pith. sign in

arxiv: 2605.18251 · v1 · pith:NZBXHGTTnew · submitted 2026-05-18 · 📡 eess.SP · cs.LG· q-bio.NC

Subject-Specific Analysis of Self-Initiated Attention Shifts from EEG with Controlled Internal and External Attention Conditions

Yuwen Zeng , Dengzhe Hou , Zhang Zhang , Sai Sun , Yongsong Huang , Chia-huei Tseng , Satoshi Shioiri This is my paper

Pith reviewed 2026-05-20 00:21 UTC · model grok-4.3

classification 📡 eess.SP cs.LGq-bio.NC
keywords EEG classificationself-initiated attentionpreparatory activitymachine learningSHAP analysisbrain-machine interfacessubject-specific patternsattention shifts
0
0 comments X

The pith

Preparatory EEG activity before attention shifts carries subject-specific information that distinguishes self-initiated from externally triggered changes.

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

The authors test whether brain signals recorded just before a person shifts their attention can reveal if the shift was started by the person themselves or prompted from outside. They use an experimental setup where the visual scene stays the same in both cases, so the only real difference is who starts the shift. By training machine learning models on these preparatory signals for each person separately, they find consistent success in telling the two apart. This matters for understanding the brain basis of voluntary actions and for building interfaces that can detect internal intentions without waiting for external commands. Their follow-up analysis highlights that signals in higher frequency ranges and from the front of the brain are particularly useful for the models, though artifacts could play a role there.

Core claim

In this controlled comparison of self-initiated and externally instructed attention shifts under identical visual conditions, preparatory EEG activity enables reliable within-subject classification. Frequency-specific topographic patterns and SHAP feature attribution reveal that higher-frequency bands and frontal regions contribute strongly to these model decisions. The findings indicate that such preparatory activity holds subject-specific discriminative information suitable for applications like personalized asynchronous brain-machine interfaces.

What carries the argument

Machine learning classification of frequency-specific EEG topographic patterns using SHAP for feature attribution, applied to preparatory activity in a paradigm that controls for visual stimulation.

If this is right

  • Reliable within-subject classification performance demonstrates subject-specific discriminative information in preparatory EEG.
  • Higher-frequency bands and frontal regions make strong contributions to model decisions.
  • The approach has potential applications in personalized and asynchronous brain-machine interface systems.

Where Pith is reading between the lines

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

  • Extending the paradigm to detect self-initiated actions in everyday settings without fixed visual stimuli could test the generality of these signals.
  • Integrating these subject-specific models into real-time systems might enable more natural control in brain-computer interfaces.
  • Comparing these findings with other voluntary behaviors, such as motor preparation, could reveal shared or distinct neural signatures.

Load-bearing premise

The experimental setup keeps visual stimulation identical so that any differences in EEG come primarily from whether the attention shift was self-initiated or externally instructed.

What would settle it

A direct test would be to run the same within-subject classification on a fresh group of participants using the identical paradigm and observe whether accuracies remain significantly above chance levels.

Figures

Figures reproduced from arXiv: 2605.18251 by Chia-huei Tseng, Dengzhe Hou, Sai Sun, Satoshi Shioiri, Yongsong Huang, Yuwen Zeng, Zhang Zhang.

Figure 1
Figure 1. Figure 1: Electrode distribution B. EEG Segmentation and Band-Specific Representations EEG preprocessing was conducted according to the pipeline described in the original study [11]. For each attention shift, EEG segments were extracted from −2.0 s to −0.5 s relative to the shift onset. This time window was selected to capture preparatory activity while reducing potential contamination from eye-movement artifacts. T… view at source ↗
Figure 2
Figure 2. Figure 2: ROC curves and AUC values for within-subject classification [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Topographic patterns for the multi-band setting and the five [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Example SHAP summary plot for a representative participant in [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: SHAP-based topographic patterns for the multi-band setting [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
read the original abstract

