OSCR

Hierarchical learning creates invariant schema within plastic neural networks.

Code ↔ Paper

6 matches between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

The 6 matches
  1. [1] § Methods › Trial firing ↔ scripts/PlotMaker.py, lines 231–303 · score 0.69 · background firing, peak firing, activated firings, steepness, boundary, neurons
  2. [2] § Methods › Top models › Backpropagation networks ↔ scripts/PlotMaker.py, lines 545–608 · score 0.65 · MLPClassifier, max_iter, tanh activation, hidden layers, model, neuron
  3. [3] § Results › Distinct populations of hidden neurons segregate into sequential layers in hierarchical models ↔ scripts/PlotMaker.py, lines 545–608 · score 0.60 · peak firing, activated firing, event cells, hidden layer, NB, HB
  4. [4] § Results › Hierarchical networks converge to a stable schema robust to further training ↔ scripts/PlotMaker.py, lines 1–34 · score 0.60 · Concept space, Network robustness, event cell, Boundary cell, firing, ENN
  5. [5] § Results › Hierarchical schema is distillable into a concise interpretable circuit ↔ scripts/PlotMaker.py, lines 231–303 · score 0.56 · background firing, peak firing, Boundary cell, steepness, Event, activated
  6. [6] § Methods › Top models › Essence neural networks ↔ enn/network.py, lines 32–152 · score 0.55 · subconcept neurons, activation function, artificial, biases, tanh, class

Paper

Loaded from Europe PMC by your browser, not stored by OSCR: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The authors' code

