A large-scale examination of inductive biases shaping high-level visual representation in brains and machines

Nature Communications, Год журнала: 2024, Номер 15(1)

Опубликована: Окт. 30, 2024

Язык: Английский

---

Distributed representations of behaviour-derived object dimensions in the human visual system

Nature Human Behaviour, Год журнала: 2024, Номер unknown

Опубликована: Сен. 9, 2024

Язык: Английский

---

NeuralDiffuser: Neuroscience-inspired Diffusion Guidance for fMRI Visual Reconstruction

IEEE Transactions on Image Processing, Год журнала: 2025, Номер 34, С. 552 - 565

Опубликована: Янв. 1, 2025

Reconstructing visual stimuli from functional Magnetic Resonance Imaging (fMRI) enables fine-grained retrieval of brain activity. However, the accurate reconstruction diverse details, including structure, background, texture, color, and more, remains challenging. The stable diffusion models inevitably result in variability reconstructed images, even under identical conditions. To address this challenge, we first uncover neuroscientific perspective methods, which primarily involve top-down creation using pre-trained knowledge extensive image datasets, but tend to lack detail-driven bottom-up perception, leading a loss faithful details. In paper, propose NeuralDiffuser incorporates primary feature guidance provide detailed cues form gradients. This extension process for achieves both semantic coherence detail fidelity when reconstructing stimuli. Furthermore, have developed novel strategy tasks that ensures consistency repeated outputs with original images rather than various outputs. Extensive experimental results on Natural Senses Dataset (NSD) qualitatively quantitatively demonstrate advancement by comparing it against baseline state-of-the-art methods horizontally, as well conducting longitudinal ablation studies. Code can be available https://github.com/HaoyyLi/NeuralDiffuser.

Язык: Английский

---

Self-supervision deep learning models are better models of human high-level visual cortex: The roles of multi-modality and dataset training size

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2025, Номер unknown

Опубликована: Янв. 13, 2025

Abstract With the rapid development of Artificial Neural Network based visual models, many studies have shown that these models show unprecedented potence in predicting neural responses to images cortex. Lately, advances computer vision introduced self-supervised where a model is trained using supervision from natural properties training set. This has led examination their prediction performance, which revealed better than supervised for with language or image-only supervision. In this work, we delve deeper into models’ ability explain representations object categories. We compare differed objectives examine they diverge predict fMRI and MEG recordings while participants are presented different Results both self-supervision was advantageous comparison classification training. addition, predictor later stages perception, shows consistent advantage over longer duration, beginning 80ms after exposure. Examination effect data size large dataset did not necessarily improve predictions, particular models. Finally, correspondence hierarchy each cortex showed image only conclude consistently recordings, type reveals property activity, language-supervision explaining onsets, explains long very early latencies response, naturally sharing corresponding hierarchical structure as brain.

Язык: Английский

---

Origins of food selectivity in human visual cortex

Trends in Neurosciences, Год журнала: 2025, Номер unknown

Опубликована: Фев. 1, 2025

Язык: Английский

---

A simple clustering approach to map the human brain's cortical semantic network organization during task

NeuroImage, Год журнала: 2025, Номер unknown, С. 121096 - 121096

Опубликована: Фев. 1, 2025

Constructing task-state large-scale brain networks can enhance our understanding of the organization functions during cognitive tasks. The primary goal network partitioning is to cluster functionally homogeneous regions. However, a region often serves multiple functions, complicating process. This study proposes novel clustering method for based on specific selecting semantic representation as target function evaluate validity proposed method. Specifically, we analyzed functional magnetic resonance imaging (fMRI) data from 11 subjects, each exposed 672 concepts, and correlated this with rating related these concepts. We identified distinct concept comprehension task validated robustness through methods. found that derived multidimensional activation exhibit high reliability cross-semantic model consistency (semantic ratings word embeddings extracted GPT-2), particularly in associated functions. Moreover, exhibits significant differences resting-state task-based obtained using traditional Further analysis revealed between networks, including disparities their capabilities, information modalities they rely acquire information, varying associations general domains. introduces approach analyzing tailored establishing standard parcellation seven future research, potentially enriching complex processes neural bases.

