Modeling attention and binding in the brain through bidirectional recurrent gating.
8 blocks of code here have a permanent identifier (oscr:b:…), listed with the matches below: citable for good, with a page of its own. Select a match and press Cite to copy it. What a block identifier is.
The 8 matches
- [1] § Methods › Evaluation › Bell-curve fitting ↔ src/curves_utils.py, lines 236 to 271 · score 0.86 · Euclidean Distance, half width, preferred orientation, 179 deg, bell, asymptote · block
oscr:b:MV726Q4XYE2Y2-8· Report a mismatch - [2] § Methods › Model ↔ src/modelv2.py, lines 182 to 280 · score 0.85 · AdaptiveAvgPool2d, MaxPool2d, ConvTranspose2d, Conv2d, Unflatten, Concat · block
oscr:b:MWPP42B76T0J8-4· Report a mismatch - [3] § Methods › Model ↔ src/resatt_.py, lines 107 to 180 · score 0.83 · AdaptiveAvgPool2d, MaxPool2d, ConvTranspose2d, Conv2d, Unflatten, UpSample · block
oscr:b:0QYXHXHYJKP81-M· Report a mismatch - [4] § Methods › Optimization ↔ main_mmshape.py, lines 98 to 123 · score 0.62 · weight decay, OneCycleLR, Cosine, Adam, scheduling, optimizer · block
oscr:b:VGFBJ9MWY229Z-H· Report a mismatch - [5] § Methods › Evaluation ↔ src/conductor.py, lines 328 to 432 · score 0.61 · cross entropy, pixel error, class weighting, validation, loss, accuracy · block
oscr:b:40TRJ2ABSVKMT-N· Report a mismatch - [6] § Methods › Optimization ↔ main_stl10.py, lines 163 to 185 · score 0.61 · weight decay, OneCycleLR, Cosine, Adam, scheduling, optimizer · block
oscr:b:QVMKRXV5VVY05-0· Report a mismatch - [7] § Results › Neurophysiological results › Attention-invariant tuning ↔ src/curves_utils.py, lines 155 to 160 · score 0.55 · McAdams, orientation tuning, Maunsell, unattended, curves, attending · block
oscr:b:D0DMG15ZKA45M-T· Report a mismatch - [8] § Methods › Optimization ↔ src/conductor.py, lines 80 to 198 · score 0.51 · cross entropy, class weights, sum, MSE, loss, Optimization · block
oscr:b:12XVQJ3ATG8JA-1· Report a mismatch
Paper
Loaded from Europe PMC by your browser; OSCR does not store it: doi.org · Europe PMC
The paper is loaded when this pane is shown.
The authors' code
Python · 310 lines · 12 KB · GPL-3.0 · 2 matches
- # # built-in modules
- import os
- import argparse
- from pprint import pformat
- from collections import OrderedDict
- import random
- # # Torch modules
- import torch
- # # internal imports
- from prelude import save_dicts, startup_folders, get_device, save_results_to_csv, load_dicts
- from src.composer import CurveTracing, transforms, conv2d, bezier_generator
- from src.model import AttentionModel
- from src.utils import plot_all, plot_loss_all, DataLoader
- from src.utils import build_loaders, get_n_parameters
- from src.conductor import AttentionTrain
- import matplotlib
- import numpy
- import math
- import matplotlib.pyplot as plt
- from matplotlib import cm
- from scipy.ndimage import gaussian_filter1d
- c = matplotlib.colormaps["tab10"]
- def gaussian(x, mu, std):
- """
- Gaussian normalized function
- """
- y = torch.exp( - ((x - mu) / std)**2 / 2 ) / (std * math.sqrt(2 * math.pi))
- return y
- def roll_to_mean(x):
- d = x.size(-1)
- argmean = torch.argmax(x, dim=-1).item()
- roll = -argmean + d//2 + d%2 - 1
