This paper investigates the expected excess risk of In-Context Learning (ICL) for multi-class classification. We model each task as a sequence of labeled prompt samples and a query input, where a pre-trained model estimates the conditional class probabilities of the query. The expected excess risk is defined as the average truncated Kullback-Leibler (KL) divergence between the predicted and ground-truth conditional class distributions, averaged over a specified family of tasks. We establish a new oracle inequality for the expected excess risk based on KL divergence in multiclass classification. This allows us to derive tight upper and lower bounds for the expected excess risk in transformer-based models, demonstrating that the ICL estimator achieves the minimax optimal rate--up to a logarithmic factor--for conditional probability estimation. From a technical standpoint, our results introduce a novel method for controlling generalization error using the uniform empirical covering entropy of the log-likelihood function class. Furthermore, we show that multilayer perceptrons (MLPs) can also perform ICL and achieve this optimal rate under specific assumptions, suggesting that transformers may not be the exclusive architecture capable of effective ICL.

Gradient-based causal discovery shows great potential for deducing causal structure from data in an efficient and scalable way. Those approaches however can be susceptible to distributional biases in the data they are trained on. We identify two such biases: Marginal Distribution Asymmetry, where differences in entropy skew causal learning toward certain factorizations, and Marginal Distribution Shift Asymmetry, where repeated interventions cause faster shifts in some variables than in others. For the bivariate categorical setup with Dirichlet priors, we illustrate how these biases can occur even in controlled synthetic data. To examine their impact on gradient-based methods, we employ two simple models that derive causal factorizations by learning marginal or conditional data distributions - a common strategy in gradient-based causal discovery. We demonstrate how these models can be susceptible to both biases. We additionally show how the biases can be controlled. An empirical evaluation of two related, existing approaches indicates that eliminating competition between possible causal factorizations can make models robust to the presented biases.

The Expectation-Maximisation (EM) algorithm is a central tool in statistics and machine learning, widely used for latent-variable models such as Gaussian Mixture Models (GMMs). Despite its ubiquity, EM is typically treated as a non-differentiable black box, preventing its integration into modern learning pipelines where end-to-end gradient propagation is essential. In this work, we present and compare several differentiation strategies for EM, from full automatic differentiation to approximate methods, assessing their accuracy and computational efficiency. As a key application, we leverage this differentiable EM in the computation of the Mixture Wasserstein distance $\mathrm{MW}_2$ between GMMs, allowing $\mathrm{MW}_2$ to be used as a differentiable loss in imaging and machine learning tasks. To complement our practical use of $\mathrm{MW}_2$, we contribute a novel stability result which provides theoretical justification for the use of $\mathrm{MW}_2$ with EM, and also introduce a novel unbalanced variant of $\mathrm{MW}_2$. Numerical experiments on barycentre computation, colour and style transfer, image generation, and texture synthesis illustrate the versatility and effectiveness of the proposed approach in different settings.

The automated identification of clusters or isolated points is a fundamental step in many classification and spatial analysis pipelines [1, 2, 3] to identify structures in unlabeled data. Clustering typically begins by assigning labels to data points, indicating their membership to one or more groups. However, the strategies used to define these groups can vary significantly across clustering methods, depending on the underlying assumptions about data structure, density, or similarity. Clustering and classification algorithms can be broadly categorized into partitioning-based, hierarchical, and density-based methods. Partitioning methods, such as K-means [4, 5], Spectral Clustering [6], and Support Vector Machines (SVMs) [7], divide the data into distinct groups by optimizing specific criteria. K-means partitions data into a fixed number of spherical clusters by minimizing within-cluster variance. Spectral Clustering extends partitioning by leveraging the eigenstructure of similarity graphs to identify clusters with complex, non-convex shapes through an embedding step followed by a partitioning algorithm. Similarly, SVMs perform classification by implicitly mapping data into higher-dimensional feature spaces using the kernel trick, effectively partitioning data through linear separation in that transformed space.

Mental disorders are clinically significant patterns of behavior that are associated with stress and/or impairment in social, occupational, or family activities. People suffering from such disorders are often misjudged and poorly diagnosed due to a lack of visible symptoms compared to other health complications. During emergency situations, identifying psychiatric issues is that's why challenging but highly required to save patients. In this paper, we have conducted research on how traditional machine learning and large language models (LLM) can assess these psychiatric patients based on their behavioral patterns to provide a diagnostic assessment. Data from emergency psychiatric patients were collected from a rescue station in Germany. Various machine learning models, including Llama 3.1, were used with rescue patient data to assess if the predictive capabilities of the models can serve as an efficient tool for identifying patients with unhealthy mental disorders, especially in rescue cases.

Vehicle safety assessment is crucial for consumer information and regulatory oversight. The New Car Assessment Program (NCAP) assigns standardized safety ratings, which traditionally emphasize passive safety measures but now include active safety technologies such as Advanced Driver-Assistance Systems (ADAS). It is crucial to understand how these various systems interact empirically. This study explores whether particular ADAS features like Forward Collision Warning, Lane Departure Warning, Crash Imminent Braking, and Blind Spot Detection, together with established vehicle attributes (e.g., Curb Weight, Model Year, Vehicle Type, Drive Train), can reliably predict a vehicle's likelihood of earning the highest (5-star) overall NCAP rating. Using a publicly available dataset derived from NCAP reports that contain approximately 5,128 vehicle variants spanning model years 2011-2025, we compared four different machine learning models: logistic regression, random forest, gradient boosting, and support vector classifier (SVC) using a 5-fold stratified cross-validation approach. The two best-performing algorithms (random forest and gradient boost) were hyperparameter optimized using RandomizedSearchCV. Analysis of feature importance showed that basic vehicle characteristics, specifically curb weight and model year, dominated predictive capability, contributing more than 55% of the feature relevance of the Random Forest model. However, the inclusion of ADAS features also provided meaningful predictive contributions. The optimized Random Forest model achieved robust results on a held-out test set, with an accuracy of 89.18% and a ROC AUC of 0.9586. This research reveals the use of machine learning to analyze large-scale NCAP data and highlights the combined predictive importance of both established vehicle parameters and modern ADAS features to achieve top safety ratings.

