pith. sign in

arxiv: 1906.12077 · v1 · pith:DZYGDFYCnew · submitted 2019-06-28 · 📊 stat.ML · cs.LG· eess.SP· math.OC· stat.AP· stat.CO

Large scale Lasso with windowed active set for convolutional spike sorting

Laurent Dragoni , R\'emi Flamary , Karim Lounici , Patricia Reynaud-Bouret This is my paper

Pith reviewed 2026-05-25 13:56 UTC · model grok-4.3

classification 📊 stat.ML cs.LGeess.SPmath.OCstat.APstat.CO
keywords spike sortingLassoactive setconvolutional modellarge-scale optimizationneuroscienceparallel algorithmssignal recovery
0
0 comments X

The pith

A windowed active-set algorithm solves the Lasso for convolutional spike sorting with linear time complexity.

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

The paper develops a new active-set solver for the Lasso problem under a convolutional forward model that is standard in spike sorting. The goal is to handle the very large datasets that arise when recording from many electrodes over long periods, which exceed the reach of conventional Lasso solvers. The proposed method restricts active-set operations to sliding temporal windows, yielding linear scaling in signal length and straightforward parallel execution across time segments. Theoretical analysis gives complexity bounds, while experiments check that recovered spikes remain accurate and stable under changes in the regularization strength. If the approach works as described, spike sorting becomes feasible for recordings whose size previously made them intractable.

Core claim

The authors introduce a windowed active-set procedure that solves the convolutional Lasso exactly while attaining linear complexity in the temporal dimension and admitting efficient parallel implementation across time windows. Complexity guarantees are derived, and numerical tests confirm that the recovered spike times and amplitudes match the accuracy of standard active-set methods on moderate-sized problems.

What carries the argument

The windowed active-set procedure applied to the convolutional Lasso, which limits variable activation searches to local temporal windows to enforce linear scaling.

If this is right

  • Spike sorting can be performed on signals whose length grows linearly rather than quadratically in runtime.
  • The procedure maps naturally onto parallel hardware because windows can be processed independently.
  • Theoretical bounds establish that the number of active-set iterations remains controlled by the window size.
  • Numerical recovery of spike times stays accurate across a range of regularization values.
  • The same solver can be used for online processing as new data arrive.

Where Pith is reading between the lines

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

  • Real-time spike sorting during live experiments becomes practical if the linear scaling holds on hardware.
  • The windowing idea could be tested on other sparse-recovery problems that have a convolutional or Toeplitz structure.
  • Closed-loop neuroscience setups that require immediate feedback on neural activity could adopt the method directly.
  • Performance on real rather than simulated electrophysiological recordings would be the next concrete check.

Load-bearing premise

The convolutional model faithfully represents the observed electrode signals and the window restriction on the active set preserves the exact same solution that an unrestricted active-set method would return.

What would settle it

A controlled experiment on a synthetic dataset whose true spikes are known, comparing the windowed solver against an exact (but slower) Lasso solver and checking whether any spikes are missed or added once the recording length exceeds a few thousand samples.

Figures

Figures reproduced from arXiv: 1906.12077 by Karim Lounici, Laurent Dragoni, Patricia Reynaud-Bouret, R\'emi Flamary.

Figure 1
Figure 1. Figure 1: Illustration of the convolutive model with [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Influence of λ and the signal-to-noise ratio on the performances of the Lasso estimator. Results are averaged over 5 draws. B. Influence of the noise and the regularization parameter Proper calibration of the regularization parameter λ is crucial for the success of the estimation. We want to visualize the influence of this choice, especially for various noise levels. Using real shapes of action potentials … view at source ↗
read the original abstract

Spike sorting is a fundamental preprocessing step in neuroscience that is central to access simultaneous but distinct neuronal activities and therefore to better understand the animal or even human brain. But numerical complexity limits studies that require processing large scale datasets in terms of number of electrodes, neurons, spikes and length of the recorded signals. We propose in this work a novel active set algorithm aimed at solving the Lasso for a classical convolutional model. Our algorithm can be implemented efficiently on parallel architecture and has a linear complexity w.r.t. the temporal dimensionality which ensures scaling and will open the door to online spike sorting. We provide theoretical results about the complexity of the algorithm and illustrate it in numerical experiments along with results about the accuracy of the spike recovery and robustness to the regularization parameter.

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 proposes a windowed active-set algorithm to solve the Lasso problem for a convolutional model in the context of large-scale spike sorting. It asserts that the method achieves linear complexity in the temporal dimension, supports efficient parallel implementation, includes theoretical complexity results, and demonstrates accuracy in spike recovery along with robustness to the regularization parameter via numerical experiments.