- return torch.roll(x, roll, dims=-1), roll
- def try_gaussian_fit(x):
- d = x.numel()
- v = torch.arange(d)
- mean = torch.argmax(x).item()
- std = torch.argmin((torch.cumsum(x, dim=0) - 0.859 * x.sum()).square()).item() - mean
- return gaussian(v, mean, std)
- def does_it_fit(x):
- y = try_gaussian_fit(x)
- x = x / torch.linalg.norm(x)
- y = y / torch.linalg.norm(y)
- z = torch.dot(y, x)
- return z
- def parameters_of_orientation_tuning(x, k=180):
- a = torch.max(x).item()
- d = torch.argmax(x).item()
- b = (torch.argmin((torch.cumsum(x, dim=0) - 0.859 * x.sum()).square()).item() - d)/k
- c = torch.min(x).item()
- return a, b, c, d
- def scale_(x: torch.Tensor, dim=((-2, -1)), eps=1e-6):
- return x / (x.abs().flatten(start_dim=dim[0], end_dim=dim[1]).max(dim=-1, keepdim=True).values + eps)
- def gabor(h: float, w: float, sigma: float, theta: float, k: float):
- pi = torch.pi
- rh, rw = int(h//2), int(w//2)
- xx, yy = torch.meshgrid(torch.arange(-rh, rh), torch.arange(-rw, rw), indexing='ij')
- cc = torch.zeros(h, w)
- x = xx * math.cos(theta) + yy * math.sin(theta)
- y = -xx * math.sin(theta) + yy * math.cos(theta)
- gaussian = torch.exp(- sigma**2 / (8 * k**2) * (4 * x**2 + y**2)) * sigma**2 / (4*pi * k**2)
- sinusoid = torch.cos(sigma * x) * math.exp(k**2 / 2)
- stimulus = torch.clamp(scale_(gaussian * sinusoid), 0.0, 1.0)
- cc[rh-1:rh+1, rw-1:rh+1] = 1.0
- return stimulus.unsqueeze(0), cc.unsqueeze(0)
- def make_bar(h: float, w: float, theta: float, k = 19, sigma: float = 3.0):
- if k > 0 and sigma > 0:
- blur = transforms.GaussianBlur(k, sigma=(sigma, sigma))
- else:
- blur = lambda x: x
- p = min(h//2, w//2)
- pp = p//4
- x = torch.zeros(1, h, w)
- c = torch.zeros(1, h, w)
- x[:, p-2:p+1, pp:3*pp+p] = 1.0
- c[:, p-3:p+2, pp-2:pp+3] = 1.0
- x = transforms.functional.rotate(x, theta, interpolation=transforms.InterpolationMode.BILINEAR)
- c = transforms.functional.rotate(c, theta, interpolation=transforms.InterpolationMode.BILINEAR)
- x = blur(x)
- c = blur(c)
- x = scale_(x, dim=(-2, -1))
- c = scale_(c, dim=(-2, -1))
- return x, c
- def generate_bezier_rot(b: int, h: float, w: float, ds: CurveTracing):
- radius = 0.3
- bezier_order = 1
- resolution = ds.resolution
- stimulus_kernel = ds.stimulus_kernel
- fixate_kernel = ds.fixate_kernel
- saccade_kernel = ds.saccade_kernel
- samples = torch.zeros(3, b, 1, h, w)
- t_ = torch.linspace(0., 1.0, resolution) # parameter
- c_ = torch.rand(b, bezier_order + 1, 2, 1) # coordinates
- for j in range(b):
- c_[j, 0, 0] = 0.5 + radius * math.cos(math.pi * j / b)
- c_[j, 0, 1] = 0.5 + radius * math.sin(math.pi * j / b)
- c_[j, 1, 0] = 0.5 - radius * math.cos(math.pi * j / b)
- c_[j, 1, 1] = 0.5 - radius * math.sin(math.pi * j / b)
- for i in range(b):
- im_ = torch.zeros(3, 1, h, w)
- b_ = bezier_generator(c_[i], t_) # Bezier curve
- b_h, b_w = (b_[0] * h).int().tolist(), (b_[1] * w).int().tolist()
- im_[0, :, b_h, b_w] = 1.0
- im_[1, :, b_h[0], b_w[0]] = 1.0
- im_[2, :, b_h[-1], b_w[-1]] = 1.0
- im_[0] = (conv2d(im_[0], stimulus_kernel, padding='same') > 0.0)
- im_[1] = (conv2d(im_[1], fixate_kernel, padding='same') > 0.0)
- im_[2] = (conv2d(im_[2], saccade_kernel, padding='same') > 0.0)
