1071 papers
This paper demonstrates that commonly used bias metrics fail to reliably capture allocational harms from large language models because they overlook the critical discrepancy between model predictions and the actual decisions made when allocating limited resources.
The paper introduces Goldfish, a suite of over 1,000 small monolingual language models for 350 languages that outperform large multilingual models in grammaticality and perplexity, particularly for low-resource languages where existing models often struggle.
This paper proposes a resource-efficient and interpretable framework that mitigates biases in large language models at decoding time by leveraging small, fine-tuned expert models to generate debiasing signals, achieving significant reductions in gender, race, and religion biases while preserving overall model performance.
The paper introduces SpecFuse (referred to as SpecEM in the abstract), a training-free ensemble framework that enhances large language model performance by enabling segment-level semantic collaboration through speculative decoding and dynamically adjusting model weights via an online feedback mechanism to prioritize stronger contributors.
This paper introduces LEO, a streamlined multimodal large language model architecture that employs a lightweight fusion strategy of post-adaptation projectors, tile-level sequence interleaving, and dynamic tiling to significantly enhance visual understanding across diverse benchmarks and specialized domains like autonomous driving.
This survey provides a comprehensive overview of the emerging ecosystem of large language models and tools that support researchers across the scientific lifecycle, covering key tasks from literature search and idea generation to content creation, experimentation, and evaluation, while addressing associated datasets, methods, limitations, and ethical concerns.
This paper proposes a methodology to fine-tune Large Language Models for generating fluent English-Spanish code-switched text by leveraging back-translated parallel corpora, demonstrating that while traditional metrics fail to correlate with human preferences, LLM-based evaluation aligns well with human judgment and the approach significantly advances CS text generation capabilities.
The paper introduces CAReDiO, a novel data optimization framework that enhances cultural alignment in Large Language Models by iteratively optimizing data for representativeness and distinctiveness, enabling efficient and superior alignment across 15 diverse cultures with as few as 200 training samples.
The paper introduces Reasoning Reward Models (ReasRMs), specifically the RM-R1 family, which reformulate reward modeling as a reasoning task using a chain-of-rubrics mechanism and a two-stage training pipeline to achieve superior interpretability and performance compared to existing large-scale models.
This paper introduces EVA, a novel event-by-event asynchronous-to-synchronous (A2S) framework inspired by language modeling that generates highly expressive features, outperforming prior methods in recognition tasks and achieving state-of-the-art results in detection for event-based vision.
This paper proposes K-CAST, a novel fine-grained conditional activation steering method that dynamically mitigates content biases in large language models, significantly improving formal reasoning accuracy by up to 15% while maintaining robustness across prompts and languages.
This paper introduces AdAEM, a novel self-extensible evaluation framework that automatically generates adaptive test questions by probing the internal value boundaries of diverse LLMs to overcome the limitations of static benchmarks and provide more informative, distinguishable insights into models' value differences and alignment dynamics.
This paper introduces a deterministic, automated pipeline that transforms raw domain corpora into completion-style benchmarks to provide scalable, unbiased, and LLM-independent evaluations of domain expertise in both base and instruction-tuned models, effectively addressing issues of benchmark contamination and multiple-choice bias.
The paper introduces Sysformer, a novel approach that safeguards frozen large language models by learning to adapt system prompts in the embedding space to significantly improve safety robustness against harmful inputs and jailbreaking attacks without requiring costly model fine-tuning.
This paper introduces VLMQ, a post-training quantization framework tailored for vision-language models that leverages a gradient-driven importance factor to address visual over-representation and modality gaps, thereby achieving state-of-the-art performance across various model sizes and low-bit settings.
This paper presents a case study demonstrating that Hybrid Retrieval-Augmented Generation (RAG) consistently outperforms direct long-context prompting across English, French, and German technical agricultural manuals, achieving over 85% accuracy with models like Gemini 2.5 Flash and Qwen 2.5 7B in cross-lingual question answering.
The paper proposes CMRAG, a novel framework that enhances visual document retrieval and question answering by unifying text and image modalities through a shared embedding space and a normalized co-modality retrieval method, thereby outperforming existing single-modality approaches.
MERLIN is a two-stage, curriculum-based framework that integrates multilingual encoders with LLMs using efficient DoRA fine-tuning to significantly enhance cross-lingual reasoning performance, particularly in low-resource languages where existing methods and even GPT-4o-mini fall short.
This paper addresses the inconsistency and structural biases in existing latency metrics for simultaneous speech-to-text translation by introducing a comprehensive meta-evaluation, proposing new metrics (YAAL and LongYAAL) and a resegmentation tool (SoftSegmenter), and implementing these solutions within the OmniSTEval toolkit to enable more reliable system assessments.
This paper demonstrates that while standard decoder-only models underperform compared to encoder-only architectures in cross-modal adaptation for partial differential equations, introducing novel bidirectionality-mimicking techniques like Parallel Flipping and Sequence Doubling effectively closes this performance gap.