Python · 723 lines · 26 KB · no license · 5 matches

  1. '''
  2. Path: bdENS/repo/scripts/PlotMaker.py
  4. Boundary-Detecting ENN Project
  5. Results Plots
  6. - Figure 2a: Boundary Cell Firing
  7. - Figure 2b: Event Cell Firing
  8. - Figure 2c: Average Firing Rates
  9. - Figure 2d: Trial Firings
  10. - Figure 2e: Neuron Average Firing
  11. - Figure 3b: MLP Plots
  12. - Figure 3c: ENN Plots
  13. - Figure 3e: Network Robustness Plots
  14. - Figure 3f: Concept Space Plots
  15. Author: James R. Elder
  16. Institution: UTSW
  18. DOO: 11-25-2024
  19. LU: 11-25-2024
  21. Reference(s):
  22. - bdENS/release/FigureScripts/Fig2abc.py
  23. - bdENS/release/FigureScripts/Fig2d.py
  24. - bdENS/release/FigureScripts/Fig2e.py
  25. - bdENS/release/FigureScripts/Fig3b.py
  26. - bdENS/release/FigureScripts/Fig3c.py
  27. - bdENS/release/FigureScripts/Fig3e.py
  28. - bdENS/release/FigureScripts/Fig3f.py
  30. - Python 3.12.2
  31. - bdENSenv
  32. - 3.12.x-anaconda
  33. - linux-gnu (BioHPC)
  34. '''
  36. # Project variables
  37. user = "s181641"
  38. projID = "bdENS"
  39. labID = "Lin_lab"
  40. labAffiliation = "greencenter"
  41. projDir = "/work/{}/{}/{}/".format(labAffiliation, user, projID)
  43. # Subproject variables
  44. subprojID = "NetworkVisualization"
  45. branch = "repo"
  47. # System imports
  48. import time
  49. today = time.strftime("%m-%d-%Y")
  50. import os
  51. if os.getcwd() != projDir + branch:
  52. print("Changing to project directory")
  53. os.chdir(projDir + branch) # Change to the project directory
  54. import sys
  55. sys.path.append(projDir + branch) # Add the project directory to the system path
  57. # 3rd party imports
  58. import numpy as np
  59. import matplotlib.pyplot as plt
  61. # ENN imports
  63. # Local imports
  64. from utils import LumberJack as LJ
  65. from utils import NetworkVisualizer as NV
  66. from utils import commonFunctions as cF
  67. from utils import commonClasses as cC
  69. # Logging
  70. logMan = LJ.LoggingManager(subprojID)
  71. cF.updatePlotrcs(font="serif")
  73. # Variables
  74. globers = {
  75. "randomSeed": 42,
  76. "nTrials": 135,
  77. "nBins": 150,
  78. "msPerBin": 10,
  79. "boundaryBin": 50,
  80. }
  82. eventers = {
  83. "nNeurons": 36,
  84. "cellType": "HB",
  85. "peakFiring": 30.1,
  86. "peakStd": 5.5,
  87. }
  89. trialers = {
  90. "nNBTrials": 30,
  91. "nSBTrials": 75,
  92. "nHBTrials": 30
  93. }
  95. boundaryers = {
  96. "nNeurons": 42,
  97. "cellType": "B",
  98. "peakFiring": 19.7,
  99. "peakStd": 4.9
  100. }
  102. boundaryRecording = cF.loadNeuron(7, boundaryers["cellType"])
  103. eventRecording = cF.loadNeuron(2, eventers["cellType"])
  105. eventCellMeanRecordings = np.zeros((eventers["nNeurons"], globers["nBins"]))
  106. boundaryCellMeanRecordings = np.zeros((boundaryers["nNeurons"], globers["nBins"]))
  108. for i in range(1, eventers["nNeurons"]+1):
  109. neuronRecording = cF.loadNeuron(i, 'HB')
  110. neuronRecording = np.mean(neuronRecording[105:], axis=0)
  111. eventCellMeanRecordings[i-1] = neuronRecording
  113. eventCellsMeanRecordings = np.mean(eventCellMeanRecordings, axis=0)
  115. for i in range(1, boundaryers["nNeurons"]+1):
  116. neuronRecording = cF.loadNeuron(i, 'B')
  117. neuronRecording = np.mean(neuronRecording[30:], axis=0)
  118. boundaryCellMeanRecordings[i-1] = neuronRecording
  120. boundaryCellsMeanRecordings = np.mean(boundaryCellMeanRecordings, axis=0)
  122. figWidth = 8
  123. figHeight = 2.5
  124. fig, ax = plt.subplots(1,3, figsize=(figWidth, figHeight))
  126. randomSBTrials = np.random.choice(np.arange(trialers["nNBTrials"], trialers["nNBTrials"] + trialers["nSBTrials"]), 30, replace=False)
  127. boundaryRecordingSubSample = np.append(boundaryRecording[:30], boundaryRecording[randomSBTrials])
  128. boundaryRecordingSubSample = np.append(boundaryRecordingSubSample, boundaryRecording[105:])
  129. boundaryRecordingSubSample = np.reshape(boundaryRecordingSubSample, (-1, 150))
  130. ax[0].imshow(boundaryRecordingSubSample, aspect='auto', cmap="gray_r")
  131. ax[0].set_title("Boundary Cell (B)")
  133. randomSBTrials = np.random.choice(np.arange(trialers["nNBTrials"], trialers["nNBTrials"] + trialers["nSBTrials"]), 30, replace=False)
  134. eventRecordingSubSample = np.append(eventRecording[:30], eventRecording[randomSBTrials])
  135. eventRecordingSubSample = np.append(eventRecordingSubSample, eventRecording[105:])
  136. eventRecordingSubSample = np.reshape(eventRecordingSubSample, (-1, 150))
  137. ax[1].imshow(eventRecordingSubSample, aspect='auto', cmap="gray_r")
  138. ax[1].set_title("Event Cell (E)")
  140. ax[2].plot(boundaryCellsMeanRecordings, color=cC.cellColors["Boundary"], label="B", linestyle="-", linewidth=2.5)
  141. ax[2].plot(eventCellsMeanRecordings, color=cC.cellColors["Event"], label="E", linestyle="-", linewidth=2.5)
  142. ax[2].legend(loc="upper right", ncol=1, bbox_to_anchor=(1.1, 1.05))
  143. # Add early at the top of the plot on the left, and late at the top of the plot on the right
  144. ax[2].annotate("Early", xy=(globers["boundaryBin"], 0.27), xytext=(globers["boundaryBin"]+10, 0.27), textcoords="data", ha="center", va="center", fontsize=8, rotation=90)
  145. ax[2].annotate("Late", xy=(globers["boundaryBin"]+22, 0.275), xytext=(globers["boundaryBin"]+32, 0.275), textcoords="data", ha="center", va="center", fontsize=8, rotation=90)
  146. ax[2].set_title("Mean Firing Rate")
  148. for i in range(3):
  149. ax[i].spines['top'].set_visible(False)
  150. ax[i].spines['right'].set_visible(False)
  151. ax[i].set_xticks([0, 50, 100, 150], ["-0.5", "0.0", "0.5", "1.0"])
  152. ax[i].set_xlim(0, globers["nBins"])