Significance. If the central claims on exact Lasso recovery and linear complexity hold without approximation errors from windowing, the work would enable scalable processing of high-dimensional neural recordings and potentially support online spike sorting, addressing a key bottleneck in neuroscience data analysis. The provision of numerical validation on accuracy is a positive aspect.

major comments (2)
  1. [Algorithm section] Algorithm description (likely §3 or equivalent): the claim that the windowed active-set procedure preserves the exact optimality conditions and solution properties of standard Lasso active-set methods is central to the accuracy and recovery guarantees, yet no explicit verification or proof is supplied showing that windowing introduces no approximation errors affecting the recovered spikes.
  2. [Theoretical results] Theoretical complexity results (likely §4): the asserted linear complexity w.r.t. temporal dimensionality and parallel efficiency lack detailed derivation steps or bounds that would allow independent verification, particularly regarding dependence on window size and number of electrodes; this underpins the scaling claims.
minor comments (2)
  1. [Abstract and Experiments] The abstract and introduction would benefit from explicit dataset sizes, number of channels, and spike counts used in the numerical experiments to allow reproducibility assessment.
  2. [Model and Algorithm] Notation for the convolutional model and active-set updates could be clarified with a small example or pseudocode to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed review and constructive comments. We address each major point below and will revise the manuscript to strengthen the presentation of the theoretical guarantees.

read point-by-point responses

  1. Referee: [Algorithm section] Algorithm description (likely §3 or equivalent): the claim that the windowed active-set procedure preserves the exact optimality conditions and solution properties of standard Lasso active-set methods is central to the accuracy and recovery guarantees, yet no explicit verification or proof is supplied showing that windowing introduces no approximation errors affecting the recovered spikes.

    Authors: We agree that an explicit verification is needed to confirm equivalence. The window selection rule in the algorithm is designed to ensure that all active-set updates and correlation checks remain identical to the full problem, but the original submission did not include a formal argument. In revision we will add a dedicated subsection (or appendix) proving that the windowed procedure satisfies the same KKT conditions and yields the identical sparse solution as the standard active-set method, with no approximation error. revision: yes

  2. Referee: [Theoretical results] Theoretical complexity results (likely §4): the asserted linear complexity w.r.t. temporal dimensionality and parallel efficiency lack detailed derivation steps or bounds that would allow independent verification, particularly regarding dependence on window size and number of electrodes; this underpins the scaling claims.

    Authors: The complexity claims rest on the fact that each window processes a fixed-size subproblem whose cost is independent of total recording length, yielding overall linear scaling; parallel efficiency follows from independent window processing. We acknowledge that the original text provides only a high-level outline. In the revision we will expand §4 with complete derivation steps, explicit big-O bounds, and the precise dependence on window size, electrode count, and sparsity level to allow independent verification. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper introduces a windowed active-set algorithm to solve the Lasso problem under a convolutional spike-sorting model. The abstract and supplied description present the method as a direct algorithmic construction whose complexity and exactness properties are asserted to follow from the active-set framework and parallel implementation. No equations or steps are shown that reduce a claimed prediction or uniqueness result to a fitted parameter or prior self-citation by construction. The central claims rest on the algorithmic design itself together with external numerical validation, satisfying the criteria for a self-contained derivation.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Review performed from abstract only; the central modeling choice is the convolutional Lasso formulation itself.

free parameters (1)
  • regularization parameter
    Abstract reports experiments on robustness to this parameter, indicating it is a tunable input whose value affects the reported recovery accuracy.
axioms (1)
  • domain assumption Spike sorting can be accurately cast as a convolutional Lasso inverse problem.
    This modeling premise underpins the entire algorithmic development described in the abstract.

pith-pipeline@v0.9.0 · 5675 in / 1168 out tokens · 33428 ms · 2026-05-25T13:56:45.643753+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages · 2 internal anchors

  1. [1]

    Surrogate data methods based on a shuffling of the trials for synchrony detection : the centering issue,

    M. Albert, Y . Bouret, M. Fromont, and P. Reynaud-Bouret, “Surrogate data methods based on a shuffling of the trials for synchrony detection : the centering issue,” Neural Computation, vol. 28, no. 11, pp. 2352– 2392, 2016

    work page 2016

  2. [2]

    Reconstructing the functional connectivity of multiple spike trains using hawkes models,

    R. Lambert, C. Tuleau-Malot, T. Bessaih, V . Rivoirard, Y . Bouret, N. Leresche, and P. Reynaud-Bouret, “Reconstructing the functional connectivity of multiple spike trains using hawkes models,” Journal of Neuroscience Methods, vol. 297, pp. 9–21, 2018

    work page 2018

  3. [3]

    Dynamics and effective topology underlying synchronization in networks of cortical neurons,

