machine-learning

A tweet from The Turing Post recommending Alexander Jung's book 'Machine Learning: The Basics' as a compact refresher covering the data-model-loss framework, including hypothesis spaces, model selection, ERM, regularization, probabilistic models, clustering, federated learning, privacy, and explainability.

This arXiv paper introduces CohortHijack, a robustness audit that removes non-target cells from single-cell query cohorts to test how annotation tools can be manipulated without altering the target cell's expression profile. It shows that structured removal and search strategies can change refined labels in popular pipelines while preserving the target, identifying query cohort composition as a vulnerability surface.

Introduces Fast Evidential Rule Learning (FERL), a method for interpretable classification that produces evidential outputs and can abstain when uncertain, with theoretical stability guarantees and strong empirical results across tabular and concept-bottleneck benchmarks.

This arXiv paper presents a shape-constrained predictive modeling approach to estimate nanoparticle size and dispersity in continuous flow nanodrug production, reducing the need for extensive experimental screening.

This paper introduces PMOT, a potential-flow framework for general p-cost optimal transport using continuous normalizing flows, with theoretical zero-loss exactness and promising results on synthetic and high-dimensional benchmarks.

This position paper argues that many LLM failures stem from evaluating outputs independently and proposes a self-consistency framework that treats diverse techniques as special cases of consistency optimization.

Introduces Quantum-Structured World Models (QSWMs), a quantum-inspired framework for predictive world modeling with structured latent states, and evaluates them on elementary cellular automata against classical baselines.

This paper systematically evaluates 15 machine learning models, including the TabPFN foundation model, for post-wildfire debris-flow prediction using USGS basin-scale data, finding TabPFN achieves the best performance (threat score 0.637) and that synthetic data augmentation improves most models.

This paper introduces MS-MLB, an open machine learning benchmark for classifying multiple sclerosis from whole blood RNA expression data using the public GSE17048 cohort. It provides a reproducible, leakage-controlled evaluation pipeline and reports Gradient Boosting as the top performer.

The paper proposes DoctorAgents, an agentic AI framework that uses specialized LLM agents to iteratively generate, validate, and refine end-to-end machine learning pipelines for small, heterogeneous clinical temporal datasets, outperforming established AutoML baselines.

Otter is a 15.3M-parameter human chess AI that extends Maia 2 by conditioning move predictions on game history and time pressure, achieving higher accuracy than Maia 2 with fewer parameters. Trained on 6.1 billion positions from Lichess games, it demonstrates that treating chess as a time-aware, sequential activity improves prediction of human play.

The paper investigates whether guideline-based categorical encodings of continuous predictors can replace continuous inputs in stroke outcome prediction models without sacrificing accuracy, finding comparable performance in most treatment cohorts.

Proposes Woodpecker Distillation, a weak-to-strong training framework that uses weak probe models to identify and repair local reasoning bugs in stronger models via contrastive local interventions, improving performance on math reasoning benchmarks.

Presents counterfactual recoverability as an outcome-grounded decision variable for on-policy distillation, showing that selectively supervising based on recoverability outperforms divergence-only methods across AIME and GPQA benchmarks.

This paper introduces a probabilistic ensemble model based on a conditional diffusion model for real-time tsunami inundation forecasting, offering uncertainty quantification in contrast to deterministic warnings. Validated with 2011 Tohoku-oki data, it demonstrates that generative AI can shift tsunami forecasting from deterministic to probabilistic approaches.

This paper introduces ArborEnum, the first algorithm to exactly enumerate decision-tree Rashomon sets over continuous features without binarization, along with relaxed and anytime approximations that achieve orders-of-magnitude speedups while preserving near-perfect recall.

This paper introduces random feature maps for scalable kernel machines on Grassmannian manifolds, using bounded rank-one projections to approximate rotation-invariant Grassmannian kernels with reduced computational and memory costs.

This paper compares five feature selection methods for EHR diagnosis codes in opioid use disorder prediction, finding that NTK sensitivity offers the best accuracy-stability balance while LLM-guided selection adds complementary clinical signals.

Introduces LiNC, a lightweight noise correction method that learns per-sample trust parameters to distinguish clean and noisy labels using a Gaussian Mixture Model, achieving robust accuracy gains on medical imaging datasets under high label noise.

This preliminary study applies a fully convolutional neural network to automatically detect neutron resonances in transmission spectra, achieving ~93% classification accuracy but failing to generalize to unseen isotopes. The authors suggest future work with larger datasets and physics-informed features.