  153. ax[i].set_xlabel("Time, relative to\nboundary (s)")
  154. ax[i].axvline(globers["boundaryBin"], color="black", linestyle="-")
  155. if i != 2:
  156. ax[i].set_ylabel("Trial")
  157. ax[i].set_ylim(90, 0)
  158. ax[i].set_yticks([0, 30, 60, 90], [])
  159. ax[i].axhspan(0, trialers["nNBTrials"], color=cC.conceptColors[0], alpha=0.1)
  160. ax[i].axhspan(trialers["nNBTrials"], trialers["nNBTrials"] + 30, color=cC.conceptColors[1], alpha=0.1)
  161. ax[i].axhspan(60, 90, color=cC.conceptColors[2], alpha=0.1)
  162. else:
  163. ax[i].set_yticks([0, 0.1, 0.2, 0.3], ["0", "1", "2", "3"])
  164. ax[i].set_ylabel("Firing Rate (Hz)")
  165. ax[i].axvline(globers["boundaryBin"]+22, color="black", linestyle="--")
  167. fig.tight_layout()
  168. fig.savefig(logMan.mediaDir + "/Fig2abc.png",
  169. transparent=True,
  170. dpi=300)
  171. plt.close(fig)
  173. # Figure 2d: Trial Firings
  175. # ENN imports
  176. from enn.network import Network
  177. import enn.learnBoundaries as lB
  179. # 3rd party imports
  180. from sklearn.neural_network import MLPClassifier
  181. from sklearn.model_selection import train_test_split
  183. """
  184. # Variables
  185. globers = {
  186. "randomSeed": 42,
  187. "nTrials": 135,
  188. "nBins": 150,
  189. "msPerBin": 10,
  190. "boundaryBin": 50,
  191. "trainTestSplit": 0.5,
  192. "nHiddenNeurons": 3,
  193. "nHiddenLayers": 2
  194. }
  196. mlpers = {
  197. "maxIter": 10000,
  198. "activation": "tanh",
  199. }
  201. eventers = {
  202. "nNeurons": 36,
  203. "cellType": "HB",
  204. "peakFiring": 30.1,
  205. "peakStd": 5.5,
  206. }
  208. trialers = {
  209. "nNBTrials": 30,
  210. "nSBTrials": 75,
  211. "nHBTrials": 30
  212. }
  214. enners = {
  215. "subconceptSelector": 0,
  216. }
  218. boundaryers = {
  219. "nNeurons": 42,
  220. "cellType": "B",
  221. "peakFiring": 19.7,
  222. "peakStd": 4.9
  223. }
  225. plotters = {
  226. "aTrial": 0.1,
  227. "cBoundary": "#E8BDCB",
  228. "cEvent": "#E5C8EE"
  229. }
  231. # Functions
  232. def activateFiring(_avgPeakFiring, _backgroundFiring):
  233. _scaleFactor = 0.5
  234. _xOffset = 1.5
  235. _steepness = 0.1 / np.std(np.mean(_backgroundFiring, axis=1))
  236. _yOffset = 0.5
  238. _normalizedFiring = _avgPeakFiring / np.mean(_backgroundFiring)
  240. return _scaleFactor * np.tanh(_steepness * (_normalizedFiring - _xOffset)) + _yOffset
  243. eventCellActivatedFirings = np.zeros((globers["nTrials"], eventers["nNeurons"]))
  244. lateBoundaryInds = []
  245. earlyEventInds = []
  247. cutoff = 22
  248. for i in range(eventers["nNeurons"]):
  249. recording = cF.loadNeuron(i+1, eventers["cellType"])
  250. preBoundaryFiring = recording[:,:globers["boundaryBin"]]
  251. peakFiring = recording[:,int(globers["boundaryBin"]+eventers["peakFiring"]-2*eventers["peakStd"]):int(globers["boundaryBin"] + eventers["peakFiring"]+2*eventers["peakStd"])]
  252. avgPeakFiring = np.mean(peakFiring, axis=1)
  253. if np.argmax(np.mean(recording[105:], axis=0)) <= globers["boundaryBin"] + cutoff:
  254. earlyEventInds.append(i)
  255. eventCellActivatedFirings[:,i] = activateFiring(avgPeakFiring, preBoundaryFiring)
  257. eventScorer = [-1, -1, 1]
  258. eventAvgClassFiring = [np.mean(eventCellActivatedFirings[:30], axis=0),
  259. np.mean(eventCellActivatedFirings[30:105], axis=0),
  260. np.mean(eventCellActivatedFirings[105:], axis=0)]
  261. eventCellEventness = np.dot(eventScorer, eventAvgClassFiring)
  262. eventnessInds = np.argsort(-eventCellEventness)
  263. eventCellActivatedFirings = eventCellActivatedFirings[:,eventnessInds]
  265. boundaryCellActivatedFirings = np.zeros((globers["nTrials"], boundaryers["nNeurons"]))
  267. for i in range(boundaryers["nNeurons"]):
  268. recording = cF.loadNeuron(i+1, boundaryers["cellType"])
  269. preBoundaryFiring = recording[:,:globers["boundaryBin"]]
  270. peakFiring = recording[:,int(globers["boundaryBin"]+boundaryers["peakFiring"]-2*boundaryers["peakStd"]):int(globers["boundaryBin"] + boundaryers["peakFiring"]+2*boundaryers["peakStd"])]
  271. avgPeakFiring = np.mean(peakFiring, axis=1)
  272. if np.argmax(np.mean(recording[30:], axis=0)) > globers["boundaryBin"] + cutoff:
  273. lateBoundaryInds.append(i)
  274. boundaryCellActivatedFirings[:,i] = activateFiring(avgPeakFiring, preBoundaryFiring)
  276. boundaryScorer = [-1, 1, 1]
  277. boundaryAvgClassFiring = [np.mean(boundaryCellActivatedFirings[:30], axis=0),
  278. np.mean(boundaryCellActivatedFirings[30:105], axis=0),
  279. np.mean( boundaryCellActivatedFirings[105:], axis=0)]
  280. boundaryCellBoundariness = np.dot(boundaryScorer, boundaryAvgClassFiring)
  281. boundaryInds = np.argsort(-boundaryCellBoundariness)
  282. boundaryCellActivatedFirings = boundaryCellActivatedFirings[:,boundaryInds]
  284. earlyCellFirings = []
  285. lateCellFirings = []
  287. for i in range(boundaryCellActivatedFirings.shape[1]):
  288. if i in lateBoundaryInds:
  289. lateCellFirings.append(boundaryCellActivatedFirings[:,i])
  290. else:
  291. earlyCellFirings.append(boundaryCellActivatedFirings[:,i])
  293. for i in range(eventCellActivatedFirings.shape[1]):
  294. if i in earlyEventInds:
  295. earlyCellFirings.append(eventCellActivatedFirings[:,i])
  296. else:
  297. lateCellFirings.append(eventCellActivatedFirings[:,i])
  299. earlyCellFirings = np.array(earlyCellFirings).T
  300. lateCellFirings = np.array(lateCellFirings).T
  302. earlyCellSubFirings = []
  303. lateCellSubFirings = []
  305. # Randomly sample 30 of the SB trials
  306. for i in range(earlyCellFirings.shape[1]):
  307. sbTrials = np.random.choice(range(30, 105), 30, replace=False)
  308. earlyCellSubFirings.append(np.append(earlyCellFirings[:30,i], earlyCellFirings[sbTrials,i]))
