pith. sign in

arxiv: 2509.15832 · v3 · submitted 2025-09-19 · 🧬 q-bio.NC

Overcoming Output Dimension Collapse: When Sparsity Enables Zero-shot Brain-to-Image Reconstruction at Small Data Scales

Kenya Otsuka , Yoshihiro Nagano , Yukiyasu Kamitani This is my paper

Pith reviewed 2026-05-18 16:06 UTC · model grok-4.3

classification 🧬 q-bio.NC
keywords brain-to-image reconstructionsparse linear regressionoutput dimension collapsezero-shot reconstructionsmall data scalesstudent-teacher frameworkfMRI decodingprediction error
0
0 comments X

The pith

Sparse multivariate linear regression avoids output dimension collapse in brain-to-image reconstruction when training samples are few relative to feature dimensions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper analyzes two common translators for mapping brain activity to latent image features in zero-shot reconstruction tasks limited by scarce brain-image pairs. It shows that naive multivariate linear regression produces output dimension collapse at small data scales, defined as the ratio of training samples to feature dimensionality, creating a generalization bottleneck. Sparse multivariate linear regression, analyzed via a student-teacher framework, derives prediction error expressions showing that variable selection can lower error by exploiting sparsity in the brain-to-feature mapping. This matters for readers because it targets the data scarcity issue that limits externalizing subjective visual experiences from brain signals. The work supplies a best prediction diagnostic computable without brain data and guidelines for translator and feature design.

Core claim

Naive multivariate linear regression suffers output dimension collapse as a structural property when the number of training samples is small compared to latent feature dimensionality, acting as a practical bottleneck in zero-shot brain-to-image reconstruction; sparse linear regression models in the student-teacher framework yield explicit prediction error formulas in terms of data scale and sparsity parameters, clarifying that variable selection reduces error at small data scales when the brain-to-feature mapping is sufficiently sparse.

What carries the argument

Output dimension collapse, the rank deficiency arising in naive multivariate linear regression predictions when samples fall below output dimensionality, which the paper shows becomes a generalization bottleneck quantifiable by the best prediction diagnostic.

If this is right

  • The best prediction diagnostic can be computed from feature statistics alone to flag when output dimension collapse is harming reconstruction.
  • Sparse regression supplies quantitative thresholds on data scale and sparsity parameters for when it outperforms naive methods.
  • Feature spaces should be selected or engineered to make brain-to-feature mappings sparse enough to benefit from variable selection.
  • Translators for zero-shot reconstruction at limited data scales should incorporate sparsity-aware variable selection rather than full multivariate regression.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The student-teacher error derivations could be tested directly on simulated brain responses to confirm the sparsity thresholds before real-data application.
  • Similar dimension-collapse analysis might apply to other brain-decoding domains such as auditory or language reconstruction where output spaces are also high-dimensional.
  • If sparsity holds across subjects, the approach could reduce the number of training stimuli needed for personalized reconstruction pipelines.

Load-bearing premise

The brain-to-feature mapping has enough sparsity for variable selection to improve prediction accuracy over naive regression at small data scales.

What would settle it

Measure whether sparse regression yields measurably lower prediction error than naive regression on held-out brain-to-image pairs precisely when the ratio of training samples to feature dimensionality is low, checking if the gap matches the derived student-teacher error expressions.

Figures

Figures reproduced from arXiv: 2509.15832 by Kenya Otsuka, Yoshihiro Nagano, Yukiyasu Kamitani.