- samples[:, i] = im_
- return samples
- def make_stimuli(h: int, w: int, b: int, ds:CurveTracing , bar: bool = False):
- assert h%4 == 0 and w%4 == 0
- x = torch.zeros(b, 1, h, w)
- xc = torch.zeros(b, 1, h, w)
- zc = torch.zeros(b, 1, h, w)
- t = torch.zeros(b, 1, h, w)
- theta = torch.linspace(0., 179., b)
- better_bars = generate_bezier_rot(b, h//2, w//2, ds)
- for i in range(b):
- if bar:
- xx, cc = better_bars[0, i], better_bars[1, i] # make_bar(h//2, w//2, theta[i].item())
- else:
- xx, cc = gabor(h//2, w//2, torch.pi/4, theta[i].item()*torch.pi/180.0, 2.0)
- x[i, :, :h//2, :w//2] = xx
- x[i, :, -h//2:, -w//2:] = xx
- xc[i, :, :h//2, :w//2] = cc
- zc[i, :, -h//2:, -w//2:] = cc
- t[i, :, :h//2, :w//2] = xx
- return x, xc, zc, t
- def parameters_of_orientation_tuning(x, n=180):
- a, d = x.max(dim=-1)
- b = (torch.argmin((torch.cumsum(x, dim=-1) - 0.859 * x.sum(dim=-1, keepdim=True)).square(), dim=-1) - d)/n
- c = torch.min(x, dim=-1).values
- mu = x.mean(dim=-1)
- md = x.median(dim=-1).values
- return a, b, c, d, mu, md
- def modulation_index(a: torch.Tensor, u: torch.Tensor):
- """
- attentional modulation index: (attended - unattended)/(attended + unattended)
- ref: McAdams and Maunsell · Effects of Attention on Orientation Tuning
- """
- return torch.nan_to_num((a - u) / (a + u), nan=0.0, posinf=0.0, neginf=0.0)
- def normed_mse_loss(predictions: torch.Tensor, targets: torch.Tensor, reduction: str = 'mean', donorm: bool = True) -> float:
- if donorm:
- predictions = (predictions + 1.0) / 2.0
- targets = (targets + 1.0) / 2.0
- corrects = (predictions - targets).square().sum()
- return corrects / targets.numel() if reduction == 'mean' else corrects * targets.size(0) / targets.numel()
- def roelf_accuracy(y_t: torch.Tensor, y_d: torch.Tensor, y_p: torch.Tensor):
- y_tp = (y_t * y_p).sum(dim=(-3, -2, -1))
- y_dp = (y_d * y_p).sum(dim=(-3, -2, -1))
- return (y_tp > y_dp).sum().item() / y_t.size(0)
- def get_rec_field_act(model: AttentionModel, x: torch.Tensor, e: float = 1e-3):
- # x is the stimulus or receptive field as (batch, channel, h, w)
- # get the receptive field
- list_ind = [[] for _ in range(model.n_convs)]
- base_act = [[] for _ in range(model.n_convs)]
- for i in range(model.n_convs):
- conv = model.conv_blocks[i].conv
- pool = model.conv_blocks[i].pool if model.conv_blocks[i].pool is not None else lambda x: x
- fun = model.conv_blocks[i].fun
- x = pool(fun(conv(x)))
- base_act[i] = x
- list_ind[i] = (x.abs() > e)
- return list_ind, base_act
- def get_activity(model: AttentionModel, x: torch.Tensor):
- if x.ndim == 4: # (batch, channel, h, w)
- x = x.unsqueeze(1) # (batch, n_iter, channel, h, w)
- batch_size, n_iter = x.size(0), x.size(1)
- all_acts = []
- model.eval()
- with torch.no_grad():
- model.initiate_forward(batch_size=batch_size)
- for i in range(n_iter):
- *_, act_ = model.one_forward(x[:, i])
- all_acts.append(act_)
- return all_acts
- def polar_plot(xt, xd, theta_res, results_folder, plotname):
- theta = torch.linspace(0, 2*torch.pi, 2*theta_res)
- fig, ax = plt.subplots(figsize=(3, 2), subplot_kw={'projection': 'polar'})
- r_max = max(xt.max().item(), xd.max().item())