  309. earlyCellSubFirings[i] = np.append(earlyCellSubFirings[i], earlyCellFirings[105:,i])
  311. for i in range(lateCellFirings.shape[1]):
  312. sbTrials = np.random.choice(range(30, 105), 30, replace=False)
  313. lateCellSubFirings.append(np.append(lateCellFirings[:30,i], lateCellFirings[sbTrials,i]))
  314. lateCellSubFirings[i] = np.append(lateCellSubFirings[i], lateCellFirings[105:,i])
  316. earlyCellSubFirings = np.array(earlyCellSubFirings).T
  317. lateCellSubFirings = np.array(lateCellSubFirings).T
  319. xAll, yAll = cF.loadEmbeddings()
  320. xAll_NB = xAll[yAll == 0]
  321. xAll_SB = xAll[yAll == 1]
  322. xAll_HB = xAll[yAll == 2]
  324. nTrials = 20
  326. enn_l1Outputs = np.zeros((nTrials, 90, 3))
  327. enn_l2Outputs = np.zeros((nTrials, 90, 3))
  328. mlp_l1Outputs = np.zeros((nTrials, 90, 3))
  329. mlp_l2Outputs = np.zeros((nTrials, 90, 3))
  331. mlpAccs = np.zeros(nTrials)
  332. convThresh = 0.5
  333. for i in range(nTrials):
  334. xTrain, xTest, yTrain, yTest = train_test_split(xAll, yAll,
  335. train_size=globers["trainTestSplit"],
  336. random_state=globers["randomSeed"]+i)
  338. xAll_NB_TrialsInds = np.random.choice(range(len(xAll_NB)), 30, replace=False)
  339. xAll_SB_TrialsInds = np.random.choice(range(len(xAll_SB)), 30, replace=False)
  340. xAll_HB_TrialsInds = np.random.choice(range(len(xAll_HB)), 30, replace=False)
  342. xAll_Trials = np.concatenate((xAll_NB[xAll_NB_TrialsInds],
  343. xAll_SB[xAll_SB_TrialsInds],
  344. xAll_HB[xAll_HB_TrialsInds]), axis=0)
  346. mdlENN = lB.main_0(xTrain, yTrain, enners["subconceptSelector"], globers["nHiddenNeurons"])
  347. mdlMLP = MLPClassifier(hidden_layer_sizes=(globers["nHiddenNeurons"],globers["nHiddenNeurons"]), max_iter=mlpers["maxIter"],
  348. random_state=globers["randomSeed"]+i, activation=mlpers["activation"])
  349. mdlMLP.fit(xTrain, yTrain)
  350. mlpAccs[i] = mdlMLP.score(xTest, yTest)
  352. enn_l1Output = mdlENN.layers[0].compute_output(xAll_Trials, activate=True) # Tanh activation
  353. enn_l2Output = mdlENN.layers[1].compute_output(enn_l1Output, activate=True) # Tanh activation
  355. # Flip the sign of the firings rates if the average firing rate on NB is positive
  356. for i2 in range(3):
  357. if np.mean(enn_l1Output[:30, i2]) > 0:
  358. enn_l1Output[:, i2] = -enn_l1Output[:, i2]
  359. if np.mean(enn_l2Output[:30, i2]) > 0:
  360. enn_l2Output[:, i2] = -enn_l2Output[:, i2]
  362. enn_l1Outputs[i] = enn_l1Output
  363. enn_l2Outputs[i] = enn_l2Output
  365. if mlpAccs[i] > convThresh:
  366. mlp_l1Output_unAct = mdlMLP.coefs_[0].T @ xAll_Trials.T + mdlMLP.intercepts_[0].reshape(-1, 1)
  367. mlp_l1Output = np.tanh(mlp_l1Output_unAct) # Tanh activation
  368. mlp_l2Output_unAct = mdlMLP.coefs_[1].T @ mlp_l1Output + mdlMLP.intercepts_[1].reshape(-1, 1)
  369. mlp_l2Output = np.tanh(mlp_l2Output_unAct) # Tanh activation
  371. mlp_l1Output = mlp_l1Output.T
  372. mlp_l2Output = mlp_l2Output.T
  374. for i2 in range(3):
  375. if np.mean(mlp_l1Output[:30, i2]) > 0:
  376. mlp_l1Output[:, i2] = -mlp_l1Output[:, i2]
  377. if np.mean(mlp_l2Output[:30, i2]) > 0:
  378. mlp_l2Output[:, i2] = -mlp_l2Output[:, i2]
  380. mlp_l1Outputs[i] = mlp_l1Output
  381. mlp_l2Outputs[i] = mlp_l2Output
  385. # Remove unconverged networks
  386. mlp_l1Outputs = mlp_l1Outputs[mlpAccs > convThresh]
  387. mlp_l2Outputs = mlp_l2Outputs[mlpAccs > convThresh]
  388. enn_l1Outputs = enn_l1Outputs[mlpAccs > convThresh]
  389. enn_l2Outputs = enn_l2Outputs[mlpAccs > convThresh]
  391. mlpL1Firings = mlp_l1Outputs.reshape(-1, 90, 3)
  392. mlpL2Firings = mlp_l2Outputs.reshape(-1, 90, 3)
  394. ennL1Firings = enn_l1Outputs.reshape(-1, 90, 3)
  395. ennL2Firings = enn_l2Outputs.reshape(-1, 90, 3)
  397. mlpL1Cells = []
  398. mlpL2Cells = []
  399. ennL1Cells = []
  400. ennL2Cells = []
  402. for i in range(ennL1Firings.shape[0]):
  403. for i2 in range(3):
  404. mlpL1Cells.append(mlpL1Firings[i,:,i2])
  405. mlpL2Cells.append(mlpL2Firings[i,:,i2])
  406. ennL1Cells.append(ennL1Firings[i,:,i2])
  407. ennL2Cells.append(ennL2Firings[i,:,i2])
  409. mlpL1Cells = np.array(mlpL1Cells)
  410. mlpL2Cells = np.array(mlpL2Cells)
  411. ennL1Cells = np.array(ennL1Cells)
  412. ennL2Cells = np.array(ennL2Cells)
  414. mlpL1Cells = np.reshape(mlpL1Cells, (-1, 90))
  415. mlpL2Cells = np.reshape(mlpL2Cells, (-1, 90))
  416. ennL1Cells = np.reshape(ennL1Cells, (-1, 90))
  417. ennL2Cells = np.reshape(ennL2Cells, (-1, 90))
  419. # Sort the cells by average firing rate [Update this for Milo]
  420. # Need to put Boundary like cells first, then Event like cells, then the rest
  421. def sortCells(_cells):
  422. _threshold = 1.5
  423. _boundaryScorer = [-1, 1, 1]
  424. _eventScorer = [-1, -1, 1]
  425. _boundaryCells = []
  426. _eventCells = []
  427. _otherCells = []
  429. for i in range(_cells.shape[0]):
  430. boundaryScore = np.dot(_boundaryScorer, [np.mean(_cells[i,:30]), np.mean(_cells[i,30:60]), np.mean(_cells[i,60:])])
  431. eventScore = np.dot(_eventScorer, [np.mean(_cells[i,:30]), np.mean(_cells[i,30:60]), np.mean(_cells[i,60:])])
  432. if boundaryScore > eventScore and boundaryScore > _threshold:
  433. _boundaryCells.append(_cells[i])
  434. elif eventScore > boundaryScore and eventScore > _threshold:
  435. _eventCells.append(_cells[i])
  436. else:
  437. _otherCells.append(_cells[i])
  439. _boundaryCells = np.array(_boundaryCells)
  440. _boundaryCells = np.reshape(_boundaryCells, (-1, 90))
  442. _eventCells = np.array(_eventCells)
  443. _eventCells = np.reshape(_eventCells, (-1, 90))
  445. _otherCells = np.array(_otherCells)
  446. _otherCells = np.reshape(_otherCells, (-1, 90))
  448. # Sort within the boundary and event cells
