Neural Networks

A developer has released a compact, hackable implementation of a generative pretrained transformer (GPT) model that operates on byte-level tokens, enabling it to process not only text but also DNA sequences, images, audi…

Large language models (LLMs) frequently exhibit miscalibration, where their reported confidence scores do not match actual accuracy rates, with studies showing mean calibration scores as low as 23.9% in biomedical models…

Developers can optimize spaCy's text processing speed by selectively disabling pipeline components like the dependency parser and lemmatizer when only named entity recognition is needed. The technique reduces processing …

OpenAI's GPT-3.5-Turbo model drops from 90% accuracy to 50% when the correct answer is placed in the middle of a 20,000-token prompt rather than at the start or end, according to research by Liu et al. (2023) published a…

RQM+ and Jordi Labs released Lumo, an in-silico tool that uses machine learning to predict LC-MS and GC-MS response factors for extractables and leachables analysis. The tool routes compounds to chemistry class-specific …

TorchRec's KeyedJaggedTensor, designed to efficiently combine sparse features in recommendation systems without padding, introduces GPU-CPU synchronization that degrades system performance. The data type's output shape c…

Researchers have developed a data-efficient flood depth prediction method that uses domain-aware coreset selection and tabular foundation models, achieving near-real-time accuracy with only 0.7% of the training data typi…

Researchers have developed ReGeN, a reference-guided generative pipeline that creates synthetic multivariate time series data by decomposing observed sequences into periodic backbones, stochastic residuals, and cross-var…

A study of behavioral foundation models trained on user event sequences found that a small embedder using only 2% of parameters is compute-optimal across all tested budgets, as embedder parameters are more expensive per …

Researchers have developed DiffSlack, a differentiable projection layer that enforces nonlinear inequality constraints in neural networks by reformulating them as equalities with learnable slack variables. Tested on vehi…

Researchers introduced trajectory extrapolation error, a measure of how much a language model's hidden states deviate from a linear path during word processing, and found it independently predicts human reading times bey…

Researchers have developed the Multi-Granularity Reasoning Network (MGRN), a new framework for natural language inference that explicitly leverages hierarchical semantic features across multiple levels of language unders…

Researchers introduced generic triple-latent sequence models that maintain a running token state and compressed pair-memory pathway to capture higher-order token interactions without benchmark-specific parsing. The tripl…

Researchers at arXiv propose a hybrid pre-training objective for language models that combines a Joint Embedding Predictive Architecture (JEPA) latent-space prediction loss with masked language modelling (MLM). Pre-train…

Researchers have developed a framework for unpaired RGB-thermal novel view synthesis that uses a 3D feed-forward transformer to independently estimate camera poses for each modality, then aligns them with the Procrustes …

Researchers have demonstrated that human preference scores for text-to-image content can be accurately predicted before generation begins, using minimal computational overhead. The study found that predicting these score…

Researchers have demonstrated that formal concept lattices, derived from Formal Concept Analysis, can serve as principled semantic scaffolds to guide neural network learning by organizing concepts hierarchically from gen…

Researchers have developed DFPL, a framework for incomplete image-tabular classification that addresses the missing-modality problem through fine-grained prototype learning. The method introduces Shared-Specific Prototyp…

Researchers have developed a deep learning algorithm that restores three-dimensional capillary anatomy from a single optical coherence tomographic angiography (OCTA) volume, addressing imaging artifacts that hinder relia…

Researchers have developed LightVesselNet, a neural network with only 75,000 parameters that performs retinal blood vessel segmentation competitively with much larger models. The ultra-lightweight architecture, which use…