The paper introduces Quantile-Guided Diffusion for Probabilistic Time-Series Forecasting, showing that integrating predicted quantiles as distributional controls can guide a pretrained diffusion forecaster. The authors find that applying residual control throughout the denoising network yields the most consistent improvements, leading to the Control-TS model that learns a quantile-control pathway while keeping the backbone frozen. Across eight benchmarks, Control-TS improves the backbone’s CRPS, competes with state‑of‑the‑art models, and retains gains in capacity‑matched comparisons, with further analyses on quantile granularity, predictor choice, and forecast quality. The study also presents the Conditional Linear Predictive Score (CLPS) to evaluate downstream forecasting utility of generated futures.
By Shadi Omari, Bashar Khoury, Assaf Schuster
The paper introduces compact representations called sliced Wasserstein sketches that enable efficient, geometrically meaningful comparisons of probability distributions. It presents an exact decomposition of the squared distance error into residual and partition‑mismatch components, providing computable error certificates for any centroid sketch. Experiments demonstrate that optimal sketches reduce storage while maintaining classification performance, improve generalization with limited labels, and can be used as compact pooling layers in end‑to‑end learning.
By Elaheh Akbari, James Liu, Keaton Hamm, Soheil Kolouri
Gibbs‑Gap Policy Optimization (GGPO) is a new off‑policy reinforcement learning method that avoids high‑variance learner‑to‑behavior importance weights by minimizing a log‑mean‑exp contrast of KL‑adjusted rewards across responses to the same prompt. Derived from KL‑regularized policy optimization, GGPO guarantees that a learner’s single‑response expected reward is at least the behavior policy’s best‑of‑K reward minus bounded terms, and it includes a reference‑refresh mechanism that empirically improves performance. Experiments on benchmark problems and mathematical reasoning tasks show GGPO outperforming GRPO and AGRO, especially under training‑inference mismatch and with fixed offline data.
whyItMatters":"The paper demonstrates a practical improvement in off‑policy RL for language models by reducing variance and ensuring better reward guarantees, which is directly supported by the reported empirical gains and theoretical bounds."
By Xuanchang Zhang, Tong Zhang
arXiv:2610. 21001v1 Announce Type: new Abstract: Column generation folklore warns that initial columns hurt: a warm start may bias the dual and slow the root of branch-price-and-cut.
By Haoran Liu, Yu Yang
The paper investigates how language models’ internal judgments can be separated from the way those judgments are expressed. By comparing hidden-state vectors under different instructions, the authors identify a low-dimensional expression subspace that can be removed to isolate the model’s true judgment. This isolation improves predictive performance across tasks such as correctness, entailment, and safety, even when the model’s final answers degrade.
By Chaemin Jang, Donggyu Lee, Jeesoo Kim, Dongman Lee, Jihee Kim
RDL-Bench is a new benchmark for task-aware question answering over relational databases, introduced in arXiv:2610.21382v1. It categorizes user questions into reasoning-oriented queries, which depend on the current database state, and prediction-oriented queries, which rely on future outcomes under temporal constraints. Each benchmark instance includes a natural-language question, task metadata, schema details, temporal context, executable evidence or target-construction logic, and denotation-level ground truth, supporting evaluation in both native solver and task-aware modular system settings.
By Yao Cheng, Siqiang Luo
The paper introduces a graph deep learning framework that learns to forecast spatiotemporal data across multiple hierarchical structures, using variations among hierarchies as multi‑task regularization. By jointly training on different aggregation levels, the model reduces over‑specialization and improves forecasting accuracy. A novel top‑down operator ensures coherent forecasts with linear time and space complexity, making the approach computationally efficient.
By Valentina Moretti, Cesare Alippi, Andrea Cini
The paper presents a quantum machine learning approach to content-based media recommendation, mapping BBC iPlayer programme synopses to variational quantum circuits within the Quantum Natural Language Processing framework. The quantum model, trained via supervised pairwise similarity learning, matches a classical BiLSTM baseline in performance while using far fewer parameters (47k vs. 17.8M). Experiments on Quantinuum's System Model H2 confirm that the similarity signal survives realistic noise, demonstrating the feasibility of deploying variational quantum circuits for recommendation tasks.
By Mina Abbaszadeh, Ouissal Moumou, Anna Pearson, Bob Coecke, Rob Cooper, Alessandro Piscopo, Mehrnoosh Sadrzadeh
The paper investigates why direct pairwise AUC surrogate training, though aligned with the ranking metric, can fail to provide corrective updates during stochastic training. It introduces the concept of Corrective Power, showing that in certain regions severe ranking errors persist without generating strong gradient signals, a phenomenon linked to degeneracies in the U‑statistics of pairwise gradients. The authors explain empirical observations—such as the effectiveness of squared‑hinge loss, the benefit of cross‑entropy warm‑up, and the advantage of parameter‑efficient fine‑tuning—through this theoretical lens, and validate their claims with experiments on eight computer‑vision and eight NLP tasks using various models.
By Jia Chen, Zhiyong Yang, Shilong Bao, Qianqian Xu, Qingming Huang
Lethe is a new benchmark for Edge Unlearning (EU) in Graph Neural Networks, addressing the lack of comprehensive evaluation tools for forgetting edges. It includes 15 EU methods tested on 8 datasets, focusing on large graphs and varying task difficulty. The benchmark replaces traditional accuracy metrics with empirical link inference attacks to better assess EU effectiveness.
By Andrea D'Angelo, Francesco Gullo, Davide Mottin, Sayan Ranu, Giovanni Stilo
The paper introduces the Universal Weight Encoder (UWE), a scalable representation for weight-space learning that uses a typed graph of interfaces and modules to respect parameter symmetries across diverse neural network architectures. UWE combines local equivariant updates with global attention over invariant interface summaries, enabling a single encoder to process multiple supported architectures efficiently. Experiments show UWE matches or surpasses architecture‑specific state‑of‑the‑art methods on standard benchmarks and excels on a large zoo of 1,200 transformers, ranking unseen architectures accurately while using significantly less memory.