Self-initiated attention shifts play a critical role in voluntary behavior but are difficult to study due to the absence of explicit temporal markers. While previous studies have examined their neural correlates, it remains unclear how multi-dimensional electroencephalography (EEG) features contribute to their characterization within an interpretable computational framework. In this study, we build on an experimental paradigm developed in our previous work, which enables controlled comparison between task-constrained self-initiated shifts and externally instructed shifts under identical visual stimulation. Within this setting, we investigate whether preparatory EEG activity can distinguish these two types of attention shifts. We adopt a machine learning-based approach and conduct two complementary analyses: (1) a performance-oriented assessment of frequency-specific topographic patterns, and (2) a model-based feature attribution analysis using SHapley Additive exPlanations (SHAP). These analyses provide a structured view of how spectral features across regions of interest contribute to model behavior. Our results demonstrate reliable within-subject classification performance, indicating that preparatory EEG activity contains subject-specific discriminative information within this paradigm. The analysis shows that higher-frequency bands and frontal regions contribute strongly to model decisions, although such contributions should be interpreted cautiously due to the potential influence of non-neural artifacts in high-frequency EEG signals. Overall, this work highlights the value of interpretable machine learning for analyzing subject-specific EEG signal patterns in a controlled experimental setting, with potential applications in personalized and asynchronous brain-machine interface systems.

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 paper develops a machine-learning pipeline to classify preparatory EEG activity preceding self-initiated versus externally instructed attention shifts within a controlled visual paradigm. It reports reliable within-subject classification accuracies and uses SHAP values to identify dominant contributions from higher-frequency bands and frontal electrodes, while noting the need for cautious interpretation due to possible non-neural artifacts.

Significance. If the reported discriminative information is shown to arise from neural rather than artifactual sources, the work would supply a concrete, subject-specific feature set for asynchronous BCI applications and a template for interpretable analysis of voluntary behavior in EEG.

major comments (2)
  1. [Abstract] Abstract and Results: The SHAP analysis attributes strong importance to higher-frequency bands, yet the manuscript provides no quantitative ablation (e.g., performance drop when bands >30 Hz are removed) to demonstrate that classification remains reliable on lower-frequency neural signals alone.
  2. [Methods] Methods: Artifact rejection and trial-exclusion criteria are described at a general level but do not address potential systematic differences in subtle EMG or EOG activity between self-initiated and externally cued trials; such differences could inflate within-subject performance without reflecting preparatory neural activity.
minor comments (2)
  1. [Figures] Figure captions and axis labels should explicitly state the frequency ranges used for each topographic map to improve reproducibility.
  2. [Results] The manuscript would benefit from a short table summarizing per-subject accuracies, chance levels, and number of trials retained after preprocessing.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and positive overall assessment. We address each major comment below and outline the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract] Abstract and Results: The SHAP analysis attributes strong importance to higher-frequency bands, yet the manuscript provides no quantitative ablation (e.g., performance drop when bands >30 Hz are removed) to demonstrate that classification remains reliable on lower-frequency neural signals alone.

    Authors: We agree that a quantitative ablation would strengthen the interpretation. In the revised manuscript we will add an ablation analysis retraining the classifiers on features restricted to bands below 30 Hz and report the resulting within-subject accuracies together with the corresponding SHAP attributions. revision: yes

  2. Referee: [Methods] Methods: Artifact rejection and trial-exclusion criteria are described at a general level but do not address potential systematic differences in subtle EMG or EOG activity between self-initiated and externally cued trials; such differences could inflate within-subject performance without reflecting preparatory neural activity.

    Authors: We acknowledge that the current Methods section describes artifact handling at a general level. In revision we will expand this section with the precise rejection thresholds and trial-exclusion rules used, and we will add a supplementary comparison of EOG and high-frequency EMG power between the two conditions to evaluate possible systematic differences. revision: yes

Circularity Check

0 steps flagged

No circularity: results derived from new EEG recordings and ML training

full rationale

The paper collects fresh subject-specific EEG data under a controlled paradigm, trains classifiers on frequency-band and topographic features from that data, and reports empirical within-subject accuracies plus SHAP attributions. The reference to prior work is confined to describing the experimental setup (identical visual stimulation, self-initiated vs. instructed shifts); it supplies no fitted parameters, no pre-computed predictions, and no uniqueness theorem that the current classification performance is forced to reproduce. The derivation chain therefore terminates in independent measurements and standard supervised learning rather than reducing to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Based solely on the abstract, the paper relies on standard domain assumptions in EEG research and the validity of the prior experimental paradigm; no explicit free parameters or new invented entities are described.

axioms (2)
  • domain assumption The experimental paradigm provides identical visual stimulation across self-initiated and externally instructed attention-shift conditions
    Stated in the abstract as the basis for controlled comparison.
  • domain assumption Preparatory EEG activity contains information that precedes and can distinguish the type of attention shift
    Implicit foundation for analyzing pre-shift signals with ML.

pith-pipeline@v0.9.0 · 5816 in / 1447 out tokens · 71389 ms · 2026-05-20T00:21:58.409358+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.

  • IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    We adopt a machine learning-based approach and conduct two complementary analyses: (1) a performance-oriented assessment of frequency-specific topographic patterns, and (2) a model-based feature attribution analysis using SHapley Additive exPlanations (SHAP).

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

26 extracted references · 26 canonical work pages · 1 internal anchor

  1. [1]

    The attention system of the human brain: 20 years after,

    S. E. Petersen and M. I. Posner, “The attention system of the human brain: 20 years after,”Annu. Rev. Neurosci., vol. 35, pp. 73–89, 2012

    work page 2012

  2. [2]

    Placing willed attention in context: A review of attention and free will,

    J. G. Nadra and G. R. Mangun, “Placing willed attention in context: A review of attention and free will,”Front. Cognit., vol. 2, Art. no. 1205618, 2023

    work page 2023

  3. [3]

    Visual attention spreads broadly but selects information locally,

    S. Shioiri, H. Honjyo, Y . Kashiwase, K. Matsumiya, and I. Kuriki, “Visual attention spreads broadly but selects information locally,”Sci. Rep., vol. 6, Art. no. 35513, 2016

    work page 2016

  4. [4]

    Time courses of attentional modulation in neural amplification and synchronization measured with steady-state visual-evoked potentials,

    Y . Kashiwase, K. Matsumiya, I. Kuriki, and S. Shioiri, “Time courses of attentional modulation in neural amplification and synchronization measured with steady-state visual-evoked potentials,”J. Cogn. Neu- rosci., vol. 24, no. 8, pp. 1779–1793, 2012

    work page 2012

  5. [5]

    Attention without constraint: Alpha lateralization in uncued willed attention,

    J. G. Nadra, J. J. Bengson, A. B. Morales, and G. R. Mangun, “Attention without constraint: Alpha lateralization in uncued willed attention,”eNeuro, vol. 10, no. 6, 2023

    work page 2023

  6. [6]

    Critical brain states related with self-initiated attentional shift,

    W. Wu et al., “Critical brain states related with self-initiated attentional shift,”J. Vis., vol. 22, no. 14, p. 3883, 2022

    work page 2022

  7. [7]

    Investigation of critical brain states to initiate attentional shift,

    W. Wu et al., “Investigation of critical brain states to initiate attentional shift,”IEICE Tech. Rep., vol. 121, no. 312, pp. 49–52, 2021

    work page 2021

  8. [8]

    Alpha- band electroencephalographic activity over occipital cortex indexes visuospatial attention bias and predicts visual target detection,

    G. Thut, A. Nietzel, S. A. Brandt, and A. Pascual-Leone, “Alpha- band electroencephalographic activity over occipital cortex indexes visuospatial attention bias and predicts visual target detection,”J. Neurosci., vol. 26, no. 37, pp. 9494–9502, 2006

    work page 2006

  9. [9]

    Anticipatory biasing of visuospatial attention indexed by retinotopically specific alpha-band electroencephalography over occipital cortex,

    M. S. Worden, J. J. Foxe, N. Wang, and G. V . Simpson, “Anticipatory biasing of visuospatial attention indexed by retinotopically specific alpha-band electroencephalography over occipital cortex,”J. Neurosci., vol. 20, no. 6, pp. 1–6, 2000

    work page 2000

  10. [10]

    Spontaneous alpha and theta oscillations are related to cognitive control,

    G. M. Clements, C. H. Bowie, E. M. Garofalo, and M. J. Kuhlmann, “Spontaneous alpha and theta oscillations are related to cognitive control,”Front. Hum. Neurosci., vol. 15, Art. no. 621620, 2021

    work page 2021

  11. [11]

    Task-constrained self-initiated attention shifts are indexed by frontal-midline theta ramping,

    D. Hou, S. Sun, Y . Hatori, C. Tseng, and S. Shioiri, “Task-constrained self-initiated attention shifts are indexed by frontal-midline theta ramping,”Front. Hum. Neurosci., vol. 19, Art. no. 1708257, 2025

    work page 2025

  12. [12]

    Spontaneous neural fluctuations predict decisions to attend,

    J. J. Bengson, T. A. Kelley, X. Zhang, J. L. Wang, and G. R. Mangun, “Spontaneous neural fluctuations predict decisions to attend,”J. Cogn. Neurosci., vol. 26, no. 11, pp. 2578–2584, 2014

    work page 2014

  13. [13]