Figure 1
Figure 1. Figure 1: Translator–Generator pipeline. The translator converts brain activity into latent features, and the generator transforms latent features into reconstructed images. A naive multivariate linear regression model and a sparse multivariate linear regression model are commonly used as translators. Whereas a naive model uses a single shared set of input variables to predict all output dimensions, a sparse model a… view at source ↗
Figure 2
Figure 2. Figure 2: Output dimension collapse. (A) The predictions become restricted to a low-dimensional subspace determined by the training outputs. (B) The best prediction within the training feature subspace provides the lower bound for prediction error in naive multivariate linear regression. of the training data row-wise, respectively. For a test input brain activity xte, the prediction is yˆte = Wˆ ⊤xte (3) = Ytr ⊤Xtr … view at source ↗
Figure 3
Figure 3. Figure 3: The best predictions for different training data sizes. (A) The best prediction error 1 dout ∥yˆ opt te − yte∥ 2 for each test sample across training data sizes 1200, 600, 300, and 150. The vertical axis represents the error of the best prediction, and the horizontal axis represents the sample index sorted by the error at n =150. (B) Repre￾sentative reconstructions from the best predictions for different t… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of the best prediction and the brain prediction. (A) Left: Percentage of the best prediction error to the brain prediction error ∥yˆ opt te − yte∥ 2/∥yˆte − yte∥ 2 . Blue for natural images, orange for artificial shapes. The error bars denote the standard deviation across samples. Right: Sample-wise comparison of best prediction error 1 dout ∥yˆ opt te −yte∥ 2 and brain prediction error 1 dout ∥… view at source ↗
Figure 5
Figure 5. Figure 5: Sparse brain-to-feature mapping. (A) If the brain and the latent features exhibit local or selective activity, their relationship should become sparse mapping. (B) Data generation process that reflects the sparse structure of the brain-to-feature mapping. The smaller the non-zero weight ratio a, the sparser the input-output mapping is. ods such as CFS (Hall, 2000) and SIS (Fan and Lv, 2008)), as well as ap… view at source ↗
Figure 6
Figure 6. Figure 6: Prediction error of sparse regression. Each heatmap shows the prediction error R of sparse regression with data scale n/d = 0.2 (left) and n/d = 0.8 (right). The horizontal axis represents selection rate psel, and the vertical axis represents hit rate π. The diagonal line indicates the case where psel = π, which corresponds to random selection. The upper-left shaded region defined by π ≥ psel/a represents … view at source ↗
Figure 7
Figure 7. Figure 7: Correlation-based filter. The depth labeled n is the sample axis: each slice represents one training sample. For each feature in y, compute its correlation with all inputs in x across the n samples, select the top dsel by absolute correlation, and fit the model using only the selected inputs. = 0.01 = 0.02 = 0.04 = 0.08 = 0.16 = 0.32 = 0.01 = 0.02 = 0.04 = 0.08 = 0.16 = 0.32 Hit rate Selection rate Hit rat… view at source ↗
Figure 8
Figure 8. Figure 8: Hit rate versus selection rate with a correlation-based filter. Each curve shows the estimated prediction error R for different non-zero weight ratios a. The horizontal axis shows psel, and the vertical axis shows π. Left: data scale n/d = 0.2, right: data scale n/d = 0.8. where Φ is the cumulative distribution function of the standard normal dis￾tribution. A detailed derivation is provided in Appendix A.4… view at source ↗
Figure 9
Figure 9. Figure 9: The estimated prediction error from theory. The black line shows the prediction error of naive linear regression and is independent of the non-zero weight ratio a, so only one curve is displayed. The colored lines show the prediction error of sparse regression for different non-zero weight ratios a. The horizontal axis represents the data scale n/d, and the vertical axis represents the estimated prediction… view at source ↗
Figure 10
Figure 10. Figure 10: The simulated prediction error under four data-generating processes. Each plot shows the simulated prediction error R of sparse regression under four differ￾ent data-generating processes: (A) Baseline (the same as the setting in our theory), (B) Gaussian-distributed non-zero weights, (C) Correlated signals modeled by a Toeplitz co￾variance with ρ = 0.5, (D) Input noise. The horizontal axis represents the … view at source ↗
read the original abstract

Advances in brain-to-image reconstruction are enabling us to externalize the subjective visual experiences encoded in the brain as images. A key challenge in this task is data scarcity: a translator that maps brain activity to latent image features is trained on a limited number of brain-image pairs, making the translator a bottleneck for zero-shot reconstruction beyond the training stimuli. In this paper, we mathematically analyze the behavior of two translators commonly used in recent reconstruction pipelines: naive multivariate linear regression and sparse multivariate linear regression. We define the data scale as the ratio of the number of training samples to the latent feature dimensionality and characterize the behavior of each model across data scales. Building on a standard structural property of naive multivariate regression, we first show that the resulting ``output dimension collapse'' can become a practical generalization bottleneck in brain-to-image reconstruction. We introduce the best prediction diagnostic, which is computable without brain activity, to quantify the practical impact of this collapse. We then analyze sparse linear regression models in a student--teacher framework and derive expressions for the prediction error in terms of data scale and other sparsity-related parameters. Our analysis clarifies when variable selection can reduce prediction error at small data scales by exploiting the sparsity of the brain-to-feature mapping. Our findings provide quantitative guidelines for diagnosing output dimension collapse and for designing effective translators and feature representations for zero-shot reconstruction.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript analyzes naive and sparse multivariate linear regression translators for brain-to-image reconstruction under data scarcity. It identifies output dimension collapse in naive regression at small data scales (n/d << 1) as a generalization bottleneck and introduces a best prediction diagnostic computable without brain activity. In a student-teacher framework, it derives closed-form prediction error expressions for sparse regression in terms of data scale and sparsity parameters, clarifying when variable selection reduces error by exploiting brain-to-feature mapping sparsity and providing quantitative guidelines for translator and feature design.

