Topic Modeling

SEEK-VEC: Augmenting topic modeling with spectral ensemble learning

SEEK-VEC improves topic modeling by combining multiple models through spectral ensembling, enhancing performance in identifying vocabulary and relationships, especially with weak signals.

Thoughts-as-Planning: Latent World Models for Chain-of-Thoughts Optimization via Reinforcement Planning

This paper introduces a framework for optimizing reasoning chains in LLMs by modeling them as sequential decisions in a latent space, showing improved efficiency and interpretability.

Verifying LLM-extracted text with token alignment

This paper introduces a method to verify LLM-extracted text by aligning it with ordered subword tokenization, showing improved accuracy over word-level tokenization.

An Empirical Analysis of Fine-Tuning Large Language Models on Bioinformatics Literature: PRSGPT and BioStarsGPT

This paper presents a 9-step pipeline for fine-tuning LLMs on bioinformatics data, demonstrating PRSGPT and BioStarsGPT with performance metrics and QA datasets.

DeepResearch_Eco: A Recursive Agentic Workflow for Complex Scientific Question Answering in Ecology

DeepResearch_Eco enhances ecological research by recursively synthesizing scientific literature, integrating more sources per word than conventional methods.

Large language models implicitly learn to straighten neural sentence trajectories to construct a predictive representation of natural language

The paper investigates how transformer models straighten neural sentence trajectories to improve next-word prediction, showing curvature decreases with model performance and linguistic context.

DrBERT: A Robust Pre-trained Model in French for Biomedical and Clinical domains

This paper introduces DrBERT, a French biomedical pre-trained language model trained on both public and private medical data, showing improved performance through further pre-training on targeted data.

Enriching Biomedical Knowledge for Low-resource Language Through Translation

The paper translates biomedical data into Vietnamese to create resources for low-resource languages, introducing models and benchmarks that achieve state-of-the-art results.

Understanding and Improving Word Embeddings through a Neuroscientific Lens

The paper investigates whether word embeddings align with brain representations and improves them using brain features, showing increased coherence with human semantic judgments.