  449. _boundaryCells = _boundaryCells[np.argsort(-np.mean(_boundaryCells[:,30:], axis=1))]
  450. _eventCells = _eventCells[np.argsort(-np.mean(_eventCells[:,60:], axis=1))]
  451. _otherCells = _otherCells[np.argsort(-np.mean(_otherCells, axis=1))]
  453. print("Boundary Cells: {}".format(_boundaryCells.shape))
  454. print("Event Cells: {}".format(_eventCells.shape))
  455. print("Other Cells: {}".format(_otherCells.shape))
  457. # Append all cells
  458. _allCells = np.concatenate((_boundaryCells, _eventCells, _otherCells), axis=0)
  459. print("All Cells: {}".format(_allCells.shape))
  461. return _allCells
  463. ennL1Cells = sortCells(ennL1Cells)
  464. ennL2Cells = sortCells(ennL2Cells)
  466. mlpL1Cells = sortCells(mlpL1Cells)
  467. mlpL2Cells = sortCells(mlpL2Cells)
  469. figWidth = 7
  470. figHeight = 2.5
  471. fig, ax = plt.subplots(1, 6, figsize=(figWidth, figHeight), tight_layout=True, sharey=True)
  472. _aspect=1
  474. ax[0].imshow(earlyCellSubFirings, cmap="gray_r",
  475. aspect=_aspect,
  476. vmin=0, vmax=1, interpolation="none")
  477. ax[1].imshow(ennL1Cells.T, cmap="gray_r",
  478. aspect=_aspect,
  479. vmin=-1, vmax=1, interpolation="none")
  480. ax[2].imshow(mlpL1Cells.T, cmap="gray_r",
  481. aspect=_aspect, vmin=-1, vmax=1, interpolation="none")
  483. ax[3].imshow(lateCellSubFirings, cmap="gray_r",
  484. aspect=_aspect, vmin=0, vmax=1, interpolation="none")
  485. ax[4].imshow(ennL2Cells.T, cmap="gray_r",
  486. aspect=_aspect, vmin=-1, vmax=1, interpolation="none")
  487. ax[5].imshow(mlpL2Cells.T, cmap="gray_r",
  488. aspect=_aspect, vmin=-1, vmax=1, interpolation="none")
  490. titles = ["MTL", "ENN", "Backprop"]
  491. for i in range(6):
  492. ax[i].set_yticks([0,30,60,90], labels=[])
  493. ax[i].axhspan(0, 30, color=cC.conceptColors[0], alpha=0.1)
  494. ax[i].axhspan(30, 30 + 30, color=cC.conceptColors[1], alpha=0.1)
  495. ax[i].axhspan(30 + 30, 30 + 30 + 30, color=cC.conceptColors[2], alpha=0.1)
  496. ax[i].set_ylim(89, 0)
  497. ax[i].set_xticks([], labels=[])
  498. ax[i].set_title(titles[i%3])
  500. ax[1].set_xlabel("Early / Layer 1 Neurons")
  501. ax[4].set_xlabel("Late / Layer 2 Neurons")
  503. fig.tight_layout()
  504. fig.savefig(logMan.mediaDir + "/Fig2d.png",
  505. transparent=True,
  506. dpi=300)
  507. plt.close(fig)
  508. """
  510. # Figure 2e: Neuron Average Firing
  511. # Functions
  512. def activateFiring(_avgPeakFiring, _backgroundFiring):
  513. _scaleFactor = 0.5
  514. _xOffset = 1.5
  515. _steepness = 0.1 / np.std(np.mean(_backgroundFiring, axis=1))
  516. _yOffset = 0.5
  518. _normalizedFiring = _avgPeakFiring / np.mean(_backgroundFiring)
  520. return _scaleFactor * np.tanh(_steepness * (_normalizedFiring - _xOffset)) + _yOffset
  522. def processNeurons(_nNeurons, _cellType, _peakFiring, _peakStd, globers):
  523. _activatedFirings = np.zeros((globers["nTrials"], _nNeurons))
  525. for i in range(_nNeurons):
  526. _recording = cF.loadNeuron(i + 1, _cellType)
  527. _preBoundaryFiring = _recording[:, :globers["boundaryBin"]]
  528. _peakFiringRange = _recording[:, int(globers["boundaryBin"] + _peakFiring - 2*_peakStd):int(globers["boundaryBin"] + _peakFiring + 2*_peakStd)]
  529. _avgPeakFiring = np.mean(_peakFiringRange, axis=1)
  530. _activatedFirings[:, i] = activateFiring(_avgPeakFiring, _preBoundaryFiring)
  532. return _activatedFirings
  534. def computeStats(_activatedFirings):
  535. avgNB = np.mean(_activatedFirings[:30], axis=0)
  536. avgSB = np.mean(_activatedFirings[30:30 + 75], axis=0)
  537. avgHB = np.mean(_activatedFirings[30 + 75:], axis=0)
  539. stdNB = np.std(_activatedFirings[:30], axis=0)
  540. stdSB = np.std(_activatedFirings[30:30 + 75], axis=0)
  541. stdHB = np.std(_activatedFirings[30 + 75:], axis=0)
  543. return [avgNB, avgSB, avgHB], [stdNB, stdSB, stdHB]
  545. eventers = {
  546. "nNeurons": 36,
  547. "cellType": "HB",
  548. "peakFiring": 30.1,
  549. "peakStd": 5.5,
  550. }
  552. boundaryers = {
  553. "nNeurons": 42,
  554. "cellType": "B",
  555. "peakFiring": 19.7,
  556. "peakStd": 4.9
  557. }
  559. globers = {
  560. "randomSeed": 42,
  561. "nTrials": 135,
  562. "nBins": 150,
  563. "msPerBin": 10,
  564. "boundaryBin": 50,
  565. "nHiddenNeurons": 3,
  566. "trainTestSplit": 0.5,
  567. }
  569. eventCellActivatedFirings = processNeurons(eventers["nNeurons"], eventers["cellType"], eventers["peakFiring"], eventers["peakStd"], globers)
  570. boundaryCellActivatedFirings = processNeurons(boundaryers["nNeurons"], boundaryers["cellType"], boundaryers["peakFiring"], boundaryers["peakStd"], globers)
  572. eventCellActivatedFiringsAvg, eventCellActivatedFiringsStd = computeStats(eventCellActivatedFirings)
  573. boundaryCellActivatedFiringsAvg, boundaryCellActivatedFiringsStd = computeStats(boundaryCellActivatedFirings)
  575. xAll, yAll = cF.loadEmbeddings()
  576. xAll_NB = xAll[yAll == 0]
  577. xAll_SB = xAll[yAll == 1]
  578. xAll_HB = xAll[yAll == 2]
  580. nNetworks = 20
  581. enn_l1Outputs = np.zeros((nNetworks, 90, 3))
  582. enn_l2Outputs = np.zeros((nNetworks, 90, 3))
  583. mlp_l1Outputs = np.zeros((nNetworks, 90, 3))
  584. mlp_l2Outputs = np.zeros((nNetworks, 90, 3))
  586. mlpAccs = np.zeros(nNetworks)
  587. convThresh = 0.5
  589. for i in range(nNetworks):
  590. xTrain, xTest, yTrain, yTest = train_test_split(xAll, yAll,
  591. train_size=globers["trainTestSplit"],
  592. random_state=42+i)
  594. mdlMLP = MLPClassifier(hidden_layer_sizes=(3,3), max_iter=10000, random_state=42+i, activation="tanh")
  595. mdlMLP.fit(xTrain, yTrain)
  596. mlpAccs[i] = mdlMLP.score(xTest, yTest)
  597. mdlENN = lB.main_0(xTrain, yTrain, 0, globers["nHiddenNeurons"])
  599. xAll_NB_TrialsInds = np.random.choice(range(len(xAll_NB)), 30, replace=False)
