classification

An LLM judge consistently returned a fixed confidence score of 0.72 in evaluations, but switching to categorical labels improved score distribution, showing that models are better at classification than numerical estimation for assessments.

This paper presents a black-box method for LLM safety classification using dynamical systems and Koopman operators on prompt-response embedding dynamics to detect unsafe outputs.

The article discusses a new paper arguing that the current classification of human ancestors into genera like Homo, Australopithecus, and Paranthropus is outdated and needs revision based on new fossil and genetic evidence.

This research paper compares various machine learning classifiers for heart disease prediction, finding that Support Vector Machine and Simple Cart achieve the best performance on UCI and Kaggle datasets respectively, highlighting ML's potential to aid in early clinical diagnosis.

ChiroEcho is a deep learning framework that extends automated bat vocalisation classification by combining species and genus predictions with geographic data, increasing operational coverage of European bat species from 73% to 85%.

This paper conducts a controlled audit of memory-policy classification for personalized agents, demonstrating that explicit state elicitation does not significantly improve policy accuracy on a frozen counterfactual dataset for models like Llama-3.3-70B and GPT-OSS-120B.

This paper introduces a data-driven framework combining supervised classification and generative modeling to reconstruct separatrices in multistable dynamical systems, using neural networks and score-based generative models to approximate boundaries of basins of attraction.

Evaluation of gpt-4o-mini and gpt-4o on an event classification system showed gpt-4o performed better, but both models had unreliable confidence scores for real-world decision-making.

Google AI Devs demonstrated Gemini 3.7 Flash by using it to extract and classify plants from historical botanical PDFs with an interactive visualization.

The paper proposes a robust dual-model collaborative random vector functional link network (KRPRVFL) to improve classification accuracy in the presence of noisy labels and outliers, leveraging kernel risk-sensitive mean p-power criterion and collaborative learning.

This paper proposes an adaptive and efficient KNN classifier via granular-ball computing that dynamically determines the k value using granular ball neighborhoods, improving accuracy and robustness while reducing computational cost. The method is open-sourced on GitHub.

This paper introduces GENADA, a generative adversarial attack framework that learns to produce deceptive perturbations for time series classifiers in a single forward pass, achieving comparable attack quality to iterative baselines with lower inference time.

This paper proposes a new asymmetric robust bounded sparse smooth (aR) loss function for l1-norm penalized geometric twin support vector machine (aRSGTSVM) to handle classification and regression tasks with label and feature noise, achieving feature selection and robustness. Experiments on synthetic and UCI datasets plus China stock market index tracking demonstrate superiority.

This paper introduces LabelFusion-TS, which fuses a fine-tuned RoBERTa encoder, a prompted LLM, and time-series transformers over market data to classify Federal Reserve communication as hawkish, dovish, or neutral. The fused system achieves 70.2% weighted F1, outperforming a zero-shot LLM and showing early evidence that market time series help financial text classification.

The paper compares large language models and embedding models across 37 tasks, finding that while aggregate performance is similar, embedding models are far cheaper and faster, supporting a division of labor for cost-efficiency.

The paper proposes intuitionistic fuzzy deep RVFL (IF-dRVFL) and ensemble deep RVFL (IF-edRVFL) frameworks that use sample neighborhood information to improve robustness against noise and outliers in classification tasks, outperforming existing SOTA fuzzy and non-fuzzy approaches on benchmark datasets.

Introduces FreSH, a frequency-segmented hierarchical multi-expert framework for multivariate time series classification, achieving state-of-the-art accuracy on UEA benchmarks with reduced model size and computational cost.

This paper studies how multilingual training helps identify figurative language in proverbs across seven languages, introducing a multidimensional annotation framework and finding that about 50% of translated multilingual data is sufficient for near-optimal performance.

The article describes a technique for classifying e-commerce queries with LLMs by having the model hallucinate hypothetical classifications, then mapping them to the real taxonomy using embeddings, which is cheaper and simpler than constrained structured outputs.

This paper presents HNR-DAC, a two-stage framework for scientific claim verification over cited papers, combining hard-negative reranking and distribution-aligned classification. It achieves strong results on NLPCC 2026 Task 10 Track 2, ranking third on the leaderboard with the highest Macro-F1.