble smooths noisy pseudo labels to promote selective temporal consistency. We show that our temporal ensemble is asymptotically correct and our label smoothing technique can reduce the optimality gap of self-training. Our extensive experiments validate that our approach consistently improves self-training performances by 8% to 16% across diverse distribution shift scenarios without a computational overhead. Besides, our method exhibits attractive properties, such as improved calibration performance and robustness to different hyperparameter choices.

The applicability domain refers to the range of data for which the prediction of the predictive model is expected to be reliable and accurate and using a model outside its applicability domain can lead to incorrect results. The ability to define the regions in data space where a predictive model can be safely used is a necessary condition for having safer and more reliable predictions to assure the reliability of new predictions. However, defining the applicability domain of a model is a challenging problem, as there is no clear and universal definition or metric for it. This work aims to make the applicability domain more quantifiable and pragmatic. Eight applicability domain detection techniques were applied to seven regression models, trained on five different datasets, and their performance was benchmarked using a validation framework. We also propose a novel approach based on non-deterministic Bayesian neural networks to define the applicability domain of the model. Our method exhibited superior accuracy in defining the Applicability Domain compared to previous methods, highlighting its potential in this regard.

Temporal Heterogeneous Networks play a crucial role in capturing the dynamics and heterogeneity inherent in various real-world complex systems, rendering them a noteworthy research avenue for link prediction. However, existing methods fail to capture the fine-grained differential distribution patterns and temporal dynamic characteristics, which we refer to as spatial heterogeneity and temporal heterogeneity. To overcome such limitations, we propose a novel \textbf{C}ontrastive Learning-based \textbf{L}ink \textbf{P}rediction model, \textbf{CLP}, which employs a multi-view hierarchical self-supervised architecture to encode spatial and temporal heterogeneity. Specifically, aiming at spatial heterogeneity, we develop a spatial feature modeling layer to capture the fine-grained topological distribution patterns from node- and edge-level representations, respectively. Furthermore, aiming at temporal heterogeneity, we devise a temporal information modeling layer to perceive the evolutionary dependencies of dynamic graph topologies from time-level representations. Finally, we encode the spatial and temporal distribution heterogeneity from a contrastive learning perspective, enabling a comprehensive self-supervised hierarchical relation modeling for the link prediction task. Extensive experiments conducted on four real-world dynamic heterogeneous network datasets verify that our \mymodel consistently outperforms the state-of-the-art models, demonstrating an average improvement of 10.10\%, 13.44\% in terms of AUC and AP, respectively.

Understanding node representations in graph-based models is crucial for uncovering biases ,diagnosing errors, and building trust in model decisions. However, previous work on explainable AI for node representations has primarily emphasized explanations (reasons for model predictions) rather than interpretations (mapping representations to understandable concepts). Furthermore, the limited research that focuses on interpretation lacks validation, and thus the reliability of such methods is unclear. We address this gap by proposing a novel interpretation method-Node Coherence Rate for Representation Interpretation (NCI)-which quantifies how well different node relations are captured in node representations. We also propose a novel method (IME) to evaluate the accuracy of different interpretation methods. Our experimental results demonstrate that NCI reduces the error of the previous best approach by an average of 39%. We then apply NCI to derive insights about the node representations produced by several graph-based methods and assess their quality in unsupervised settings.

Curriculum learning (CL) describes a machine learning training strategy in which samples are gradually introduced into the training process based on their difficulty. Despite a partially contradictory body of evidence in the literature, CL finds popularity in deep learning research due to its promise of leveraging human-inspired curricula to achieve higher model performance. Yet, the subjectivity and biases that follow any necessary definition of difficulty, especially for those found in orderings derived from models or training statistics, have rarely been investigated. To shed more light on the underlying unanswered questions, we conduct an extensive study on the robustness and similarity of the most common scoring functions for sample difficulty estimation, as well as their potential benefits in CL, using the popular benchmark dataset CIFAR-10 and the acoustic scene classification task from the DCASE2020 challenge as representatives of computer vision and computer audition, respectively. We report a strong dependence of scoring functions on the training setting, including randomness, which can partly be mitigated through ensemble scoring. While we do not find a general advantage of CL over uniform sampling, we observe that the ordering in which data is presented for CL-based training plays an important role in model performance. Furthermore, we find that the robustness of scoring functions across random seeds positively correlates with CL performance. Finally, we uncover that models trained with different CL strategies complement each other by boosting predictive power through late fusion, likely due to differences in the learnt concepts. Alongside our findings, we release the aucurriculum toolkit (https://github.com/autrainer/aucurriculum), implementing sample difficulty and CL-based training in a modular fashion.