  600. xAll_SB_TrialsInds = np.random.choice(range(len(xAll_SB)), 30, replace=False)
  601. xAll_HB_TrialsInds = np.random.choice(range(len(xAll_HB)), 30, replace=False)
  603. xAll_Trials = np.concatenate((xAll_NB[xAll_NB_TrialsInds],
  604. xAll_SB[xAll_SB_TrialsInds],
  605. xAll_HB[xAll_HB_TrialsInds]), axis=0)
  607. enn_l1Output = mdlENN.layers[0].compute_output(xAll_Trials, activate=True) # Tanh activation
  608. enn_l2Output = mdlENN.layers[1].compute_output(enn_l1Output, activate=True) # Tanh activation
  610. # Flip the sign of the firings rates if the average firing rate on NB is positive
  611. for i2 in range(3):
  612. if np.mean(enn_l1Output[:30, i2]) > 0:
  613. enn_l1Output[:, i2] = -enn_l1Output[:, i2]
  614. if np.mean(enn_l2Output[:30, i2]) > 0:
  615. enn_l2Output[:, i2] = -enn_l2Output[:, i2]
  617. enn_l1Outputs[i] = enn_l1Output
  618. enn_l2Outputs[i] = enn_l2Output
  620. if mlpAccs[i] > convThresh:
  621. mlp_l1Output_unAct = mdlMLP.coefs_[0].T @ xAll_Trials.T + mdlMLP.intercepts_[0].reshape(-1, 1)
  622. mlp_l1Output = np.tanh(mlp_l1Output_unAct) # Tanh activation
  623. mlp_l2Output_unAct = mdlMLP.coefs_[1].T @ mlp_l1Output + mdlMLP.intercepts_[1].reshape(-1, 1)
  624. mlp_l2Output = np.tanh(mlp_l2Output_unAct) # Tanh activation
  626. mlp_l1Output = mlp_l1Output.T
  627. mlp_l2Output = mlp_l2Output.T
  629. for i2 in range(3):
  630. if np.mean(mlp_l1Output[:30, i2]) > 0:
  631. mlp_l1Output[:, i2] = -mlp_l1Output[:, i2]
  632. if np.mean(mlp_l2Output[:30, i2]) > 0:
  633. mlp_l2Output[:, i2] = -mlp_l2Output[:, i2]
  635. mlp_l1Outputs[i] = mlp_l1Output
  636. mlp_l2Outputs[i] = mlp_l2Output
  638. # Remove unconverged networks
  639. mlp_l1Outputs = mlp_l1Outputs[mlpAccs > convThresh]
  640. mlp_l2Outputs = mlp_l2Outputs[mlpAccs > convThresh]
  641. enn_l1Outputs = enn_l1Outputs[mlpAccs > convThresh]
  642. enn_l2Outputs = enn_l2Outputs[mlpAccs > convThresh]
  644. enn_l1AvgNBFiring = np.mean(enn_l1Outputs[:,:30,:], axis=1)
  645. enn_l1AvgSBFiring = np.mean(enn_l1Outputs[:,30:30+30,:], axis=1)
  646. enn_l1AvgHBFiring = np.mean(enn_l1Outputs[:,30+30:,:], axis=1)
  648. enn_l2AvgNBFiring = np.mean(enn_l2Outputs[:,:30,:], axis=1)
  649. enn_l2AvgSBFiring = np.mean(enn_l2Outputs[:,30:30+30,:], axis=1)
  650. enn_l2AvgHBFiring = np.mean(enn_l2Outputs[:,30+30:,:], axis=1)
  652. ennl1StdNBFiring = np.std(enn_l1Outputs[:,:30,:], axis=1)
  653. ennl1StdSBFiring = np.std(enn_l1Outputs[:,30:30+30,:], axis=1)
  654. ennl1StdHBFiring = np.std(enn_l1Outputs[:,30+30:,:], axis=1)
  656. ennl2StdNBFiring = np.std(enn_l2Outputs[:,:30,:], axis=1)
  657. ennl2StdSBFiring = np.std(enn_l2Outputs[:,30:30+30,:], axis=1)
  658. ennl2StdHBFiring = np.std(enn_l2Outputs[:,30+30:,:], axis=1)
  660. mlp_l1AvgNBFiring = np.mean(mlp_l1Outputs[:,:30,:], axis=1)
  661. mlp_l1AvgSBFiring = np.mean(mlp_l1Outputs[:,30:30+30,:], axis=1)
  662. mlp_l1AvgHBFiring = np.mean(mlp_l1Outputs[:,30+30:,:], axis=1)
  664. mlp_l2AvgNBFiring = np.mean(mlp_l2Outputs[:,:30,:], axis=1)
  665. mlp_l2AvgSBFiring = np.mean(mlp_l2Outputs[:,30:60,:], axis=1)
  666. mlp_l2AvgHBFiring = np.mean(mlp_l2Outputs[:,30+30:,:], axis=1)
  668. mlp_l1StdNBFiring = np.std(mlp_l1Outputs[:,:30,:], axis=1)
  669. mlp_l1StdSBFiring = np.std(mlp_l1Outputs[:,30:60,:], axis=1)
  670. mlp_l1StdHBFiring = np.std(mlp_l1Outputs[:,30+30:,:], axis=1)
  672. mlp_l2StdNBFiring = np.std(mlp_l2Outputs[:,:30,:], axis=1)
  673. mlp_l2StdSBFiring = np.std(mlp_l2Outputs[:,30:60,:], axis=1)
  674. mlp_l2StdHBFiring = np.std(mlp_l2Outputs[:,30+30:,:], axis=1)
  676. # Plot the results
  677. figWidth = 8
  678. figHeight = 4
  679. fig, ax = plt.subplots(1, 3, figsize=(figWidth, figHeight), subplot_kw={'projection': '3d'})
  681. ax[0].scatter(eventCellActivatedFiringsAvg[0], eventCellActivatedFiringsAvg[1], eventCellActivatedFiringsAvg[2], color=cC.cellColors["Event"], label='Event Cell', alpha=1)
  682. ax[0].scatter(boundaryCellActivatedFiringsAvg[0], boundaryCellActivatedFiringsAvg[1], boundaryCellActivatedFiringsAvg[2], color=cC.cellColors["Boundary"], label='Boundary Cell', alpha=1)
  683. ax[0].legend(bbox_to_anchor=(0., 2), loc='upper left', ncols=2)
  685. ax[1].scatter(enn_l1AvgNBFiring, enn_l1AvgSBFiring, enn_l1AvgHBFiring, color='blue', label='Layer 1', alpha=1)
  686. ax[1].scatter(enn_l2AvgNBFiring, enn_l2AvgSBFiring, enn_l2AvgHBFiring, color='red', label='Layer 2', alpha=1)
  687. #ax[1].legend(bbox_to_anchor=(0.5, 2), loc='upper left', ncols=2)
  689. ax[2].scatter(mlp_l1AvgNBFiring, mlp_l1AvgSBFiring, mlp_l1AvgHBFiring, color='blue', label='Layer 1', alpha=1)
  690. ax[2].scatter(mlp_l2AvgNBFiring, mlp_l2AvgSBFiring, mlp_l2AvgHBFiring, color='red', label='Layer 2', alpha=1)
  691. ax[2].legend(bbox_to_anchor=(-0.5, 2), loc='upper left', ncols=2)
  693. ax[0].set_title("MTL")
  694. ax[1].set_title("ENN")
  695. ax[2].set_title("Backprop")
  697. ax[0].set_xlim([0, 1])
  698. ax[0].set_ylim([0, 1])
  699. ax[0].set_zlim([0, 1])
  701. ax[0].set_xticks([0, 0.5, 1])
  702. ax[0].set_yticks([0, 0.5, 1])
  703. ax[0].set_zticks([0, 0.5, 1])
  705. for i in range(3):
  706. ax[i].set_xlabel("{} Firing".format(r'$\overline{NB}$'))
  707. ax[i].set_ylabel("{} Firing".format(r'$\overline{SB}$'))
  708. ax[i].set_zlabel("{} Firing".format(r'$\overline{HB}$'))
  710. for i in range(1,3):
  711. ax[i].set_xlim([-1, 1])
  712. ax[i].set_ylim([-1, 1])
  713. ax[i].set_zlim([-1, 1])
  715. ax[i].set_xticks([-1, 0, 1])
  716. ax[i].set_yticks([-1, 0, 1])
  717. ax[i].set_zticks([-1, 0, 1])
  719. fig.subplots_adjust(wspace=0.75)
  720. fig.savefig(logMan.mediaDir + "/Fig2e.png",
  721. transparent=True,
  722. dpi=300)
  723. plt.close(fig)