- ax.plot(theta, torch.cat([xt, xt])/r_max, c='#E91EF9')
- ax.plot(theta, torch.cat([xd, xd])/r_max, c='#07BAFC')
- ax.set_rmax(1.05)
- # ax.set_rticks([0.25, 0.5, 0.75]) # Less radial ticks
- # ax.set_xticks(torch.linspace(0, 2*torch.pi, 8)) # Less radial ticks
- # ax.set_xticklabels(["$0^\circ$", None, "$90^\circ$", None, "$180^\circ$", None, "$270^\circ$", None]) # Less radial ticks
- # ax.set_xticklabels([0, 90, 180, 270]) # Less radial ticks
- # ax.set_yticklabels([]) # Less radial ticks
- # ax.set_rlabel_position(0.0) # Move radial labels away from plotted line
- ax.grid(True)
- # ax.set_title(f"Scaled response of Cell {c} Layer {i} ", va='bottom')
- # if argus.res_fold is not None:
- plt.savefig(os.path.join(results_folder, 'Tuning_Curve' + plotname + '.svg'))
- plt.close()
- def plot_curves(n_layers, curve_tar_act, curve_dis_act, results_folder = None, plotname = None):
- for j in range(n_layers):
- plt.figure(figsize=(6, 4))
- # mean = torch.cat([mean_tar_act[j][:2], mean_dis_act[j][:2]]).mean().cpu()
- plt.plot(curve_tar_act[j].detach().cpu(), color="r")
- plt.plot(curve_dis_act[j].detach().cpu(), color="b")
- plt.title(f"Layer {j}")
- if results_folder is None or plotname is None:
- plt.show()
- else:
- plt.savefig(os.path.join(results_folder, f"{plotname}_{j}.svg"), format="svg")
- plt.close()
- class FitBellCurve:
- def __init__(self):
- self.d = 180
- self.x = torch.linspace(0, 179, self.d)
- self.z = torch.ones_like(self.x) * 0.5
- def roll(self, y: torch.Tensor, b: int):
- h = self.d//2
- return y.roll(h - b)
- def normal(self, mu: float, std: float):
- # Normalized Gaussian function
- return torch.exp(-((self.x-mu)**2)/(2*std**2)) / (std * math.sqrt(2*math.pi))
- def mean_euc(self, y_t: torch.Tensor, y_p: torch.Tensor):
- # Calculate the mean euclidean distance between two curves
- return (y_t - y_p).square().sum().sqrt() / self.d
- def __call__(self, y: torch.Tensor):
- d = y.min().item() # Asymptote (baseline)
- y = y - d # Remove the asymptote
- a = y.max().item() # Amplitude (peak)
- y = y / a # Normalize the amplitude
- b = y.argmax().item() # Preferred orientation (peak position)
- y = self.roll(y, b) # center the curve on the preferred orientation
- idx = torch.argwhere((y - self.z).sign().diff())
- # Find the half width at half maximum to calculate the standard deviation
- if idx.numel() > 1:
- c = (idx[1].item() - idx[0].item()) / (2 * math.sqrt(2 * math.log(2)))
- else:
- c = self.d//2
- yy = self.normal(self.d//2, c) # Generate a normal curve for the given width
- yy = yy / yy.max() # give the fitted curve the same amplitude as the original
- yy = yy + d # Add same baseline to the fitted curve
- e = self.mean_euc(y, yy) # Calculate the mean euclidean distance between the curves
- return (a, b, c, d), e
- class Roelfsema:
- def __init__(self, model, tasks, logger):
- import matplotlib.pyplot as plt
- self.task_name = 'CurveTracing' if 'CurveTracing' in tasks else 'SwitchBox'
- self.model = model
- self.tasks = tasks
- self._loader = tasks[self.task_name]["dataloaders"][-1]
- self._loader.dataset.training = False
- self.n_samples = len(self._loader.dataset)
- self.logger = logger
