Human EEG and artificial neural networks reveal disentangled representations of object real-world size in natural images DOI Open Access
Zitong Lu, Julie D. Golomb

Опубликована: Июль 11, 2024

Remarkably, human brains have the ability to accurately perceive and process real-world size of objects, despite vast differences in distance perspective. While previous studies delved into this phenomenon, distinguishing from other visual perceptions, like depth, has been challenging. Using THINGS EEG2 dataset with high time-resolution brain recordings more ecologically valid naturalistic stimuli, our study uses an innovative approach disentangle neural representations object retinal perceived depth a way that was not previously possible. Leveraging state-of-the-art dataset, EEG representational similarity results reveal pure representation brains. We report timeline processing: appeared first, then size, finally, size. Additionally, we input both these images object-only without natural background artificial networks. Consistent findings, also successfully disentangled all three types networks (visual-only ResNet, visual-language CLIP, language-only Word2Vec). Moreover, multi-modal comparison framework across reveals as stable higher-level dimension space incorporating semantic information. Our research provides detailed clear characterization processing process, which offers further advances insights understanding construction brain-like models.

Язык: Английский

“Identifying and characterizing scene representations relevant for categorization behavior” DOI Creative Commons
Johannes Singer, Agnessa Karapetian, Martin N. Hebart

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2023, Номер unknown

Опубликована: Авг. 18, 2023

1. Abstract Scene recognition is a core sensory capacity that enables humans to adaptively interact with their environment. Despite substantial progress in the understanding of neural representations underlying scene recognition, relevance these for behavior given varying task demands remains unknown. To address this, we aimed identify behaviorally relevant representations, characterize them terms visual features, and reveal how they vary across different tasks. We recorded fMRI data while human participants viewed scenes linked brain responses three tasks acquired separate sessions: manmade/natural categorization, basic-level fixation color discrimination. found correlations between categorization response times scene-specific responses, quantified as distance hyperplane derived from multivariate classifier. Across tasks, effects were largely distinct parts ventral stream. This suggests are depending on task. Next, using deep networks proxy feature early/intermediate layers mediated relationship both indicating contribution low-/mid-level features representations. Finally, observed opposite patterns brain-behavior task, interference do not align content Together, results spatial extent, content, task-dependence mediate complex scenes.

Язык: Английский

Процитировано

4

Decoding electroencephalographic responses to visual stimuli compatible with electrical stimulation DOI Creative Commons
Simone Romeni, Laura Toni, Fiorenzo Artoni

и другие.

APL Bioengineering, Год журнала: 2024, Номер 8(2)

Опубликована: Июнь 1, 2024

Electrical stimulation of the visual nervous system could improve quality life patients affected by acquired blindness restoring some sensations, but requires careful optimization parameters to produce useful perceptions. Neural correlates elicited perceptions be used for fast automatic optimization, with electroencephalography as a natural choice it can non-invasively. Nonetheless, its low signal-to-noise ratio may hinder discrimination similar patterns, preventing use in electrical stimulation. Our work investigates first time discriminability electroencephalographic responses stimuli compatible stimulation, employing newly dataset whose encompass concurrent variation several features, while neuroscience research tends study neural single features. We then performed above-chance single-trial decoding multiple features our crafted using relatively simple machine learning algorithms. A scheme information from stimulus presentations was implemented, substantially improving performance, suggesting that such methods should systematically future applications. The significance present relies determination which decoded stimulation-compatible and at granularity they discriminated. pave way optimize parameters, thus increasing effectiveness current neuroprostheses.

Язык: Английский

Процитировано

1

Contribution of low-level image statistics to EEG decoding of semantic content in multivariate and univariate models with feature optimization DOI Creative Commons

Eric Lützow Holm,

Diego Fernández Slezak, Enzo Tagliazucchi

и другие.

NeuroImage, Год журнала: 2024, Номер 293, С. 120626 - 120626

Опубликована: Апрель 25, 2024

Spatio-temporal patterns of evoked brain activity contain information that can be used to decode and categorize the semantic content visual stimuli. However, this procedure biased by low-level image features independently present in stimuli, prompting need understand robustness different models regarding these confounding factors. In study, we trained machine learning distinguish between concepts included publicly available THINGS-EEG dataset using electroencephalography (EEG) data acquired during a rapid serial presentation paradigm. We investigated contribution decoding accuracy multivariate model, utilizing broadband from all EEG channels. Additionally, explored univariate model obtained through data-driven feature selection applied spatial frequency domains. While exhibited better accuracy, their predictions were less robust effect statistics. Notably, some maintained even after random replacement training with semantically unrelated samples presented similar content. conclusion, our findings suggest optimization impacts sensitivity factors, regardless resulting classification performance. Therefore, choice for should ideally informed criteria beyond classifier performance, such as neurobiological mechanisms under study.

Язык: Английский

Процитировано

1

An easy‐to‐follow handbook for electroencephalogram data analysis with Python DOI Creative Commons
Zitong Lu,

Wanru Li,

Lu Nie

и другие.

Brain‐X, Год журнала: 2024, Номер 2(2)

Опубликована: Июнь 1, 2024

Abstract This easy‐to‐follow handbook offers a straightforward guide to electroencephalogram (EEG) analysis using Python, aimed at all EEG researchers in cognitive neuroscience and related fields. It spans from single‐subject data preprocessing advanced multisubject analyses. contains four chapters: Preprocessing Single‐Subject Data, Basic Python Data Operations, Multiple‐Subject Analysis, Advanced Analysis. The chapter provides standardized procedure for preprocessing, primarily the MNE‐Python package. Operations introduces essential operations handling, including reading, storage, statistical analysis. Analysis guides readers on performing event‐related potential time‐frequency analyses visualizing outcomes through examples face perception task dataset. explores three methodologies, Classification‐based decoding, Representational Similarity Inverted Encoding Model, practical visual working memory dataset NeuroRA other powerful packages. We designed our easy comprehension be an tool anyone delving into with (GitHub website: https://github.com/ZitongLu1996/Python‐EEG‐Handbook ; For Chinese version: https://github.com/ZitongLu1996/Python‐EEG‐Handbook‐CN ).

Язык: Английский

Процитировано

1

Human EEG and artificial neural networks reveal disentangled representations of object real-world size in natural images DOI Open Access
Zitong Lu, Julie D. Golomb

Опубликована: Июль 11, 2024

Remarkably, human brains have the ability to accurately perceive and process real-world size of objects, despite vast differences in distance perspective. While previous studies delved into this phenomenon, distinguishing from other visual perceptions, like depth, has been challenging. Using THINGS EEG2 dataset with high time-resolution brain recordings more ecologically valid naturalistic stimuli, our study uses an innovative approach disentangle neural representations object retinal perceived depth a way that was not previously possible. Leveraging state-of-the-art dataset, EEG representational similarity results reveal pure representation brains. We report timeline processing: appeared first, then size, finally, size. Additionally, we input both these images object-only without natural background artificial networks. Consistent findings, also successfully disentangled all three types networks (visual-only ResNet, visual-language CLIP, language-only Word2Vec). Moreover, multi-modal comparison framework across reveals as stable higher-level dimension space incorporating semantic information. Our research provides detailed clear characterization processing process, which offers further advances insights understanding construction brain-like models.

Язык: Английский

Процитировано

1