Significance. If the derivations are sound and the sparsity assumption holds for real brain data, the work supplies theoretical tools to diagnose and mitigate a key bottleneck in zero-shot reconstruction pipelines. The best prediction diagnostic and explicit error expressions versus data scale offer practical value for designing translators at small scales, a common regime in neuroimaging. The student-teacher analysis enables parameter-free insights into sparsity benefits without requiring large-scale simulations.

major comments (2)
  1. [Student-teacher framework] Student-teacher framework section: The central claim that sparse regression reduces prediction error at small scales rests on the teacher mapping being sufficiently sparse; the paper should include an empirical estimate or sensitivity analysis of sparsity levels in actual brain-to-latent-feature mappings (e.g., from NSD or similar datasets) to confirm the assumed sparsity pattern transfers, as mismatch would invalidate the quantitative guidelines.
  2. [Naive multivariate linear regression analysis] Analysis of naive regression and output dimension collapse: While the structural property is standard, the manuscript must specify the precise data-scale threshold (with equation) at which collapse becomes the dominant error source versus other factors like noise, to make the 'practical generalization bottleneck' claim load-bearing rather than asymptotic.
minor comments (2)
  1. [Abstract] Abstract describes derivations but provides no equations; adding the key closed-form error expressions would improve accessibility.
  2. [Best prediction diagnostic] The 'best prediction diagnostic' is introduced without an explicit formula or pseudocode; a dedicated subsection or equation would clarify its computation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below with point-by-point responses, indicating planned revisions where appropriate. Our goal is to strengthen the quantitative aspects of the analysis while preserving the theoretical focus of the work.

read point-by-point responses

  1. Referee: [Student-teacher framework] Student-teacher framework section: The central claim that sparse regression reduces prediction error at small scales rests on the teacher mapping being sufficiently sparse; the paper should include an empirical estimate or sensitivity analysis of sparsity levels in actual brain-to-latent-feature mappings (e.g., from NSD or similar datasets) to confirm the assumed sparsity pattern transfers, as mismatch would invalidate the quantitative guidelines.

    Authors: We agree that demonstrating robustness to the sparsity level would make the guidelines more actionable. The student-teacher derivations are parameterized by the sparsity level of the teacher mapping and explicitly show the conditions (in terms of data scale and sparsity) under which variable selection reduces error; the central claim is therefore conditional rather than absolute. To address the request, we will add a sensitivity analysis that varies the teacher sparsity parameter across a range of plausible values and reports the resulting prediction error curves at small data scales. Regarding empirical estimates from datasets such as NSD, our manuscript is primarily a theoretical analysis; we will include a short discussion of how effective sparsity could be estimated post hoc using the best-prediction diagnostic on real brain-to-feature pairs, but a full empirical study lies outside the current scope. revision: partial

  2. Referee: [Naive multivariate linear regression analysis] Analysis of naive regression and output dimension collapse: While the structural property is standard, the manuscript must specify the precise data-scale threshold (with equation) at which collapse becomes the dominant error source versus other factors like noise, to make the 'practical generalization bottleneck' claim load-bearing rather than asymptotic.

    Authors: We accept this suggestion. The manuscript currently emphasizes the asymptotic regime n/d ≪ 1. In the revision we will insert an explicit threshold equation that identifies the data-scale value at which the output-dimension-collapse contribution to the prediction error exceeds the contribution from additive noise. The equation will be expressed in terms of the noise variance, the feature dimensionality, and the regression coefficients, thereby making the practical-bottleneck claim quantitative rather than purely asymptotic. revision: yes

Circularity Check

0 steps flagged

Derivation is self-contained theoretical analysis in student-teacher model

full rationale

The paper derives closed-form prediction error expressions for naive and sparse multivariate linear regression within an explicitly assumed student-teacher generative model, building on standard properties of linear regression (e.g., output dimension collapse when n/d << 1). These expressions are obtained directly from the model assumptions and parameters (data scale, sparsity level) without reducing to fitted quantities, self-citations, or tautological redefinitions. Sparsity is introduced as an input modeling choice rather than derived from the result itself. No load-bearing steps collapse by construction to the paper's own inputs or prior self-citations; the analysis remains externally falsifiable against the stated generative assumptions.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