- fix_attend_saccade = self.tasks[self.task_name]["params"]["fix_attend_saccade"]
- n_iter = sum(fix_attend_saccade)
- n_layers = model.n_convs
- n_fix, n_att, n_sac = fix_attend_saccade
- n_fix_att = n_fix + n_att
- def test_accuracy_curve(self, device):
- correct = 0
- threshold = 0.25
- self._loader = self.tasks[self.task_name]["dataloaders"][-1]
- self._loader.dataset.training = False
- self.model.to(device)
- self.model.eval()
- with torch.no_grad():
- for x, _, m, _, c in self._loader:
- b = x.size(0)
- x, m, c = x.to(device), m.to(device), c.to(device)
- p_m, _, _ = self.model(x, self.tasks[self.task_name]["key"])
- m_correct = (p_m[:, -1] * c[:, 2])
- m_false = (p_m[:, -1] * c[:, 5])
- m_correct = m_correct.view(b, -1).sum(dim=-1) / c[:, 2].view(b, -1).sum(dim=-1)
- m_false = m_false.view(b, -1).sum(dim=-1) / c[:, 5].view(b, -1).sum(dim=-1)
- correct += ((m_correct - m_false) > threshold).sum().item()
- self.logger.info(f"Test Final Accuracy: {correct / self.n_samples}")
- self._loader.dataset.training = True
curves_utils.py at commit 9c26438, under GPL-3.0 · at the source
Overview
- Machine Learning Group, Technical University of Berlin, Berlin, Germany
- BIFOLD—Berlin Institute for the Foundations of Learning and Data, Berlin, Germany
- BCCN—Bernstein Center for Computational Neuroscience, Berlin, Germany
- New York University, New York, NY USA
- Cold Spring Harbor Laboratory, Cold Spring Harbor, NY USA
- Department of Artificial Intelligence, Korea University, Seoul, Korea
- Max Planck Institute for Informatics, Saarbrücken, Germany
- Department of Bioengineering, Department of Neuroscience, University of Pennsylvania, Philadelphia, PA USA
- CIFAR LMB Program, Toronto, ON Canada
Abstract
Attention is a cornerstone of cognition and neural computation, enabling the brain to select relevant information, bind features into coherent objects, and guide behavior. However, we currently lack a unifying computational model that connects the diverse phenomena of attention, from spatial and feature-based selection to object-based binding, within a single, neurally plausible computational framework. Here, we propose a bidirectional recurrent gating mechanism integrated into a principled architecture of the ventral visual stream. In this architecture, feedforward pathways extract visual features, while top-down and lateral connections transmit context- and task-dependent modulatory signals that control information flow. We demonstrate that our model, trained on recognition and segmentation problems, successfully performs the canonical attention tasks of orienting, filtering, and visual search on complex scenes. It replicates key psychophysical phenomena, such as perceptual load and inattentional blindness, while its internal units develop neural properties consistent with primate physiology, including multiplicative gain modulation and border-ownership coding. Our work provides evidence that this diverse set of attentional and binding phenomena can emerge from error-backpropagation combined with architectural constraints upon information flow, offering a powerful tool for neuroscience and a compelling, bio-inspired alternative to standard AI architectures.
