AI Research Paper Feed

TLDR: This paper establishes tight long-term tail decay rates for SGD and clipped SGD in non-convex optimization, showing significantly faster decay than previously known results.

TLDR: The paper introduces Evidential Uncertainty Quantification (EUQ), a method to detect misbehaviors in large vision-language models by assessing internal conflicts and knowledge gaps, outperforming existing methods in identifying issues like hallucinations and adversarial vulnerabilities.

TLDR: ADANA, an optimizer with time-varying schedules for hyperparameters, improves large-scale language model training efficiency by up to 40% compared to AdamW.

TLDR: MerNav introduces a Memory-Execute-Review framework that significantly improves success rates and generalization in zero-shot object goal navigation tasks.

TLDR: xList-Hate is a new framework for hate speech detection that uses a checklist-based approach for improved interpretability and robustness across different datasets and conditions.

TLDR: This paper presents a unified convergence analysis for the SAG, SAGA, and IAG algorithms, providing a simpler and more comprehensive understanding of their performance.

TLDR: The paper introduces a framework using piecewise deterministic Markov processes (PDMP) for efficient Bayesian inference in complex inverse problems, demonstrating improved accuracy and efficiency over traditional methods.

TLDR: The study evaluates the soundness of generative models in chess by using adversarial sequences to reveal their limitations and improve training techniques.

TLDR: EuroLLM-22B is a large language model designed to support all 24 EU languages and 11 additional languages, showing strong multilingual performance and being openly released for research use.

TLDR: The SAGE benchmark reveals that traditional BM25 outperforms LLM-based retrievers for scientific literature retrieval, with enhancements possible through document augmentation using LLMs.

TLDR: The paper introduces Codified Finite-State Machines (CFSMs) and their probabilistic extension (CPFSMs) to improve character state modeling in role-playing with large language models, enhancing consistency and engagement by automatically generating state transitions from character profiles.

TLDR: CE-RAG4EM is a cost-efficient RAG system for entity matching that reduces computational overhead while maintaining or improving performance.

TLDR: DARWIN is an evolutionary GPT model that uses a genetic algorithm approach to iteratively improve its performance by allowing GPT agents to modify each other's training code, resulting in a 1.26% improvement in FLOPS utilization and a 2.07% improvement in perplexity over five iterations.

TLDR: TADS is a new framework for selecting high-quality, task-relevant data for multi-task multimodal pre-training, improving efficiency and performance using less data.

TLDR: PatchFlow improves defect detection in die casting using local patch features and a flow-based model, reducing error rates significantly on multiple datasets.

TLDR: This paper presents a new construction of a transformer that can solve the PARITY problem using a single layer with practical features, and establishes a lower bound proving that a single-layer, single-head transformer cannot solve PARITY.

TLDR: The paper introduces new Radon--Wasserstein gradient flows for efficient high-dimensional sampling using interacting particles with linear scaling costs.

TLDR: DLM-Scope introduces a sparse autoencoder-based framework for interpreting diffusion language models, showing unique advantages over traditional autoregressive models.

TLDR: The paper introduces a Bayesian framework to adaptively determine the optimal neighborhood scope for graph neural networks, improving performance on node classification tasks across various datasets.

TLDR: CORP is a one-shot structured pruning method for Vision Transformers that preserves accuracy without retraining by using a closed-form solution to compensate for removed structures.