PlotMaker.py at commit 8aee36c, no license · at the source

Overview

Authors: James R. Elder1,2,3, Jie Zheng4,5, Lydia B. Shimelis6, Ueli Rutishauser7,8,9,10, Milo M. Lin1,2,11,12
ORCID iDs: Milo M. Lin
  1. Green Center for Systems Biology, University of Texas Southwestern Medical Center,Dallas, TX 75390 USA
  2. Lyda Hill Dept. of Bioinformatics, University of Texas Southwestern Medical Center,Dallas, TX 75390 USA
  3. Molecular Biophysics Program, University of Texas Southwestern Medical Center,Dallas, TX 75390 USA
  4. Department of Biomedical Engineering, University of California at Davis,Davis, CA 95616 USA
  5. Department of Neurological Surgery, UC Davis Health,Davis, CA 95616 USA
  6. Biomedical Engineering and Neuroscience, Harvard University,Cambridge, MA 02138 USA
  7. Department of Neurosurgery, Cedars-Sinai Medical Center,Los Angeles, CA 90048 USA
  8. Department of Neurology, Cedars-Sinai Medical Center,Los Angeles, CA 90048 USA
  9. Center for Neural Science and Medicine, Department of Biomedical Sciences, Cedars-Sinai Medical Center,Los Angeles, CA 90048 USA
  10. Division of Biology and Biological Engineering, California Institute of Technology,Pasadena, CA 91125 USA
  11. Department of Biophysics, University of Texas Southwestern Medical Ctr.,Dallas, TX 75390 USA
  12. Center for Alzheimer’s and Neurodegenerative Diseases, University of Texas Southwestern Medical Center,Dallas, TX 75390 USA
Institutions: The University of Texas Southwestern Medical Center (United States); University of California, Davis (United States); UC Davis Health (United States); Harvard University (United States); Cedars-Sinai Medical Center (United States); California Institute of Technology (United States)
Journal: Journal of computational neuroscience, volume 54, issue 3, pages 503-514
Dates: received 5 December 2025; accepted 18 June 2026; published online 30 June 2026; in print 2026
Type: Research article · Language: English
License: CC BY-NC-ND
Identifiers: DOI 10.1007/s10827-026-00940-x · PMID 42377689 · PMCID PMC13588848 · OpenAlex W7166740900
Open access: hybrid, a free copy (OpenAlex)
Status: code verified
Categories: human (organism)
Methods: Statistics, Machine learning
Keywords: cognitive neuroscience, schema formation
MeSH: Learning*, Models, Neurological*, Neural Networks, Computer*, Neuronal Plasticity*, Neurons*, Algorithms, Animals, Humans, Nerve Net, Soft Computing (* major topic)
Topic: Neural dynamics and brain function (Cognitive Neuroscience, Neuroscience), according to OpenAlex
Funding: National Institutes of Health,United States (5T32GM131963, R00NS126233, R01GM125748)
Citations: not cited yet (Europe PMC); 50 references in the paper

Abstract

The abstract is not reproduced here: the paper's license (CC BY-NC-ND) does not allow it. Read it in the paper, at the publisher or on Europe PMC.

