Skip to content

Latest

arXiv cs.LG Papers

Tree of Thoughts as a Classical Heuristic Search Problem: Formal Foundations and Design Patterns

arXiv:2605.28566v2 Announce Type: replace-cross Abstract: Large Language Models (LLMs) have demonstrated remarkable reasoning capabilities, yet their standard generation process -- auto-regressive token prediction -- is inherently…

arXiv cs.LG Papers

CAMP: A Cycle-Aware Multi-Scale Patch Mixer for Time Series Forecasting

arXiv:2608.04051v1 Announce Type: new Abstract: Real-world time series are often governed by recurring patterns, but their dominant periods may vary across datasets, forecasting settings, and individual…

arXiv cs.LG Papers

GLOBE: Trajectory-Aligned Gradient Matching with Structured SparseOptimization for Coreset Selection

arXiv:2608.02690v1 Announce Type: new Abstract: On-device training of deep neural networks is fundamentally constrained by the computational and memory costs of large-scale datasets. Coreset selection offers…

arXiv cs.LG Papers

Target-Aligned Fusion for Decision-Sequence Learning under Dynamics Shift

arXiv:2511.09173v3 Announce Type: replace Abstract: External trajectories can improve offline decision-sequence learning, but dynamics shift may make some source subsequences inconsistent with the target environment. We…

arXiv cs.LG Papers

CT-HEG: A Bidirectional, Timestamp-Attributed Event Graph for ICU In-Hospital Mortality Prediction – An Architectural Ablation Study

arXiv:2608.02663v1 Announce Type: new Abstract: Accurate ICU mortality prediction requires modeling irregular clinical observations across heterogeneous entity types. Existing sequence models handle irregular sampling but ignore…

arXiv cs.LG Papers

Sphere Retraction Normalizations

arXiv:2608.02668v1 Announce Type: new Abstract: Residual connections are the de facto mechanism for training deep neural networks stably. Geodesic Normalization (GeoNorm) recasts them on a Riemannian…

arXiv cs.LG Papers

Verifier-Guided Model Discovery for Physical Dynamical Systems with Pretrained Symbolic Transformers

arXiv:2608.02662v1 Announce Type: new Abstract: Reliable forecasting of nonlinear physical systems underpins scientific discovery and engineering decision-making. Yet high-fidelity simulations are prohibitively costly, and machine-learning surrogates…

arXiv cs.LG Papers

GeoID-PINN: Identifiability-Aware Regional Epidemic Inference with Geographic Coupling

arXiv:2608.02633v1 Announce Type: new Abstract: Regional surveillance data reflect local transmission, reporting, seeding, and external infection pressure, which are difficult to identify separately. We introduce GeoID-PINN,…

arXiv cs.LG Papers

LLMs Can Annotate Attribution Graphs

arXiv:2608.02632v1 Announce Type: new Abstract: Circuit tracing is an exciting technique for revealing the internal computation of language models, but it requires a time-intensive manual step…

arXiv cs.LG Papers

Prescribed-Basis Coefficient-to-Coefficient Neural Operator for Partial Differential Equations

arXiv:2510.10350v3 Announce Type: replace-cross Abstract: Operator learning provides a data-driven approach to approximating solution operators of partial differential equations, but its effectiveness depends strongly on how…