Reproduced under the paper's license (CC BY), from the paper cited above.
Repository
Its files are read in the Code ↔ Paper reader above, with 8 matches between paragraphs and lines of code.
ssnio/bio-attention
9c2643861c5bf67195da3bbc5ae42ac57ed64cbc, 29 January 2026Availability: 1 check, the latest on 28 September 2026: the link answers
- 28 September 2026: the link answers
32 files
- demo_coco.py, Python, 259 lines
- main_celeba.py, Python, 143 lines
- main_coco.py, Python, 186 lines
- main_curve_tracing.py, Python, 138 lines
- main_mmshape.py, Python, 155 lines, 1 match
- main_mnist.py, Python, 210 lines
- main_psycphys_change.py, Python, 153 lines
- main_psycphys_compare.py
, Python, 160 lines - main_psycphys_contrast.p
y , Python, 161 lines - main_stl10.py, Python, 213 lines, 1 match
- notebooks/
bregman_notebook.ipynb , Jupyter, 133 lines - notebooks/
celeba_notebook.ipynb , Jupyter, 134 lines - notebooks/
coco_notebook.ipynb , Jupyter, 111 lines - notebooks/
curves_notebook.ipynb , Jupyter, 583 lines - notebooks/
mm_shapes_rec_res.ipynb , Jupyter, 104 lines - notebooks/
mnist_notebook.ipynb , Jupyter, 111 lines - notebooks/
psycphys_change_notebook , Jupyter, 162 lines.ipynb - notebooks/
psycphys_contrast_notebo , Jupyter, 153 linesok.ipynb - notebooks/
psycphys_discrimination_ , Jupyter, 156 linesnotebook.ipynb - notebooks/
shapes_bo.ipynb , Jupyter, 322 lines - notebooks/
stl10_rsa_notebook.ipynb , Jupyter, 179 lines - prelude.py, Python, 130 lines
- src/
composer.py , Python, 3,825 lines - src/
conductor.py , Python, 1,043 lines, 2 matches - src/
curves_utils.py , Python, 310 lines, 2 matches - src/
model.py , Python, 546 lines - src/
model_r.py , Python, 543 lines - src/
modelv2.py , Python, 454 lines, 1 match - src/
resatt_.py , Python, 314 lines, 1 match - src/
utils.py , Python, 227 lines - LICENSE, License, 674 lines
- README.md, Text, 62 lines
Code availability
We used Python (version 3.9) as the primary programming language. The following libraries (modules) were used and are required to execute the analysis: PyTorch, NumPy, SciPy, Matplotlib, and Pillow. For the COCO experiment, the pycocotools module is necessary. The source code for all experiments and corresponding result notebooks are available on GitHub: https://
Reproduced under the paper's license (CC BY), from the paper cited above.
Tracing map
Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.
What the map holds:
- 1 repository of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
- 30 scripts, each with its path and the digest of its content;
- 8 matches between paragraphs of the paper and lines of the code (method lexical-v1);
- neither the text of the paper nor the code itself.
Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.
Data
No dataset and no data link were found in the paper.
Data Availability Statement
All datasets are either publicly available on their corresponding online repositories or the code to generate and compose them is on GitHub (https://
We used Python (version 3.9) as the primary programming language. The following libraries (modules) were used and are required to execute the analysis: PyTorch, NumPy, SciPy, Matplotlib, and Pillow. For the COCO experiment, the pycocotools module is necessary. The source code for all experiments and corresponding result notebooks are available on GitHub: https://
Reproduced under the paper's license (CC BY), from the paper cited above.
Versions
The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.