Central claim depends on the assumption that the brain-to-feature mapping is sparse and on the student-teacher framework setup; full text would be needed to list all free parameters and axioms exhaustively.

free parameters (2)
  • sparsity level
    Parameter controlling variable selection in sparse regression analysis
  • data scale
    Ratio of training samples to latent feature dimensionality, central to characterizing collapse
axioms (2)
  • domain assumption Brain-to-feature mapping is sparse
    Invoked to explain when variable selection reduces prediction error at small scales
  • standard math Standard structural property of naive multivariate regression leads to output dimension collapse
    Used as starting point for analysis of generalization bottleneck

pith-pipeline@v0.9.0 · 5782 in / 1245 out tokens · 49078 ms · 2026-05-18T16:06:22.053100+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

32 extracted references · 32 canonical work pages

  1. [1]

    Allen, Ghislain St-Yves, Yihan Wu, Jesse L

    author Allen, E.J. , author St-Yves, G. , author Wu, Y. , author Breedlove, J.L. , author Prince, J.S. , author Dowdle, L.T. , author Nau, M. , author Caron, B. , author Pestilli, F. , author Charest, I. , author Hutchinson, J.B. , author Naselaris, T. , author Kay, K. , year 2022 . title A massive 7t fmri dataset to bridge cognitive neuroscience and arti...

    work page doi:10.1038/s41593-021-00962-x 2022

  2. [2]

    , author Chernozhukov, V

    author Belloni, A. , author Chernozhukov, V. , year 2013 . title Least squares after model selection in high-dimensional sparse models . journal Bernoulli volume 19 , pages 521--547 . :10.3150/11-BEJ410

    work page doi:10.3150/11-bej410 2013

  3. [3]

    & Chen, C

    author Chen, Z. , author Qing, J. , author Xiang, T. , author Yue, W.L. , author Zhou, J.H. , year 2023 . title Seeing beyond the brain: Conditional diffusion model with sparse masked modeling for vision decoding , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. pages 22710--22720 . :10.1109/CVPR52729....

    work page doi:10.1109/cvpr52729.2023.02175 2023

  4. [4]

    , author Horikawa, T

    author Cheng, F.L. , author Horikawa, T. , author Majima, K. , author Tanaka, M. , author Abdelhack, M. , author Aoki, S.C. , author Hirano, J. , author Kamitani, Y. , year 2023 . title Reconstructing visual illusory experiences from human brain activity . journal Science Advances volume 9 . :10.1126/sciadv.adj3906

    work page doi:10.1126/sciadv.adj3906 2023

  5. [5]

    , author Albrecht, D.G

    author De Valois, R.L. , author Albrecht, D.G. , author Thorell, L.G. , year 1982 . title Spatial frequency selectivity of cells in macaque visual cortex . journal Vision Research volume 22 , pages 545--559 . :10.1016/0042-6989(82)90113-4

    work page doi:10.1016/0042-6989(82)90113-4 1982

  6. [6]

    ImageNet:

    author Deng, J. , author Dong, W. , author Socher, R. , author Li, L.J. , author Li, K. , author Fei-Fei, L. , year 2009 . title Imagenet: A large-scale hierarchical image database , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. pages 248--255 . :10.1109/CVPR.2009.5206848

    work page doi:10.1109/cvpr.2009.5206848 2009

  7. [7]

    , year 2016

    author Dicker, L.H. , year 2016 . title Ridge regression and asymptotic minimax estimation over spheres of growing dimension . journal Bernoulli volume 22 , pages 1--37 . :10.3150/14-BEJ609

    work page doi:10.3150/14-bej609 2016

  8. [8]

    Compressed Sensing

    author Donoho, D.L. , year 2006 . title Compressed sensing . journal IEEE Transactions on Information Theory volume 52 , pages 1289--1306 . :10.1109/TIT.2006.871582

    work page doi:10.1109/tit.2006.871582 2006

  9. [9]

    and Lv, J

    author Fan, J. , author Lv, J. , year 2008 . title Sure independence screening for ultrahigh dimensional feature space . journal Journal of the Royal Statistical Society: Series B (Statistical Methodology) volume 70 , pages 849--911 . :10.1111/j.1467-9868.2008.00674.x

    work page doi:10.1111/j.1467-9868.2008.00674.x 2008

  10. [10]