Notifications are an important communication channel for delivering timely and relevant information. Optimizing their delivery involves addressing complex sequential decision-making challenges under constraints such as message utility and user fatigue. Offline reinforcement learning (RL) methods, such as Conservative Q-Learning (CQL), have been applied to this problem but face practical challenges at scale, including instability, sensitivity to distribution shifts, limited reproducibility, and difficulties with explainability in high-dimensional recommendation settings. We present a Decision Transformer (DT) based framework that reframes policy learning as return-conditioned supervised learning, improving robustness, scalability, and modeling flexibility. Our contributions include a real-world comparison with CQL, a multi-reward design suitable for non-episodic tasks, a quantile regression approach to return-to-go conditioning, and a production-ready system with circular buffer-based sequence processing for near-real-time inference. Extensive offline and online experiments in a deployed notification system show that our approach improves notification utility and overall session activity while minimizing user fatigue. Compared to a multi-objective CQL-based agent, the DT-based approach achieved a +0.72% increase in sessions for notification decision-making at LinkedIn by making notification recommendation more relevant.

--T alent identification plays a critical role in promoting student development. However, traditional approaches often rely on manual processes or focus narrowly on academic achievement, and typically delaying intervention until the higher education stage. This oversight overlooks diverse non-academic talents and misses opportunities for early intervention. T o address this gap, this study introduces T alentPredictor, a novel semi-supervised multi-modal neural network that combines Transformer, LSTM, and ANN architectures. This model is designed to predict seven different talent types--academic, sport, art, leadership, service, technology, and others--in secondary school students within an offline educational setting. Drawing on existing offline educational data from 1,041 local secondary students, T alentPredictor overcomes the limitations of traditional talent identification methods. By clustering various award records into talent categories and extracting features from students' diverse learning behaviors, it achieves high prediction accuracy (0.908 classification accuracy, 0.908 ROCAUC). This demonstrates the potential of machine learning to identify diverse talents early in student development. ALENT is a critical component in human society. It is indispensable to the development of societies and the competitiveness of countries. Last but not least, talent is always in high demand. Thus, nurturing talent is the top priority for every part of the earth, and in it, talent identification is the foundation, as you must have a target individual to nurture talent. Traditional talent identification aims to give students tests that exceed their current level. For example, give grade eight students college admissions tests and use the result of the tough test as a talent score.

In this paper, we apply both supervised and unsupervised machine learning algorithms to the study of the string landscape and swampland in 6-dimensions. Our data are the (almost) anomaly-free 6-dimensional $\mathcal{N} = (1,0)$ supergravity models, characterised by the Gram matrix of anomaly coefficients. Our work demonstrates the ability of machine learning algorithms to efficiently learn highly complex features of the landscape and swampland. Employing an autoencoder for unsupervised learning, we provide an auto-classification of these models by compressing the Gram matrix data to 2-dimensions. Through compression, similar models cluster together, and we identify prominent features of these clusters. The autoencoder also identifies outlier models which are difficult to reconstruct. One of these outliers proves to be incredibly difficult to combine with other models such that the $\text{tr}R^{4}$ anomaly vanishes, making its presence in the landscape extremely rare. Further, we utilise supervised learning to build two classifiers predicting (1) model consistency under probe string insertion (precision: 0.78, predicting consistency for 214,837 models with reasonable certainty) and (2) inconsistency under anomaly inflow (precision: 0.91, predicting inconsistency for 1,909,359 models). Notably, projecting these predictions onto the autoencoder's 2-dimensional latent layer shows consistent models clustering together, further indicating that the autoencoder has learnt interesting and complex features of the set of models and potentially offers a novel approach to mapping the landscape and swampland of 6-dimensional supergravity theories.

Visual storytelling systems struggle to maintain character identity across frames and link actions to appropriate subjects, frequently leading to referential hallucinations. These issues can be addressed through grounding of characters, objects, and other entities on the visual elements. We propose StoryReasoning, a dataset containing 4,178 stories derived from 52,016 movie images, with both structured scene analyses and grounded stories. Each story maintains character and object consistency across frames while explicitly modeling multi-frame relationships through structured tabular representations. Our approach features cross-frame object re-identification using visual similarity and face recognition, chain-of-thought reasoning for explicit narrative modeling, and a grounding scheme that links textual elements to visual entities across multiple frames. We establish baseline performance by fine-tuning Qwen2.5-VL 7B, creating Qwen Storyteller, which performs end-to-end object detection, re-identification, and landmark detection while maintaining consistent object references throughout the story. Evaluation demonstrates a reduction from 4.06 to 3.56 (-12.3%) hallucinations on average per story and an improvement in creativity from 2.58 to 3.38 (+31.0%) when compared to a non-fine-tuned model.