Patterns最新文献

Quaternions, a type of hypercomplex number, can be applied to handling three-dimensional data, i.e., color images. Here, we demonstrate, by leveraging quaternions and the two-dimensional orthogonal planes split framework, image processing workflows for natural and biomedical images, including natural and biomedical image recolorization, natural image decolorization, natural and biomedical image contrast enhancement, and computational restaining and stain separation in histological images. We also demonstrate performance gains in machine learning and deep learning pipelines for histological images. The proposed workflows can regulate color appearance and image contrast, be part of automated processing pipelines, and assist in digital pathology applications. Employing basic arithmetic and matrix operations, this work offers a computationally accessible methodology that showcases versatility and consistency across processing tasks and a range of computer vision and biomedical applications. The proposed non-data-driven methods achieve comparable or better results to those reported in the literature, showcasing the potential of robust theoretical frameworks with practical effectiveness.

In drug design, transformer networks adopted from natural language processing are applied in a variety of ways. We have used sequence-based generative compound design as a model system to explore the learning characteristics of transformers and determine if these models learned information relevant for protein-ligand interactions. The analysis reveals that sequence-based predictions of active compounds using transformer models required a proportion of at least ∼60% of the original test sequences. Moreover, predictions depended on sequence and compound similarity of training and test data and on compound memorization effects. The predictions were purely statistically driven by associating sequence patterns with molecular structures, thus rationalizing their strict dependence on detectable similarities. Moreover, the transformer models did not learn target sequence information relevant for ligand binding. While the results do not call sequence-based compound design approaches generally into question, they caution against over-interpretation of transformer models used for such applications.

Patch-seq enables the integration of electrophysiological recordings, single-cell RNA sequencing (scRNA-seq), and morphological reconstruction within the same neuron, but establishing mechanistic links between transcriptomic and physiological properties remains a major challenge. Bernaerts et al.¹ developed a new statistical-biophysical model based on biophysical simulations and modern machine learning techniques. They applied this model to gene expression and established a quantitative link between gene expression and electrophysiological activity patterns. This work is an important advance toward closing the gap between gene expression and neuronal physiology.

Generative AI technologies are creating both new challenges and opportunities for educators around the world. In this People of Data piece, we asked two of the journal's advisory board members to share how they are using generative AI technologies in teaching and their views about the future of AI in education.

As artificial intelligence matures, the impact it might have on how society functions is being actively pondered. In this opinion, through uniform-binomial mixtures, the author sheds quantitative light on the matter, showing topics that unite and divide the population on an unobserved, latent level.

Fragmented AI tools struggle with the complex diagnosis of fine-grained diseases from radiological images. A new study in Patterns introduces the "screening-to-subtyping" (S2S) framework, a holistic deep learning system integrating the entire diagnostic workflow from detection to subtyping. Validated on complex thoracic cancers, the S2S-Med system demonstrated superior accuracy, outperforming existing benchmarks. A human-AI experiment revealed that the AI's performance surpassed AI-assisted physicians and that physician trust correlated with greater improvement. The S2S framework is a significant step toward enhancing precision medicine and establishing a new paradigm for human-AI partnership.

Language has long been an essential tool for human reasoning. The rise of large language models (LLMs) has led to research on their application in complex reasoning tasks. Researchers are exploring the concept of "thought," which represents intermediate reasoning steps, allowing LLMs to emulate humanlike reasoning processes. Recent work has applied reinforcement learning (RL) to train LLMs by searching for high-quality reasoning trajectories through trial-and-error exploration. In parallel, studies also demonstrate that allowing LLMs to "think" with longer chains of intermediate tokens at test time can also substantially improve reasoning accuracy. The combination of training and test-time advancements outlines a path toward large reasoning models. This survey reviews recent progress in LLM reasoning. It covers foundational concepts behind LLMs and the key technical components that contribute to the development of large reasoning models, and it highlights popular open-source projects for building these models. The survey concludes by discussing ongoing challenges and future research directions in this field.

This paper systematically analyzes the adoption of large language models (LLMs), such as ChatGPT, across consumer complaints, corporate press releases, job postings, and United Nations (UN) press releases, covering extensive datasets from January 2022 to September 2024. By late 2024, roughly 18% of financial consumer complaints, 24% of corporate press releases, nearly 10% of job postings in small firms, and 14% of UN press releases involve LLM-assisted writing. Adoption surged rapidly post-ChatGPT release but stabilized by 2024, highlighting generative artificial intelligence (AI)'s broad societal impact and its widespread use across sectors.

Visual quality assessment (VQA) is indispensable in multimedia for evaluating algorithm effectiveness and optimizing systems, yet its neurobiological mechanisms remain poorly understood. Using functional magnetic resonance imaging (fMRI), we investigate how the brain processes varying image qualities, revealing specialized mechanisms for handling low-quality stimuli. Results show that low quality significantly impacts semantic encoding along the visual pathway: low-level regions exhibit only 35.20% of the semantic information seen in high-quality condition, while higher-level regions compensate adaptively to maintain understanding. Visual quality is not locally encoded but emerges from inter-regional information gaps, with perception arising from this hierarchical discrepancy. Leveraging this compensatory mechanism, we decode quality from fMRI and propose a neural network feature fusion strategy, boosting ResNet's VQA performance by 14.29% on the BID dataset (586 instances). Our findings provide neurobiological evidence for degraded visual processing, addressing a gap in perception neuroscience and offering theoretical foundations for improving VQA models.

Sleep staging is essential for understanding sleep physiology and diagnosing sleep-related disorders. However, traditional manual scoring is time-consuming and resource intensive, limiting its scalability for large-scale application. In this study, we introduce AISleep, an automated and interpretable unsupervised algorithm based on feature-weighted kernel density estimation (KDE), designed to stage sleep using only a single electroencephalogram (EEG) channel. AISleep was evaluated using both public benchmark datasets of healthy subjects and clinical datasets of patients with sleep disorders. It outperforms state-of-the-art (SOTA) unsupervised sleep staging algorithms in young, healthy subjects and demonstrates better generalizability compared to supervised models. Importantly, we observed that some key EEG features decline with age, which may contribute to reduced staging accuracy in older adults. This study presents a robust and interpretable unsupervised sleep staging algorithm with a lightweight design that makes it well suited to integration into portable devices, offering a practical and scalable solution for accurate, home-based sleep monitoring.