Version 2, 28 September 2026
- Funding: added Deutsche Forschungsgemeinschaft: 01GQ0850, 01GQ1115, 01IS14013A-E; Bundesministerium für Bildung und Forschung; Korea University: 2019-0-00079, 2022-0-00984; Ministry of Science and ICT, South Korea: 2022-0- 00984, 01GQ1115, 01GQ0850, 01IS18037A, 01IS18025A, 2019-0-00079; Institute for Information and Communications Technology Promotion: 2019–0–00079, 2022, 2022-0-00984
Version 1, 28 September 2026: the first record
Recorded: type, language, journal, volume, issue, pages, dates, 5 authors, 3 keywords, 9 MeSH terms, 118 references.
Cite
This paper
Salehi, S., Lei, J., Benjamin, A. S., Müller, K.-R., & Kording, K. P. (2026). Modeling attention and binding in the brain through bidirectional recurrent gating. Nature communications, 17(1), 4072. https://
BibTeX
@article{salehi2026model
author = {Salehi, Saeed and Lei, Jordan and Benjamin, Ari S and Müller, Klaus-Robert and Kording, Konrad P},
title = {{Modeling attention and binding in the brain through bidirectional recurrent gating}},
journal = {Nature communications},
year = {2026},
month = may,
volume = {17},
number = {1},
pages = {4072},
publisher = {Nature Publishing Group},
issn = {2041-1723},
doi = {10.1038/
url = {https://
pmid = {42086575},
pmcid = {PMC13144683}
}
RIS
TY - JOUR
AU - Salehi, Saeed
AU - Lei, Jordan
AU - Benjamin, Ari S
AU - Müller, Klaus-Robert
AU - Kording, Konrad P
TI - Modeling attention and binding in the brain through bidirectional recurrent gating
T2 - Nature communications
J2 - Nat Commun
PY - 2026
DA - 2026/
VL - 17
IS - 1
SP - 4072
SN - 2041-1723
PB - Nature Publishing Group
DO - 10.1038/
UR - https://
LA - en
ER -
CSL-JSON
{
"id": "10.1038/
"type": "article-journal",
"title": "Modeling attention and binding in the brain through bidirectional recurrent gating",
"container-title": "Nature communications",
"author": [
{
"family": "Salehi",
"given": "Saeed"
},
{
"family": "Lei",
"given": "Jordan"
},
{
"family": "Benjamin",
"given": "Ari S"
},
{
"family": "Müller",
"given": "Klaus-Robert"
},
{
"family": "Kording",
"given": "Konrad P"
}
],
"container-title-short":
"volume": "17",
"issue": "1",
"page": "4072",
"DOI": "10.1038/
"PMID": "42086575",
"PMCID": "PMC13144683",
"ISSN": "2041-1723",
"publisher": "Nature Publishing Group",
"URL": "https://
"language": "en",
"issued": {
"date-parts": [
[
2026,
5,
5
]
]
}
}
The tracing map gets a citation of its own once an author has validated it and it has a DOI.
Similar papers
The papers with a page that share the most with this one: the tools found in their code, their categories, datasets, cited references and authors, the rarest counting most.