    , author Elisseeff, A

    author Guyon, I. , author Elisseeff, A. , year 2003 . title An introduction to variable and feature selection . journal Journal of Machine Learning Research volume 3 , pages 1157--1182 . :10.1162/153244303322753616

    work page doi:10.1162/153244303322753616 2003

  11. [11]

    , year 2000

    author Hall, M.A. , year 2000 . title Correlation-based feature selection of discrete and numeric class machine learning , in: booktitle Proceedings of the Seventeenth International Conference on Machine Learning , address Stanford, CA, USA . pp. pages 359--366

    work page 2000

  12. [12]

    Hebart, Oliver Contier, Lina Teichmann, Adam H

    author Hebart, M.N. , author Contier, O. , author Teichmann, L. , author Rockter, A.H. , author Zheng, C.Y. , author Kidder, A. , author Corriveau, A. , author Vaziri-Pashkam, M. , author Baker, C.I. , year 2023 . title Things-data, a multimodal collection of large-scale datasets for investigating object representations in human brain and behavior . journ...

    work page doi:10.7554/elife.82580 2023

  13. [13]

    the visual cortex

    author Hubel, D.H. , author Wiesel, T.N. , year 1962 . title Receptive fields, binocular interaction and functional architecture in the cat's visual cortex . journal The Journal of Physiology volume 160 , pages 106--154 . :10.1113/jphysiol.1962.sp006837

    work page doi:10.1113/jphysiol.1962.sp006837 1962

  14. [14]

    , author Gabriel, F

    author Jacot, A. , author Gabriel, F. , author Hongler, C. , year 2018 . title Neural tangent kernel: Convergence and generalization in neural networks , in: booktitle Advances in Neural Information Processing Systems , publisher Curran Associates, Inc

    work page 2018

  15. [15]

    , author John, G.H

    author Kohavi, R. , author John, G.H. , year 1997 . title Wrappers for feature subset selection . journal Artificial Intelligence volume 97 , pages 273--324 . :10.1016/S0004-3702(97)00043-X

    work page doi:10.1016/s0004-3702(97)00043-x 1997

  16. [16]

    , year 1992

    author MacKay, D.J.C. , year 1992 . title Bayesian interpolation . journal Neural Computation volume 4 , pages 415--447 . :10.1162/neco.1992.4.3.415

    work page doi:10.1162/neco.1992.4.3.415 1992

  17. [17]

    , author Uchida, H

    author Miyawaki, Y. , author Uchida, H. , author Yamashita, O. , author Sato, M. , author Morito, Y. , author Tanabe, H.C. , author Sadato, N. , author Kamitani, Y. , year 2008 . title Visual image reconstruction from human brain activity using a combination of multiscale local image decoders . journal Neuron volume 60 , pages 915--929 . :10.1016/j.neuron...

    work page doi:10.1016/j.neuron.2008.11.004 2008

  18. [18]

    , author Majima, K

    author Nonaka, S. , author Majima, K. , author Aoki, S.C. , author Kamitani, Y. , year 2021 . title Brain hierarchy score: Which deep neural networks are hierarchically brain-like? journal iScience volume 24 . :10.1016/j.isci.2021.103013

    work page doi:10.1016/j.isci.2021.103013 2021

  19. [19]

    Natural scene reconstruction from fmri signals using generative latent diffusion

    author Ozcelik, F. , author VanRullen, R. , year 2023 . title Natural scene reconstruction from fmri signals using generative latent diffusion . journal Scientific Reports volume 13 . :10.1038/s41598-023-42891-8

    work page doi:10.1038/s41598-023-42891-8 2023

  20. [20]

    , author Kubilius, J

    author Schrimpf, M. , author Kubilius, J. , author Lee, M.J. , author Ratan Murty, N.A. , author Ajemian, R. , author DiCarlo, J.J. , year 2020 . title Integrative benchmarking to advance neurally mechanistic models of human intelligence . journal Neuron volume 108 , pages 413--423 . :10.1016/j.neuron.2020.07.040

    work page doi:10.1016/j.neuron.2020.07.040 2020

  21. [21]

    , author Banerjee, A

    author Scotti, P.S. , author Banerjee, A. , author Goode, J. , author Shabalin, S. , author Nguyen, A. , author Cohen, E. , author Dempster, A.J. , author Verlinde, N. , author Yundler, E. , author Weisberg, D. , author Norman, K.A. , author Abraham, T.M. , year 2023 . title Reconstructing the mind’s eye: fmri-to-image with contrastive learning and diffus...