Repository

Its files are read in the Code ↔ Paper reader above, with 6 matches between paragraphs and lines of code.

jre411/bdENN

License: none: the authors keep all their rights
State: the link answers, verified on 27 September 2026
Evidence: files inventoried
Commit: 8aee36c26295ac90ceec784a66da1925c225d34a, 4 February 2025
Languages: Python (24)
Size: 3,643 files, 24 scripts
Software Heritage: not archived
Found in: “Data Availability”
Holds: README
Not found: license file, CITATION.cff, environment file, tests, continuous integration, documentation
Tools: NumPy (20 files), Matplotlib (15 files), scikit-learn (10 files), SciPy (4 files), pandas (2 files), Keras (1 file)
Availability: 1 check, the latest on 27 September 2026: the link answers
  • 27 September 2026: the link answers
25 files
At the source: github.com/jre411/bdENN

The paper's code and data availability statement is in the Data section.

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;
  • 24 scripts, each with its path and the digest of its content;
  • 6 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.

Code and data availability statement

The paper has a code and data availability statement. Its license (CC BY-NC-ND) does not allow reproducing it here; in short, from what the harvester recognized in it:

Read it in the paper: doi.org/10.1007/s10827-026-00940-x.

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

  • Publisher: n/a → Springer Science+Business Media

Version 1, 27 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 5 authors, 2 keywords, 10 MeSH terms, 1 funder, 37 references.

Cite

This paper

Elder, J. R., Zheng, J., Shimelis, L. B., Rutishauser, U., & Lin, M. M. (2026). Hierarchical learning creates invariant schema within plastic neural networks. Journal of computational neuroscience, 54(3), 503-514. https://doi.org/10.1007/s10827-026-00940-x

BibTeX

@article{elder2026hierarchical,
author = {Elder, James R. and Zheng, Jie and Shimelis, Lydia B. and Rutishauser, Ueli and Lin, Milo M.},
title = {{Hierarchical learning creates invariant schema within plastic neural networks}},
journal = {Journal of computational neuroscience},
year = {2026},
month = jun,
volume = {54},
number = {3},
pages = {503--514},
publisher = {Springer Science+Business Media},
issn = {0929-5313},
doi = {10.1007/s10827-026-00940-x},
url = {https://doi.org/10.1007/s10827-026-00940-x},
pmid = {42377689},
pmcid = {PMC13588848}
}

RIS

TY - JOUR
AU - Elder, James R.
AU - Zheng, Jie
AU - Shimelis, Lydia B.
AU - Rutishauser, Ueli
AU - Lin, Milo M.
TI - Hierarchical learning creates invariant schema within plastic neural networks
T2 - Journal of computational neuroscience
J2 - J Comput Neurosci
PY - 2026
DA - 2026/06/30
VL - 54
IS - 3
SP - 503
EP - 514
SN - 0929-5313
PB - Springer Science+Business Media
DO - 10.1007/s10827-026-00940-x
UR - https://doi.org/10.1007/s10827-026-00940-x
LA - en
ER -

CSL-JSON

{
"id": "10.1007/s10827-026-00940-x",
"type": "article-journal",
"title": "Hierarchical learning creates invariant schema within plastic neural networks",
"container-title": "Journal of computational neuroscience",
"author": [
{
"family": "Elder",
"given": "James R."
},
{
"family": "Zheng",
"given": "Jie"
},
{
"family": "Shimelis",
"given": "Lydia B."
},
{
"family": "Rutishauser",
"given": "Ueli"
},
{
"family": "Lin",
"given": "Milo M."
}
],
"container-title-short": "J Comput Neurosci",
"volume": "54",
"issue": "3",
"page": "503-514",
"DOI": "10.1007/s10827-026-00940-x",
"PMID": "42377689",
"PMCID": "PMC13588848",
"ISSN": "0929-5313",
"publisher": "Springer Science+Business Media",
"URL": "https://doi.org/10.1007/s10827-026-00940-x",
"language": "en",
"issued": {
"date-parts": [
[
2026,
6,
30
]
]
}
}

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:10.1038/s42003-026-10169-0 [code]
Shared representations in brains and models reveal a two-route cortical organization during scene perception.
Journal: Communications biology
In common: scikit-learn, pandas, SciPy, 2 other tools, 3 references
[2] doi:10.1162/imag.a.1207 [code]
Investigating the temporal dynamics and modeling of mid-level feature representations in humans.
Journal: Imaging neuroscience (Cambridge, Mass.)
In common: scikit-learn, pandas, SciPy, 2 other tools, 3 references
[3] doi:10.1186/s13321-026-01177-7 [code]
A pipeline for developing AI-driven models to predict molecular initiating events: a case study on neural tube defects.
Journal: Journal of cheminformatics
In common: scikit-learn, pandas, SciPy, 2 other tools, 2 references
[4] doi:10.1371/journal.pcbi.1014656 [code]
Contrastive learning to fine-tune feature extraction models for the visual cortex.
Journal: PLoS computational biology
In common: scikit-learn, pandas, SciPy, 2 other tools, 2 references
[5] doi:10.1038/s42003-026-10094-2 [code]
E/I imbalance and internal noise cause weak neural representations and face recognition challenges in ASD.
Journal: Communications biology
In common: scikit-learn, pandas, SciPy, 2 other tools, 2 references
[6] doi:10.1016/j.isci.2026.116439 [code]
Decoding the role of transcriptomic clocks in the human prefrontal cortex.
Journal: iScience
In common: Keras, scikit-learn, pandas, 3 other tools, 1 reference
[7] doi:10.1038/s41597-026-07248-6 [code]
A large-scale fMRI dataset for vision-language semantic association.
Journal: Scientific data
In common: scikit-learn, pandas, SciPy, 2 other tools, 2 references
[8] doi:10.1016/j.bpj.2026.06.005 [code]
Curvature-based machine-learning method for automated segmentation of dendritic spines.
Journal: Biophysical journal
In common: Keras, scikit-learn, pandas, 3 other tools, 1 reference
[9] doi:10.1038/s41467-026-72687-z [code]
Chordless cycle filtrations for dimensionality detection in complex networks via topological data analysis.
Journal: Nature communications
In common: Keras, scikit-learn, pandas, 3 other tools, 1 reference
[10] doi:10.1038/s41598-026-52672-8 [code]
Validation of portable, semi-dry electrode-based electroencephalography device for its application in brain-computer interface solutions.
Journal: Scientific reports
In common: Keras, scikit-learn, SciPy, 2 other tools, 1 reference

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.

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: questions and error reports about this paper and its code, from signed-in readers and its authors. It opens with sign-in.

Reproductions: reports from readers who ran the authors' code: what they reproduced, with which environment, commit and data. It opens with sign-in.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.