All papers

The paper proposes a tree search method that optimizes agent skills through contrastive atomic rule editing and execution memory.

The paper proposes a concept-guided regularizer using sparse autoencoders to reduce catastrophic forgetting during parameter-efficient fine-tuning.

Presents ReDiS, a self-evolution framework separating exploration and candidate diagnosis to improve multimodal reasoning in remote sensing.

This paper proposes an optimal-transport bridge framework to translate audio representations directly between mismatched neural codecs at low bitrates.

Formalizes language model creativity evaluation as long-tail input-output distribution performance and tests it across multiple benchmarks and models.

Deconstructs multiple-choice option bias in language models into two factors tracking answer flips and positional skew.

Distills multiple speech attributes from four teacher models into a single shared speech encoder.

The paper presents an approach using character-level tokens, larger models, and synthetic data for the multilingual BabyLM data-efficiency challenge.

This paper proposes a self-play reinforcement learning framework that automatically generates tasks and intermediate verifiable rewards for long-horizon agent training.

The paper analyzes screening collapse, showing how recursive training on synthetic data degrades feature selection prior to model training.

Identifies translation circuits in multilingual models and optimizes fewer than five percent of parameters for low-resource translation.

This paper presents a distillation method where a teacher model minimally edits a student web agent's reasoning traces at failure points.

NepPlural provides an annotated benchmark of Nepali discourse to evaluate how reliably language models capture diverse socio-cultural personas.

The paper evaluates audio forced aligners and fine-tunes Whisper to predict human-aligned word boundaries across read, conversational, and telephone speech.

Introduces an adaptive curriculum reinforcement learning method to train multimodal models on long-horizon visual Sudoku reasoning tasks.

This paper presents ReaDy, a speculative decoding method using hybrid linear attention recurrent states to provide position-specific context during drafting.

This paper introduces a benchmark evaluating whether coding agents can reimplement and train biological AI tools from published scientific papers.

ScaleMAS coordinates multi-agent systems using an asynchronous shared datastore with limited search to improve scaling and fault tolerance.

The paper analyzes aesthetic judgments by foundation models on K-pop music videos to measure inter-model consensus across audio and visual modalities.

Evaluates deep learning optimizers under a self-tuning fixed-hardware benchmark across nine distinct workloads.

The paper evaluates how language exposure distribution and ordering affect the performance of multilingual models trained with limited data.

This paper proposes a multi-agent system designed to reduce clinical medication errors through dynamic collaborative coordination.

Measures representation drift across continual learning methods, finding n-gram memory updates cause less internal drift than LoRA.

The authors propose an embodied world model that unifies RGB video, metric depth, and robot actions into a shared dynamic representation.

Introduces a unified benchmark evaluating five public photoplethysmography foundation models across nine datasets and 26 clinical and wearable tasks.

This paper proposes an attention-based noise-aware training method to improve the robustness of Transformer models deployed on analog compute-in-memory hardware.

The paper introduces an evaluation framework that diagnoses bottlenecks and dynamically regenerates benchmark tasks for language model agents.

The paper evaluates embedding distance metrics to detect generalization boundaries and distribution shifts in AI-based code vulnerability models.

Estimates surprisal directly from continuous speech signals using a speech language model and importance sampling over abstract linguistic units.

Introduces a benchmark evaluating vision-language models on graph-theoretic reasoning where evidence is split across multiple visual graph views.

The paper distils a power-grid foundation model into a compact student network for real-time edge inference on power systems.

The paper presents an evaluation framework to test the interactive agentic capabilities of base language models before post-training.

Evaluates whether failure feedback helps self-improving agents fix specific target bugs rather than general errors.

Presents a pragmatic prompting framework to preserve speaker voice, emotion, and tone during speech-to-speech translation.

The paper reviews LLM-based first-order logic autoformalization, distinguishing ontology extraction from logical translation and outlining key benchmarking challenges.

The paper presents a multimodal benchmark for competition-level chemistry that grades reasoning step-by-step with partial credit.

Adds content-addressable associative memory to latent world models to prevent collapse and improve long-horizon planning.

The paper presents a Lean 4 runtime allowing self-extending agents to run verified computations across tools without constant human oversight.

The paper presents an evidence admission framework that verifies anomalies before updating the persistent principles of scientific language model agents.

The paper develops a text-to-speech model that simulates realistic clinical dysfluency in aphasia speech across multiple speakers for training healthcare professionals.

The paper introduces a two-stage low-rank fine-tuning method that uses residual gradients to add complementary adaptation directions without expanding the rank.

The paper introduces an event ledger and executable aggregation framework to help agents accurately query and summarize trajectory history.

This paper presents a byte-level watermarking method that allows a single detector to verify text across models with different tokenizers.

The paper introduces ASSET, a reinforcement learning method that adaptively compresses speech encoder outputs into continuous audio tokens.

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 introduces a method to convert dense models into mixture-of-experts by dynamically allocating different numbers of active experts across layers.

The paper introduces a discrete codebook representation that pairs associative indexing with vector search for long-term agent memory.

This paper develops a runtime verification architecture that decouples monitoring formalisms from execution boundaries to safely control AI agents.

The paper presents a multi-stage automated labeling framework using multimodal models and agents to annotate short videos for content moderation.

This paper predicts human annotator disagreement on harmful content across multiple languages by extracting interpretable semantic question features with models.

The paper investigates context-aware model routing rules across turns in stateful multi-step agent applications.

The paper introduces a compact tabular foundation model combining distribution-aware representations with dataset-conditioned sparse mixture-of-experts computation.

The paper proposes a reinforcement learning framework using natural language metrics to reduce object hallucination in vision-language image captioning.

The paper compares discriminative and generative language models for rhetorical structure parsing and elementary discourse unit segmentation.

Applies vision-language models and structured annotation to parse and analyze historical comic books and manga at scale.

The paper enables multimodal models to proactively ask clarifying questions when ambiguous inputs carry cross-modal safety risks.

The paper introduces a framework that incrementally updates robot spatial judgments after environmental changes without full scene recomputation.

The paper introduces a single-policy optimization framework combining privileged adaptation with stored historical next-token distributions for language model reasoning.

The paper evaluates whether providing page images to vision-language models improves the detection of errors in Japanese-English manga translation.

Proposes PRISM, a multimodal clinical prognostic model combining patient records with retrieved population-level embedding memory banks.