Every story tagged with this topic, ordered by date.
TurnSight applies turn-level hindsight self-distillation to Tool-Integrated Reasoning, enabling finer credit assignment than trajectory-level RL in long-horizon agent tasks.
Test-time scaling taxonomy clarifies distinct inference algorithms (sampling, voting, search) in reasoning LLMs, standardizing evaluation protocols and compute accounting.
Agogic demonstrates music tokenization—not model scale—drives text-to-symbolic-music generation quality, using Qwen backbone with seven representations and Frechet Music Distance.
Conditional diffusion models for synthetic histopathology images; evaluates FID/IS limitations with domain-specific metrics for medical data generation.
LLMs recover missing semantic optimizations in C/C++ code that compilers overlook, evaluated on SeGaBench with 120 cases.
Video-DeepResearch extends multimodal agents to video streams; identifies modality bias and knowledge leakage bottlenecks in current models.
ReflectRL improves LLM reasoning by learning from failed expert trajectories as reflective signals during on-policy post-training.
Information-geometric analysis of GFlowNet forward-policy training using Fisher-Rao metric and natural gradient optimization.
Unconditional separations prove low-depth quantum circuits compute tasks intractable for constant-round diffusion and transformer LMs.
Adaptive test-time scaling via fuzzy controller allocates per-query compute based on prompt complexity and model confidence.
Game-theoretic analysis shows foundation model agents achieve rational cooperation through similarity inference, challenging decoupled-agency assumptions.
TACT: taxonomy-aligned post-training framework for pedagogically adaptive LLM-based ESL tutoring with human-grounded evaluation.
Muon optimizer shows token-efficiency gains on Mamba-2 when applied selectively to output projections, with benefits localized to specific weight groups.
Logic pre-pretraining on formal derivations accelerates LM skill acquisition and improves compressibility compared to narrow symbolic tasks.
Latent Reward Registers enable preference alignment in Diffusion Transformers via learnable tokens that extract reward signals from intermediate noisy latents.
R-ItCUR algorithm recovers low-tubal-rank tensors from partial cross-concentrated sampling observations with sparse outlier corruption.
Physics-Flavored Neural Network automates kinetic phenotyping of engineered skeletal muscle tissues, replacing simplistic peak-force metrics.
PRISM meta-workflow converts multivariate time series to multi-channel images for anomaly detection, evaluating vision backbone alternatives.
Transformers execute Sequence-level Interactive Dynamic Parallel Processing (SIDPP), constructing prompt-dependent parameter transformations at inference rather than reproducing training statistics.
Standard music transformers lose equivariance to pitch transposition and time shifts as scale increases, allocating capacity to memorizing absolute patterns.
EcoFrame enables training-free adaptive frame scheduling for long video understanding in VLMs, using inference feedback to adjust sampling budgets.
First implementation of causal perception framework for competing Structural Causal Models, operationalizing fairness in agent reasoning systems.
Acceleration Matching algorithm for smooth trajectory inference from discrete snapshots.
Sparse Weight Decomposition enables efficient mechanistic interpretability of pretrained transformers without sparse retraining.
Contrastive activation addition steers temporal horizon preferences in Qwen3-32B via learned linear representations.
Omega-S penalty retains fine-tuned model knowledge without prior-task data, Fisher matrix, or weight copies.
Zero-shot LLM reranking pipeline for ADHD symptom sentence ranking from Reddit text using BM25, embeddings, and semantic rescoring.
MultiGlobeQA: 46K-example multilingual benchmark for geospatial reasoning exposing LLM failures in spatial relations across 14 function families.
Feasibility-aware generative model for AC-operable synthetic power-grid scenarios for resilience and contingency analysis.
Structure-Aware Fine-Tuning (SAFT) refines VLM reward models via self-supervised learning without ground-truth labels for RL.
GENESIS: Explainable causal discovery from observational data combining LLM reasoning with statistical methods to justify edge decisions in DAGs.
CRS-Triage: ML model for emergency patient acuity prediction with confidence scoring handling incomplete and unreliable EHR data.
Source-Conditioned Description-Length Gain detects LLM-generated plagiarism via probabilistic compression, distinguishing source reuse from similarity.
CheMatE: ModernBERT-based embedding model for joint SMILES-NLP representation learning in chemistry, addressing domain overfitting.
Quantization precision (FP16/INT8/INT4) and prompt design critically affect biomedical LLM classifier calibration on Mistral variants.
FedCritic-MIMO: federated multi-agent RL for 6G RAN resource control via serverless critic sharing.
Layer-wise analysis shows sensitivity, causality, and repair capacity dissociate when LLMs fail on perturbed input; identifies spike-and-suppress vs. late-accumulation regimes.
Oilbird: training-free speculative decoding using verifier-computed keys; improves suffix matching for tool-calling workloads.
LatentGuard: safeguard framework compressing textual reasoning into latent states for efficient, inspectable LLM content moderation.
RESUME CONTRACT: TLA+ specification for checkpoint/interrupt/resume semantics across workflow persistence layers, exposing non-compliant agent frameworks.
Geo-Embed: unified multimodal embedding model for urban/geospatial tasks spanning street-view imagery, remote sensing, text, and temporal change.
FlowForm: satellite flood image synthesis using SWE-inspired latent regularization and structure-aware conditioning for data augmentation.
Computing actual causes for neural network predictions using Halpern-Pearl causal models and Boolean SCMs to improve explainability under structured input dependencies.
MDLMPE: distribution-aware positional encoding for masked diffusion language models to handle dynamic token-availability patterns during denoising.
Faithfulness-safety tension in Large Reasoning Models: models must be interpretable for monitoring yet robust against unsafe reasoning paths.
MissClick: adversarial attack on GUI grounding models exploiting digit-serialized coordinate generation to induce large spatial click displacement.
Taxonomy of multilingual multi-agent planning failures: request-to-action grounding degradation strongest in low-resource languages.
Self-augmentation method for MLLMs using model failure signals to generate targeted image augmentations without external supervision.
GPTKB 2.0 constructs disambiguated knowledge bases directly from LLM outputs using on-the-fly entity/relation resolution.
Black-box diagnostic for LLM collectives measuring whether output diversity correlates with genuine epistemic revision.