Ensemble clustering aggregates multiple weak clusterings to achieve a more accurate and robust consensus result. The Co-Association matrix (CA matrix) based method is the mainstream ensemble clustering approach that constructs the similarity relationships between sample pairs according the weak clustering partitions to generate the final clustering result. However, the existing methods neglect that the quality of cluster is related to its size, i.e., a cluster with smaller size tends to higher accuracy. Moreover, they also do not consider the valuable dissimilarity information in the base clusterings which can reflect the varying importance of sample pairs that are completely disconnected. To this end, we propose the Similarity and Dissimilarity Guided Co-association matrix (SDGCA) to achieve ensemble clustering. First, we introduce normalized ensemble entropy to estimate the quality of each cluster, and construct a similarity matrix based on this estimation. Then, we employ the random walk to explore high-order proximity of base clusterings to construct a dissimilarity matrix. Finally, the adversarial relationship between the similarity matrix and the dissimilarity matrix is utilized to construct a promoted CA matrix for ensemble clustering. We compared our method with 13 state-of-the-art methods across 12 datasets, and the results demonstrated the superiority clustering ability and robustness of the proposed approach. The code is available at https://github.com/xuz2019/SDGCA.

The prevalence of vector similarity search in modern machine IVF indexes out-of-the-box do not have the notion of inserting learning applications and the continuously changing nature of data new vectors or deleting existing vectors once constructed. Indeed, processed by these applications necessitate efficient and effective the most common method used by practitioners today is to rebuild index maintenance techniques for vector search indexes. Designed the index from scratch to reflect any updates that have accumulated primarily for static workloads, existing vector search indexes degrade over time. However, depending on the scale of the vector in search quality and performance as the underlying data is dataset and the volume and frequency of updates, a full index rebuild updated unless costly index reconstruction is performed. To address can be prohibitively expensive. For example, it takes multiple this, we introduce Ada-IVF, an incremental indexing methodology days to rebuild an IVF index from scratch for billion-scale vector for Inverted File (IVF) indexes. Ada-IVF consists of 1) an adaptive datasets [21, 69], making it necessary to revisit how updates can maintenance policy that decides which index partitions are problematic be reflected. Devising such an update mechanism consists of readjusting for performance and should be repartitioned and 2) a local the partitioning of the high-dimensional space defined by re-clustering mechanism that determines how to repartition them.

Simplicity bias poses a significant challenge in neural networks, often leading models to favor simpler solutions and inadvertently learn decision rules influenced by spurious correlations. This results in biased models with diminished generalizability. While many current approaches depend on human supervision, obtaining annotations for various bias attributes is often impractical. To address this, we introduce Debiasify, a novel self-distillation approach that requires no prior knowledge about the nature of biases. Our method leverages a new distillation loss to transfer knowledge within the network, from deeper layers containing complex, highly-predictive features to shallower layers with simpler, attribute-conditioned features in an unsupervised manner. This enables Debiasify to learn robust, debiased representations that generalize effectively across diverse biases and datasets, improving both worst-group performance and overall accuracy. Extensive experiments on computer vision and medical imaging benchmarks demonstrate the effectiveness of our approach, significantly outperforming previous unsupervised debiasing methods (e.g., a 10.13% improvement in worst-group accuracy for Wavy Hair classification in CelebA) and achieving comparable or superior performance to supervised approaches. Our code is publicly available at the following link: Debiasify.

Plotting the residuals is a recommended procedure to diagnose deviations from linear model assumptions, such as non-linearity, heteroscedasticity, and non-normality. The presence of structure in residual plots can be tested using the lineup protocol to do visual inference. There are a variety of conventional residual tests, but the lineup protocol, used as a statistical test, performs better for diagnostic purposes because it is less sensitive and applies more broadly to different types of departures. However, the lineup protocol relies on human judgment which limits its scalability. This work presents a solution by providing a computer vision model to automate the assessment of residual plots. It is trained to predict a distance measure that quantifies the disparity between the residual distribution of a fitted classical normal linear regression model and the reference distribution, based on Kullback-Leibler divergence. From extensive simulation studies, the computer vision model exhibits lower sensitivity than conventional tests but higher sensitivity than human visual tests. It is slightly less effective on non-linearity patterns. Several examples from classical papers and contemporary data illustrate the new procedures, highlighting its usefulness in automating the diagnostic process and supplementing existing methods.

This paper introduces Dirichlet process mixtures of block $g$ priors for model selection and prediction in linear models. These priors are extensions of traditional mixtures of $g$ priors that allow for differential shrinkage for various (data-selected) blocks of parameters while fully accounting for the predictors' correlation structure, providing a bridge between the literatures on model selection and continuous shrinkage priors. We show that Dirichlet process mixtures of block $g$ priors are consistent in various senses and, in particular, that they avoid the conditional Lindley ``paradox'' highlighted by Som et al.(2016). Further, we develop a Markov chain Monte Carlo algorithm for posterior inference that requires only minimal ad-hoc tuning. Finally, we investigate the empirical performance of the prior in various real and simulated datasets. In the presence of a small number of very large effects, Dirichlet process mixtures of block $g$ priors lead to higher power for detecting smaller but significant effects without only a minimal increase in the number of false discoveries.