    work page 2023

  22. [22]

    , author Tripathy, M

    author Scotti, P.S. , author Tripathy, M. , author Torrico, C.K. , author Kneeland, R. , author Chen, T. , author Narang, A. , author Santhirasegaran, C. , author Xu, J. , author Naselaris, T. , author Norman, K.A. , author Abraham, T.M. , year 2024 . title Mindeye2: Shared-subject models enable fmri-to-image with 1 hour of data , in: editor Salakhutdinov...

    work page 2024

  23. [23]

    , author Güçlü, U

    author Seeliger, K. , author Güçlü, U. , author Ambrogioni, L. , author Güçlütürk, Y. , author van Gerven, M.A.J. , year 2018 . title Generative adversarial networks for reconstructing natural images from brain activity . journal NeuroImage volume 181 , pages 775--785 . :10.1016/j.neuroimage.2018.07.043

    work page doi:10.1016/j.neuroimage.2018.07.043 2018

  24. [24]

    doi: 10.1371/journal.pcbi.1006633

    author Shen, G. , author Horikawa, T. , author Majima, K. , author Kamitani, Y. , year 2019 . title Deep image reconstruction from human brain activity . journal PLOS Computational Biology volume 15 . :10.1371/journal.pcbi.1006633

    work page doi:10.1371/journal.pcbi.1006633 2019

  25. [25]

    , author Nagano, Y

    author Shirakawa, K. , author Nagano, Y. , author Tanaka, M. , author Aoki, S.C. , author Muraki, Y. , author Majima, K. , author Kamitani, Y. , year 2025 . title Spurious reconstruction from brain activity . journal Neural Networks volume 190 . :10.1016/j.neunet.2025.107515

    work page doi:10.1016/j.neunet.2025.107515 2025

  26. [26]

    & Chen, C

    author Takagi, Y. , author Nishimoto, S. , year 2023 . title High-resolution image reconstruction with latent diffusion models from human brain activity , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. pages 14453--14463 . :10.1109/CVPR52729.2023.01389

    work page doi:10.1109/cvpr52729.2023.01389 2023

  27. [27]

    Journal of the Royal Statistical Society Series B (Methodological) 58:509--523

    author Tibshirani, R. , year 1996 . title Regression shrinkage and selection via the lasso . journal Journal of the Royal Statistical Society: Series B (Methodological) volume 58 , pages 267--288 . :10.1111/j.2517-6161.1996.tb02080.x

    work page doi:10.1111/j.2517-6161.1996.tb02080.x 1996

  28. [28]

    , author Vedaldi, A

    author Ulyanov, D. , author Vedaldi, A. , author Lempitsky, V. , year 2018 . title Deep image prior , in: booktitle Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pp. pages 9446--9454 . :10.1109/CVPR.2018.00984

    work page doi:10.1109/cvpr.2018.00984 2018

  29. [29]

    , author Gallant, J.L

    author Vinje, W.E. , author Gallant, J.L. , year 2000 . title Sparse coding and decorrelation in primary visual cortex during natural vision . journal Science volume 287 , pages 1273--1276 . :10.1126/science.287.5456.1273

    work page doi:10.1126/science.287.5456.1273 2000

  30. [30]

    , year 2009 a

    author Wainwright, M.J. , year 2009 a. title Information-theoretic limits on sparsity recovery in the high-dimensional and noisy setting . journal IEEE Transactions on Information Theory volume 55 , pages 5728--5741 . :10.1109/TIT.2009.2032816

    work page doi:10.1109/tit.2009.2032816 2009

  31. [31]

    , year 2009 b

    author Wainwright, M.J. , year 2009 b. title Sharp thresholds for high-dimensional and noisy sparsity recovery using _1 -constrained quadratic programming (lasso) . journal IEEE Transactions on Information Theory volume 55 , pages 2183--2202 . :10.1109/TIT.2009.2016018

    work page doi:10.1109/tit.2009.2016018 2009

  32. [32]

    , author Qiao, K

    author Zhang, C. , author Qiao, K. , author Wang, L. , author Tong, L. , author Zeng, Y. , author Yan, B. , year 2018 . title Constraint-free natural image reconstruction from fmri signals based on convolutional neural network . journal Frontiers in Human Neuroscience volume 12 . :10.3389/fnhum.2018.00242

    work page doi:10.3389/fnhum.2018.00242 2018