By Micha{\l} Kozyra, Michael Bronstein
The study evaluates probabilistic AI weather models—GenCast, FourCastNet 3, and FGN—against the ECMWF IFS-ENS ensemble for tropical cyclone forecasting. Using new probabilistic metrics, the authors find that AI models, particularly FGN, outperform IFS-ENS in cyclone tracking, achieving better location scores for 90% of storms in 2025. The analysis also identifies intensity and spread limitations, model‑specific failure modes, and the influence of pretraining data, highlighting opportunities for targeted improvement.
By Nathan Chalumeau, Xavier Renard, Remi Meynadier, Hugo Rakotoarimanga, Marcin Detyniecki, Emmanuel de Bezenac
The paper introduces topology‑inspired fine‑tuning for wind‑field super‑resolution, adding auxiliary losses based on persistence diagrams and merge‑tree distances to a pretrained CNN. A comprehensive study of 16 fine‑tuning setups shows that gradient supervision improves bottleneck persistence‑diagram metrics, while fixed‑pair supervision yields the most consistent gains in both persistence‑diagram and merge‑tree distances. The combined objective reduces key topological distances by up to 31% and improves PSNR by 1.30 dB compared to the original CNN.
By Arjun Dadhwal, Brian Summa
The paper introduces Continuous Adversarial Training for Consistent Honesty (CATCH), a method that trains large language models to produce honest responses while preserving their overall capabilities. Using only a few hundred self‑reporting scenarios, CATCH is applied to Llama‑3.3‑70B‑Instruct and Qwen3‑3.6‑27B, and the resulting models show improved honesty across multiple benchmarks, including MASK, Among Us, and AuditBench. Compared to prompting, supervised fine‑tuning, and state‑of‑the‑art honesty steering, CATCH achieves superior honesty while maintaining utility through a novel choice of utility data.
By Yuan-Chen Chang, Niklas Herbster, Gauthier Gidel, Tommaso Tosato
PathFM is a trajectory-based forecasting model that uses flow matching to generate complete future paths for multivariate time series. It distills the sampling process to one or two steps, sharing a history attention cache across draws, allowing multiple path generation in roughly the time of a single quantile query. The model achieves competitive marginal accuracy on fev-bench and TIME, and outperforms baselines on the new Forecast Path Score Bench (fps‑bench) with energy, variogram, and downstream metrics, while maintaining long‑run variability in chaotic systems.
By Jiachen Yao, Xiyuan Zhang, Annan Yu, Boran Han, Abdul Fatir Ansari, Shuai Zhang, Michael Bohlke-Schneider, Bernie Wang
The paper introduces DAR (Discover-and-Relabel), a support‑persistent backdoor technique that remains effective even after models undergo clean‑data knowledge distillation. DAR discovers naturally occurring patterns in low‑dimensional feature subspaces via subspace clustering and then poisons labels of samples that satisfy these patterns. At inference, the backdoor is triggered by locally modifying the target image’s subspace, achieving attack accuracy comparable to state‑of‑the‑art methods while maintaining over 96% success after clean distillation.
By Junhoo Lee, Baekseung Kim, Seungyeon Kim, Nojun Kwak
The paper presents a machine‑learning soft sensor that estimates bottom‑hole pressure (BHP) in petroleum wells using wellhead and topside data. It employs a Long Short‑Term Memory (LSTM) model, compares it with Multi‑Layer Perceptron and Ridge Regression, and introduces transfer learning to adapt the model across different operational settings. Tested on offshore datasets from Brazil’s Pre‑salt basin, the LSTM achieved a mean absolute percentage error below 2%, outperforming the benchmarks.
By M. A. Fernandes, E. Gildin, M. A. Sampaio
The paper introduces a token‑level log‑likelihood ratio to pinpoint where large language models memorize training sequences. It proposes a probe‑steer framework, instantiated as Gated Subspace Steering (GSS), that selectively corrects memorization‑relevant activations using the leading singular vectors of a gradient‑weighted activation matrix. Experiments on four benchmarks show GSS matches or surpasses state‑of‑the‑art memorization reduction methods while using 10–100× less compute.
By Xuanqi Zhang, Haoyang Shang, Xiaoxiao Li
The paper introduces Amortized Factor Inference Networks (AFINs), a single inference network that can generalize across varying priors, likelihoods, latent dimensions, and observation counts. AFINs represent a model as a list of typed prior and likelihood factors and use an encode–merge–decode architecture to produce a Gaussian posterior approximation, with learned parameters independent of latent dimensionality or observation count. Experiments on synthetic and real datasets demonstrate that a single trained AFIN achieves competitive posterior approximations at significantly lower test‑time cost than NUTS and per‑task variational inference, with optional importance sampling further improving accuracy.
By Joohwan Ko, Justin Domke
The study evaluates 13 Geospatial Foundation Models (GFMs) for estimating above‑ground biomass (AGB) from satellite imagery, comparing them to a fully supervised state‑of‑the‑art baseline trained on the AGBD dataset. Most GFMs perform worse than the baseline, but TESSERA and AlphaEarth Foundations (AEF) features combined with specific prediction heads achieve superior results, with the best model attaining an RMSE of 50.79 Mg/ha. This AEF‑based model also generalizes better to unseen regions, reduces RMSE by up to 25% in excluded areas, and matches the ESA CCI biomass product when trained on limited regional labels.
By Ghjulia Sialelli, Linus Scheibenreif, Jan Dirk Wegner, Konrad Schindler