    An accumulator model for spontaneous neural activity prior to self-initiated movement,

    A. Schurger, J. D. Sitt, and S. Dehaene, “An accumulator model for spontaneous neural activity prior to self-initiated movement,”Proc. Natl. Acad. Sci. U.S.A., vol. 109, no. 42, pp. E2904–E2913, 2012

    work page 2012

  14. [14]

    Neural decoding of EEG signals with machine learning: A systematic review,

    M. Saeidi et al., “Neural decoding of EEG signals with machine learning: A systematic review,”Brain Sci., vol. 11, no. 11, p. 1525, 2021

    work page 2021

  15. [15]

    Attention detection using EEG signals and machine learning: A review,

    Q. Sun, Y . Zhou, P. Gong, and D. Zhang, “Attention detection using EEG signals and machine learning: A review,”Mach. Intell. Res., vol. 22, no. 2, pp. 219–238, 2025

    work page 2025

  16. [16]

    A review of classification algorithms for EEG-based brain–computer interfaces: A 10-year update,

    F. Lotte et al., “A review of classification algorithms for EEG-based brain–computer interfaces: A 10-year update,”J. Neural Eng., vol. 15, no. 3, p. 031005, 2018

    work page 2018

  17. [17]

    Explainable artificial intelligence and neuroscience: Opportunities and challenges,

    J. M. Fellous, R. S. H. Smith, and P. E. Latham, “Explainable artificial intelligence and neuroscience: Opportunities and challenges,”Front. Neurosci., vol. 13, Art. no. 1346, 2019

    work page 2019

  18. [18]

    A review of artificial intelligence for EEG-based brain- computer interfaces and applications,

    Z. Cao, “A review of artificial intelligence for EEG-based brain- computer interfaces and applications,”Brain Sci. Adv., vol. 6, no. 3, pp. 162–170, 2020

    work page 2020

  19. [19]

    A unified approach to interpreting model predictions,

    S. M. Lundberg and S.-I. Lee, “A unified approach to interpreting model predictions,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2017, pp. 4765–4774

    work page 2017

  20. [20]

    Motor imagery classification using Mu and Beta rhythms of EEG with strong uncorrelating transform based complex common spatial patterns,

    J. Kim and C. Guan, “Motor imagery classification using Mu and Beta rhythms of EEG with strong uncorrelating transform based complex common spatial patterns,”Comput. Intell. Neurosci., vol. 2016, Art. no. 1489692, 2016

    work page 2016

  21. [21]

    EEG signal analysis for seizure detection using discrete wavelet transform and random forest,

    S. Bose, V . Rama, and C. R. Rao, “EEG signal analysis for seizure detection using discrete wavelet transform and random forest,” inProc. Int. Conf. Comput. Appl. (ICCA), 2017, pp. 369–378

    work page 2017

  22. [22]

    Feature extraction and classification of motor imagery EEG signals using random forests,

    S. Lu, Y . Zhang, and X. Li, “Feature extraction and classification of motor imagery EEG signals using random forests,”Sustainability, vol. 16, no. 4, p. 2201, 2024

    work page 2024

  23. [23]

    Exploring EEG features in cross- subject emotion recognition,

    X. Li, H. Song, and Y . Zhang, “Exploring EEG features in cross- subject emotion recognition,”Front. Neurosci., vol. 12, Art. no. 162, 2018

    work page 2018

  24. [24]

    A comparative study of subject-dependent and subject-independent strategies for EEG- based emotion recognition using LSTM network,

    D. Nath, S. Ghosh, A. Saha, and A. Konar, “A comparative study of subject-dependent and subject-independent strategies for EEG- based emotion recognition using LSTM network,” inProc. Int. Conf. Comput. Sci. Appl. (ICCDA), 2020, pp. 142–147

    work page 2020

  25. [25]

    A large EEG dataset for studying cross-session variability in motor imagery brain–computer interface,

    J. Ma, J. Wang, Y . Zhang, and D. Zhang, “A large EEG dataset for studying cross-session variability in motor imagery brain–computer interface,”Sci. Data, vol. 9, no. 1, p. 531, 2022

    work page 2022

  26. [26]

    Same Brain, Different Prediction: How Preprocessing Choices Undermine EEG Decoding Reliability

    D. Hou, Z. Wu, L. Jiang, Z. Li, F. Lin, and K. D. Yamada, “Same brain, different prediction: How preprocessing choices undermine EEG decoding reliability,”arXiv preprint arXiv:2605.07212, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026