All Articles
17919 articles total
Evaluating LLM Usage for Efficient and Explainable Numerical and Classified Implicit Sentiment Analysis of Product Desirability
Research on using LLMs to quantify product desirability from qualitative feedback, showing that GPT-4o-mini provides cost-effective, high-accuracy sentiment analysis.
Heterogeneous 2D/1D Signal Representation Fusion for Underwater Acoustic Modulation Recognition Under Distribution Shift
Introduction of UAMR-ShiftBench and the SCP-TriCA model for robust underwater acoustic modulation recognition under distribution shifts.
Event-Aligned Analysis of Multi-Rater Pain Assessments Using Continuous Wearable Physiology
A framework for analyzing pain assessments by aligning continuous wearable physiological signals with rater-specific pain-change events.
Coordinate-Queryable Neural Field Reconstruction for EEG Spatial Super-Resolution with Unseen-Electrode Generation
A neural field reconstruction method for EEG spatial super-resolution that can synthesize signals for electrode locations not seen during training.
Audio-visual Contrastive Alignment for Diffusion-based Visual-conditioned Speech Enhancement
A diffusion-based speech enhancement method that uses contrastive audio-visual alignment to improve signal reconstruction in noisy environments.
Random coloured digraphs defined by a Markov logic network
A theoretical study on random coloured digraphs defined by Markov logic networks, establishing 0-1 laws for first-order logic under specific weight scaling.
Legal Reasoning Is Not Lawyering: Rethinking Legal Benchmarks for Pro Se Access to Justice
A critique of legal AI benchmarks, arguing they measure upper-bound performance and fail to account for the noisy inputs typical of pro se litigants.
A Unified Framework for Runtime Verification and Model-Based Diagnosis in LOLA
An integrated framework within the LOLA stream specification language that unifies runtime verification and model-based diagnosis for online fault localization.
Weight-Space Geometry of Offline Reasoning Training
A geometric analysis of offline reasoning training methods, finding that DPO is mechanistically distinct and achieves higher accuracy than SFT or GRPO.
Statistics that live in your SQL
A discussion on integrating statistical computations directly within SQL queries to optimize data analysis workflows.
European Commission's Metsola Overrides MEPs to Force Through Chat Control
The European Commission is pushing through 'Chat Control' legislation, potentially overriding MEPs to increase surveillance on encrypted messaging.
GTA VI finally gets a price tag
Rockstar Games has announced pricing for GTA VI, with the standard edition priced at $79.99 and the Ultimate Edition at $99.99.
OpenThoughts-Agent: Data Recipes for Agentic Models
The OpenThoughts-Agent project introduces a fully open data curation pipeline for training general-purpose agentic language models, releasing training sets and models.
Reentrant value fields as delayed coupled reaction-diffusion systems on finite graphs
A mathematical study on dynamical systems using retarded functional differential equations to describe coupled reaction-diffusion systems on finite graphs.
Beyond the Autoregressive Horizon: A Comprehensive Survey of Diffusion Models, World Modelling, and State Space Models for Code
A survey exploring alternatives to autoregressive models for code generation, including Diffusion Models, Code World Models, and State Space Models.
Quantifying Prior Dominance in RAG Systems
Researchers introduce the Normalized Context Utilization (NCU) metric to quantify how much RAG systems actually use provided context versus parametric memory.
SemChunk-C: Semantic Segmentation for C Code
SemChunk-C is a family of lightweight LLMs designed specifically for the semantic segmentation of C-family source code to improve RAG and retrieval.
FP8 is All You Need (Part 2): Efficient Ozaki-Bailey Style FFT Through Tensor-core Garner Reformulation and Kulisch Escape Route
A technical proposal to achieve FP64-equivalent 3-D FFT throughput on NVIDIA Blackwell GPUs using FP8 tensor cores and a new 'libKulisch' library.
Self-Recognition Finetuning can Prevent and Reverse Emergent Misalignment
Study on using self-generated text recognition (SGTR) finetuning to prevent and reverse emergent misalignment in large language models.
We're making Bunny DNS free: because a faster internet won't build itself
Bunny DNS is now offered for free to encourage a faster global internet.