Язык: Английский

---

The Oomplet dataset toolkit as a flexible and extensible system for large-scale, multi-category image generation

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Март 18, 2025

The modern study of perceptual learning across humans, non-human animals, and artificial agents requires large-scale datasets with flexible, customizable, controllable features for distinguishing between categories. To support this research, we developed the Oomplet Dataset Toolkit (ODT), an open-source, publicly available toolbox capable generating 9.1 million unique visual stimuli ten feature dimensions. Each stimulus is a cartoon-like humanoid character, termed "Oomplet," designed to be instance within clearly defined categories that are engaging suitable use diverse groups, including children. Experiments show adults can four five dimensions as single classification criteria in simple discrimination tasks, underscoring toolkit's flexibility. With ODT, researchers dynamically generate large, novel sets biological contexts.

Язык: Английский

---

Partial information transfer from peripheral visual streams to foveal visual streams may be mediated through local primary visual circuits

NeuroImage, Год журнала: 2025, Номер unknown, С. 121147 - 121147

Опубликована: Март 1, 2025

Visual object recognition is driven through the what pathway, a hierarchy of visual areas processing features increasing complexity and abstractness. The primary cortex (V1), this pathway's origin, exhibits retinotopic organization: neurons respond to stimuli in specific field regions. A neuron responding central stimulus won't peripheral one, vice versa. However, despite organization, task-relevant feedback about can be decoded unstimulated foveal cortex, disrupting impairs discrimination behavior. information encoded by remains unclear, as prior studies used computer-generated objects ill-suited dissociate different representation types. To address knowledge gap, we investigated nature periphery-to-fovea using real-world stimuli. Participants performed same/different task on peripherally displayed images vehicles faces. Using fMRI multivariate decoding, found that both V1 could decode separated low-level perceptual models (vehicles) but not those semantic (faces). This suggests primarily carries information. In contrast, higher resolved semantically distinct images. functional connectivity analysis revealed connections later-stage areas. These findings indicate while early late may contribute transfer from streams, higher-to-lower area involve loss.

Язык: Английский

---

Language modulates vision: Evidence from neural networks and human brain-lesion models

Research Square (Research Square), Год журнала: 2025, Номер unknown

Опубликована: Март 27, 2025

Comparing information structures in between deep neural networks (DNNs) and the human brain has become a key method for exploring their similarities differences. Recent research shown better alignment of vision-language DNN models, such as CLIP, with activity ventral occipitotemporal cortex (VOTC) than earlier vision supporting idea that language modulates visual perception. However, interpreting results from comparisons is inherently limited due to "black box" nature DNNs. To address this, we combined model–brain fitness analyses lesion data examine how disrupting communication pathway systems causally affects ability vision–language DNNs explain VOTC. Across four diverse datasets, CLIP consistently outperformed both label-supervised (ResNet) unsupervised (MoCo) models predicting VOTC activity. This advantage was left-lateralized, aligning network. Analyses 33 stroke patients revealed reduced white matter integrity region left angular gyrus correlated decreased performance increased MoCo performance, indicating dynamic influence processing on These findings support integration modulation neurocognitive vision, reinforcing concepts models. The sensitivity similarity specific lesions demonstrates leveraging manipulation promising framework evaluating developing brain-like computer

Язык: Английский

---

Natural Language-Driven Data Visualization Using Kestrel AI: A Novel Algorithm for Intelligent Analytics

Опубликована: Апрель 8, 2025

Generative AI in financial analytics is increasingly vital for interpreting vast, complex datasets to guide strategic decisions dynamic, real-time markets, yet its technical complexity often limits non-expert engagement. Traditional methods, rooted structured query language (SQL), depend heavily on specialized expertise, restricting access business analysts and decision-makers lacking programming skills. These conventional systems achieve 75 percent 80 accuracy simple queries but falter significantly with complex, nested conditions or multi-operator logic, diminishing their utility today’s fast-paced landscape. We present Kestrel AI, a pioneering algorithm harnessing advanced Natural Language Processing (NLP) Retrieval-Augmented Generation (RAG) convert natural into precise SQL commands insightful visualizations. With modular, scalable architecture HighChart integration, delivers an outstanding 92 average accuracy, substantially outperforming traditional approaches across diverse types, from basic intricate. Experimental validation shows execution times averaging 0.3 seconds, highlighting speed efficiency, while user studies reveal 85 of participants, spanning seasoned complete novices, commend intuitive interface rapid insight delivery. The system adeptly processes unstructured data, allowing users blend thorough analysis, adapts seamlessly market shifts. By simplifying data interactions enhancing access, fosters inclusive, data-driven decision-making contexts. This work sets new standard generative AI-driven analytics, offering transformative tool investment research, organizational collaboration that bridges barriers user-friendly design.

Язык: Английский

---