- [1] doi:10.3758/s13423-024-02495-3
- Are neuronal mechanisms of attention universal across human sensory and motor brain maps?Journal: Psychonomic bulletin & reviewIn common: 17 references
- [2] doi:10.1371/journal.pone.0337087 [code]
- Modulatory feedback determines attentional object segmentation in a model of the ventral stream.Journal: PloS oneIn common: Pillow, PyTorch, SciPy, 2 other tools, 12 references
- [3] doi:10.7554/elife.75726
- The normalization model predicts responses in the human visual cortex during object-based attention.Journal: eLifeIn common: 15 references
- [4] doi:10.1371/journal.pbio.3003354
- Recurrence affects the geometry of visual representations across the ventral visual stream in the human brain.Journal: PLoS biologyIn common: 15 references
- [5] doi:10.1167/jov.23.5.16 [code]
- A brain-inspired object-based attention network for multiobject recognition and visual reasoning.Journal: Journal of visionIn common: Pillow, PyTorch, Matplotlib, 1 other tool, 11 references
- [6] doi:10.1038/s41562-026-02414-7 [code]
- Optimized feature gains explain and predict successes and failures of human selective listening.Journal: Nature human behaviourIn common: PyTorch, seaborn, SciPy, 2 other tools, 11 references
- [7] doi:10.1073/pnas.2411909122 [code]
- Emergent neuronal mechanisms mediating covert attention in convolutional neural networks.Journal: Proceedings of the National Academy of Sciences of the United States of AmericaIn common: SciPy, Matplotlib, NumPy, 12 references
- [8] doi:10.1038/s41467-022-35659-7 [code]
- Efficient neural codes naturally emerge through gradient descent learning.Journal: Nature communicationsIn common: Pillow, PyTorch, seaborn, 3 other tools, 1 reference, 2 authors
- [9] doi:10.1167/jov.24.6.1 [code]
- Convolutional neural network models applied to neuronal responses in macaque V1 reveal limited nonlinear processing.Journal: Journal of visionIn common: Pillow, PyTorch, SciPy, 2 other tools, 10 references
- [10] doi:10.7554/elife.76096 [code]
- Invariant neural subspaces maintained by feedback modulation.Journal: eLifeIn common: PyTorch, SciPy, Matplotlib, 1 other tool, 11 references
Contribute
The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.
Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.
Claim this paper
Correct its record
Say what each link of this record is, remove the ones that are not the paper's, add the ones that are missing. The correction becomes a new version of the record, in its Versions section.
Validate its tracing map
You validate the map as this page shows it: 1 repository of the authors' code, each at its verified commit and with its license, 30 scripts, and 8 matches between paragraphs and code (see the Code and Map sections). It then receives a DOI on Zenodo, with you (your ORCID iD) and OSCR as its creators; the code itself is not deposited.
The map's fingerprint: sha256:7b8a4ace9bb4d74c…
Add the badge to the code's README
If you are an author or a maintainer of this paper's code, you can add this badge. It says what OSCR knows of the paper (its code, its validated tracing map) and links the code to this page. Paste one of these snippets under the README's title. Keep the two marks around it, so that the badge can be replaced in place and is never added twice. OSCR changes nothing in your repository.
Markdown
<!-- oscr-badge:start -->
[](https://openscicode.org/badge/doi/10.1038/s41467-026-72146-9)
<!-- oscr-badge:end -->reStructuredText
.. oscr-badge:start
.. image:: https://openscicode.org/badge/doi/10.1038/s41467-026-72146-9.svg
:target: https://openscicode.org/badge/doi/10.1038/s41467-026-72146-9
:alt: OSCR
.. oscr-badge:endHTML
<!-- oscr-badge:start -->
<a href="https://openscicode.org/badge/doi/10.1038/s41467-026-72146-9"><img src="https://openscicode.org/badge/doi/10.1038/s41467-026-72146-9.svg" alt="OSCR"></a>
<!-- oscr-badge:end -->On GitHub, open the README, choose Edit (the pencil), paste the snippet under the title, then "Commit changes".
If you choose to, you can have the badge kept up to date for you, with oscr badge --write from the command line or with the GitHub Action in your repository's workflows. Every other form and option (a card, AsciiDoc, LaTeX, shields.io) is on the badge page.
Request its removal
To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).
Discussion, reproductions, activity
Discussion: code errors and mismatches between the code and the paper about this paper and its code, from readers and its authors, beside the paper and its tracing map.
Report a mismatch between the code and the paper · Report a code error (signed in)
Reproductions: reports from readers who ran the authors' code and did not get the paper's result, what they ran, with which environment, commit and data.
Report a reproduction failure (signed in)
Versions of its code: the releases of its code tied to a version of this paper (the preprint, the accepted manuscript, the version of record…), each with the tracing map versioned with it.
Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.
