All papers

The paper evaluates vision-language models' spatial representations using causally controlled perturbations, showing spatial relations degrade compared to lexical features.

The paper introduces a domain-general framework using predictive world models to guide action selection and policy improvement in recursive self-improvement settings.

Studies how the non-generative decision model Jev can serve as a prior or policy component within reinforcement learning frameworks.

Introduces a multivariate time series foundation model pretrained using synthetic temporal and relational pattern primitives for zero-shot forecasting.

Analyzes collider bias in Bayesian physics-informed neural networks and introduces a hierarchical generative chain model to resolve parameter estimation drift.

Pairs attack success rate with operative understanding rate to assess whether models recognize harmful intent during safety evaluations.

Proposes an external memory framework allowing frozen language model agents to learn and iteratively refine natural-language world models compiled into executable code.

Presents a rollout selection strategy that drops high-probability positive-advantage trajectories to prevent entropy collapse during language model reinforcement learning.

Surveys and taxonomizes recurring architectural patterns across language-model-integrated software applications, ranging from chatbots to autonomous agent systems.

Proposes a conformal prediction framework that learns stochastic groupings to improve subgroup coverage reliability without inflating overall prediction sets.

Proposes a framework modeling research ideation as modular thinking flows sampled dynamically from a probabilistic supernet.

Introduces a benchmark evaluating multimodal language models on extracting data, tracking narrative flows, and answering questions from animated data videos.

Proposes a benchmark and probing methodology based on attribution theory to examine how language models assign responsibility and blame.

Presents a simulated participant environment to evaluate multi-turn AI interviewers across market research and social science scenarios.

Presents an explainer for molecular graph neural networks combining factual, counterfactual, and exemplar graph perturbations.

Audits the reliability of LLM judges across multiple prompt formats, temperatures, and orderings, introducing a trustworthy verdict metric.

Applies zonotope geometry to adaptively scale critic pessimism and disagreement handling in soft actor-critic locomotion policies.

Applies Weisfeiler-Leman graph kernel features to improve automated algorithm selection for constraint optimization problems without training GNNs.

Presents a three-stage hierarchical generation method for large attributed graphs using recursive soft community decomposition.

Proposes a training framework that teaches reasoning models to plan by reverse-engineering solution ideas from hindsight examples.

Introduces AgenticBBO-Bench, a benchmark across multiple domains to evaluate LLM agents on black-box optimization under unified finite budgets.

Reports an equal-budget study finding randomized-restart list scheduling outperforms Monte Carlo Tree Search on DAG workflow scheduling.

Presents AdaCast, a parameter generation method producing input-specific low-rank updates for foundation models in time-series forecasting.

Proposes ARGUS, an agentic framework separating deterministic genomic models from LLM reasoning to interpret noncoding single nucleotide variants reliably.

Evaluates whether LLM compliance systems adjust verdicts when governing rules are perturbed, finding many verdicts are insensitive to rule changes.

Investigates whether LLM scientific agents rely on task priors or experimental feedback when configuring neural operator adaptations.

Develops a framework fusing small randomized controlled trials with large observational data to improve treatment effect estimation precision without bias.

Presents RiCo, a neural physics simulation method that models rigid-body interactions via sparse local contact surface reasoning.

Introduces AAArena, a benchmark evaluating how AI agents revise executable code policies across 12 adversarial games under fixed budgets.

Analyzes long-tail concept learning during supervised fine-tuning using a theoretical prior barrier metric to guide data needs.

Proposes an evaluation framework to measure how LLM agents handle uncertainty and resolve conflicting knowledge during execution.

Audits legal chain-of-thought faithfulness by substituting cited authorities and observing whether model verdicts causally change.

Proposes a streaming optimal transport method to monitor LLM agent trajectories in real time against reference runs.

Proposes a higher-order tensor modeling framework to preserve synergistic cross-modal interactions in self-supervised multimodal representations.

Evaluates whether internal model representations can predict how fine-tuning on one value generalizes across other held-out values.

Develops multi-layer linear probes on model activations to detect agent deception and sabotage across diverse scenarios.

Proposes a generative flow matching method with repulsion-guided sampling to model one-to-many bifurcation trajectories in physical systems.

Recomputes METR task time horizons using splines and item-response theory to examine the validity of AI difficulty curves.

Introduces a generative model that allows flexible post-hoc text editing through learned insertion, substitution, and deletion operations.

Presents a benchmark evaluating agentic large language models on long-horizon strategic reasoning in the game Slay the Spire.

Investigates self-preservation in language models by using task banks to steer synthetic replay during continual instruction tuning.

Benchmarks machine learning and pretrained foundation models for day-ahead electricity forecasting in a university building.

Demonstrates a backdoor attack where malicious reasoning degradation is triggered incrementally using invisible prompt tokens and stage tracking.

Analyzes zero-advantage training plateaus in reinforcement learning with verifiable rewards and proposes recursive data synthesis to address them.

Proposes a recurrent harness framework that uses streaming feedback to revise textual memory for continual agent learning.

Introduces a benchmark evaluating language models on eliciting geospatial data requirements and constraints from natural language requests.

Develops a temporal encoder injecting engineering domain priors to forecast rare, high-cost events in multivariate time series.

Constructs a multimodal simulator dataset and trajectory prediction model for autonomous driving in urban roadway work zones.

Proposes an inference-time unlearning method that detects leaks and injects corrective embeddings into large reasoning model traces.

Introduces an evolutionary optimization method using multi-turn reflective mutators to improve production data agents and pipelines.

Develops a framework for evolving deterministic formal policies to serve as runtime guardrails for large language model agents.

Investigates mappings between generative model latent states and simulator states to enable controllable environment simulation.

Introduces a multimodal embodied agent policy learning framework using trajectory construction, distillation, and token-level evidence routing.

Introduces a multi-turn conversational benchmark to evaluate sales agents against resistant, psychologically grounded buyer personas.

Compares hyperbolic and Euclidean embeddings for unsupervised patent representation to improve cross-category prior-art discovery.

Presents a scalable method to maximize agent empowerment by guiding agents toward states affording numerous executable skills.

Proposes a generalized f-divergence trust-region framework to unify and stabilize reinforcement learning algorithms for LLM post-training.

Presents a simulation benchmark evaluating how embodied agents continually acquire basic skills and recombine them for complex tasks.

Presents a conversational benchmark testing whether models can edit text under counterfactual changes while preserving unaffected factual details.

Introduces a framework allowing agents to rewind long-horizon environments to earlier states while retaining learned exploratory knowledge.