All papers

The paper proposes a framework that uses reasoning graphs to locate claim-level disagreements across sampled traces and correct multi-step language model reasoning.

The paper introduces a benchmark to evaluate how typed decision models handle theory-of-mind reasoning tasks compared to standard generative language models.

The paper evaluates whether self-distillation using privileged information improves reasoning capabilities or reflects overfitting to training tasks.

The authors present an approach using local language models to automatically generate conceptual architectural models and reports for digital twins.

This paper introduces a metric to estimate the compatibility of pretrained models before running costly alignment or checkpoint merging.

This paper investigates whether cloning a familiar voice systematically transfers personality impressions to an embodied virtual character.

The authors propose a training-free method constructing ontology-grounded semantic identifiers to preserve hierarchy in medical concept embeddings.

This paper investigates attention heads responsible for contextual entrainment and evaluates whether their ablation transfers across different semantic relations.

The paper introduces an auditing framework that identifies failure steps in multi-agent systems by evaluating how information is utilized across handoffs.

The paper presents a runtime system that structures long-horizon agent memory around reusable progress objects rather than raw interaction history.

The paper analyzes how language model aggregation gains divide into candidate additions and context effects under different baseline comparison setups.

The paper formalizes a framework and theoretical generalization bounds for evolving an agent's software harness without updating model weights.

The paper examines back-translation using monolingual data as an alternative to continued pretraining for low-resource translation with language models.

The paper introduces a decoding framework that dynamically selects sampling temperature based on geometric representations of conversational scenarios.

The paper proposes an agent architecture that dynamically switches between latent reasoning and text interaction while maintaining external tool access.

The paper introduces a training-free contrastive decoding method where a small model guides a larger model toward shorter reasoning chains.

The paper presents a dataset and framework with dense progressive rewards to train language agents on long-horizon tasks.

The paper introduces action chunking for agentic robotics to execute pre-planned primitive routines without redundant model reasoning at each step.

This paper presents a mid-training method, corpus, and benchmark to help language models understand and reason about time series data.

Summary pending (next check).

The paper reviews retrieval-augmented generation paradigms and examines frameworks across retrieval, generation, and modular components.

The paper presents an 8-bit weight and activation post-training quantization technique that migrates activation outliers to weights.

The paper introduces a post-training mixed-precision quantization method using GPU-compatible 2D weight tiles for efficient inference.

The paper presents an open-source multimodal language model combining a small language backbone with vision encoders for efficient inference.

Proposes an agent memory architecture that uses biologically inspired adaptive decay functions to selectively retain and forget context.

Introduces a mixture-of-prompts distillation method to transfer knowledge from hard prompts to soft prompts, improving generalization on unseen classes.

Uses coarse CTC decoding to filter relevant hotwords into LLM prompts for improved contextual speech recognition on rare words.

The paper introduces a zero-shot LLM pipeline to generate counterfactual examples for stress-testing NLP models.

Summary pending (next check).

Introduces NosRacer, a dynamic analysis framework to detect race conditions in on-device network operating systems.

Proposes error coverage and certainty metrics to evaluate the reliability of multi-model N-version object detection architectures.

Quantifies SMT solver runtime instability across program verification queries using the coefficient of variation across multiple domains.

Empirically analyzes how refactoring commits correlate with software quality metrics like complexity, code smells, and vulnerabilities.

Presents a formal verification framework for quantitative properties of recursive infinite-state networks using pumping semirings.

Automates GPU benchmark reconstruction by profiling reference workloads and synthesizing equivalent target-architecture kernel implementations.

Develops a coexponential semantic model for concurrent services in the Clef language using delimited continuations to verify state ownership.

Formulates a simplified layout algebra for GPU kernels by explicitly typing coordinate spaces to eliminate complex admissibility checks.

Introduces an automated environment generation pipeline to synthesize diverse runnable desktop GUI software applications for training computer-use agents.

Introduces a configuration-driven platform for reproducible multi-agent reinforcement learning experiments within the ARGoS swarm robotics simulator.

Implements encrypted WebAssembly execution on Arm TrustZone to protect proprietary AI models on edge hardware.

The paper conducts an empirical study tracking how publicly flagged vulnerable AI models propagate across code repositories and downstream applications.

The paper proposes a context-aware prioritization engine that incorporates system operational details to improve enterprise vulnerability scoring and remediation decisions.

The paper provides an empirical study of cross-chain bridge verification vulnerabilities and categorizes common flaws in cross-chain smart contract validation logic.

The paper analyzes vulnerabilities from persistent artifacts shared across AI agent skill sessions and develops a runtime defense against resulting cross-session attacks.

The paper models structured sentiment analysis as dependency graph parsing implemented through sequence labeling across five languages.

The paper proposes a four-tensor attention architecture that jointly normalizes attention over semantic and temporal fibers for scene prediction.

The paper defines evaluation metrics to show that language models continue generating reasoning traces even when instructed not to think.

The paper introduces a dual-path parametric memory method that combines pooled history representations with sequential adapter updates.

The paper proposes a sparse factor framework to map multimodal behavioral data to interpretable concepts for depression assessment.

The paper demonstrates that entropy-based byte patching misses predictable calculation points and proposes a boundary-dependence signal to fix it.

The paper develops an LLM-based pipeline to automatically detect outcome switching and presentation spin in clinical trial reports.

The paper benchmarks a specialized decision model against general-purpose LLMs on English and Chinese stance detection tasks.

The paper introduces a method to unlearn target knowledge while recovering collateral capability loss in non-target tasks.

The paper proposes an event-centric memory framework that converts dialogue history into query-relevant graph working memory for conversational agents.

The paper describes training a competitive decoder-only search reranker from scratch using modest computational resources and common datasets.

The paper introduces a 7,000-task benchmark to evaluate how multimodal models selectively use or ignore conversational context.

The report presents an omni-modal model family trained by scaling reinforcement learning throughput, environments, and agentic reward compute.

The paper benchmarks a specialized multiple-choice decision model against 19 general-purpose LLMs across knowledge, reasoning, and multilingual tasks.

The paper introduces an educational tool combining Arabic knowledge graphs and passage retrieval to generate narrative-based questions.

The paper meta-analyzes 25 years of automated verbal deception detection models and identifies a persistent accuracy ceiling around 70 to 75 percent.