    D. Eytan and S. Marom, “Dynamics and effective topology underlying synchronization in networks of cortical neurons,” Journal of Neuroscience , vol. 26, no. 33, pp. 8465–8476, 2006. [Online]. Available: http://www.jneurosci.org/content/26/33/8465

    work page 2006

  4. [4]

    A review of methods for spike sorting: the detection and classification of neural action potentials,

    M. S. Lewicki, “A review of methods for spike sorting: the detection and classification of neural action potentials,” Network: Computation in Neural Systems, vol. 9, no. 4, pp. R53–R78, 1998

    work page 1998

  5. [5]

    Recovery of sparse translation-invariant signals with continuous basis pursuit,

    C. Ekanadham, D. Tranchina, and E. P. Simoncelli, “Recovery of sparse translation-invariant signals with continuous basis pursuit,” IEEE transactions on signal processing, vol. 59, no. 10, pp. 4735–4744, 2011

    work page 2011

  6. [6]

    A unified framework and method for automatic neural spike identification,

    ——, “A unified framework and method for automatic neural spike identification,” Journal of neuroscience methods , vol. 222, pp. 47–55, 2014

    work page 2014

  7. [7]

    SPySort: Neuronal Spike Sorting with Python

    C. Pouzat and G. Detorakis, “Spysort: Neuronal spike sorting with python,” CoRR, vol. abs/1412.6383, 2014. [Online]. Available: http://arxiv.org/abs/1412.6383

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  8. [8]

    On the variability of manual spike sorting,

    F. Wood, M. J. Black, C. Vargas-Irwin, M. Fellows, and J. P. Donoghue, “On the variability of manual spike sorting,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 6, pp. 912–918, 2004

    work page 2004

  9. [9]

    Accuracy of tetrode spike separation as determined by simultaneous intracellular and extracellular measurements,

    K. D. Harris, D. A. Henze, J. Csicsvari, H. Hirase, and G. Buzsaki, “Accuracy of tetrode spike separation as determined by simultaneous intracellular and extracellular measurements,” Journal of neurophysiol- ogy, vol. 84, no. 1, pp. 401–414, 2000

    work page 2000

  10. [10]

    Towards reliable spike-train recordings from thousands of neurons with multielectrodes,

    G. T. Einevoll, F. Franke, E. Hagen, C. Pouzat, and K. D. Harris, “Towards reliable spike-train recordings from thousands of neurons with multielectrodes,” Current opinion in neurobiology , vol. 22, no. 1, pp. 11–17, 2012

    work page 2012

  11. [11]

    Involvement of fast-spiking cells in ictal sequences during spontaneous seizures in rats with chronic temporal lobe epilepsy,

    A. R. Neumann, R. Raedt, H. W. Steenland, M. Sprengers, K. Bzymek, Z. Navratilova, L. Mesina, J. Xie, V . Lapointe, F. Kloosterman, K. V onck, P. A. J. M. Boon, I. Soltesz, B. L. McNaughton, and A. Luczak, “Involvement of fast-spiking cells in ictal sequences during spontaneous seizures in rats with chronic temporal lobe epilepsy,” Brain, vol. 140, pp. 23...

    work page 2017

  12. [12]

    Convolutive speech bases and their application to su- pervised speech separation,

    P. Smaragdis, “Convolutive speech bases and their application to su- pervised speech separation,” IEEE Transactions on Audio, Speech, and Language Processing, vol. 15, no. 1, pp. 1–12, Jan 2007

    work page 2007

  13. [13]

    Simultaneous analysis of lasso and dantzig selector,

    P. Bickel, Y . Ritov, and A. Tsybakov, “Simultaneous analysis of lasso and dantzig selector,” Annals of Statistics, vol. 37, no. 4, pp. 1705–1732, 2009

    work page 2009

  14. [14]

    Honest variable selection in linear and logistic regression models via l1 and l1 + l2 penalization,

    F. Bunea, “Honest variable selection in linear and logistic regression models via l1 and l1 + l2 penalization,” Electron. J. Statist. , vol. 2, pp. 1153–1194, 2008. [Online]. Available: https://doi.org/10.1214/ 08-EJS287

    work page 2008

  15. [15]

    Sup-norm convergence rate and sign concentration property of lasso and dantzig estimators,

    K. Lounici, “Sup-norm convergence rate and sign concentration property of lasso and dantzig estimators,” Electron. J. Statist. , vol. 2, pp. 90–102, 2008. [Online]. Available: https://doi.org/10.1214/08-EJS177

    work page doi:10.1214/08-ejs177 2008

  16. [16]

    Learning the Morphology of Brain Signals Using Alpha-Stable Convolutional Sparse Coding,

    M. Jas, T. Dupré La Tour, U. ¸ Sim¸ sekli, and A. Gramfort, “Learning the Morphology of Brain Signals Using Alpha-Stable Convolutional Sparse Coding,” in Advances in neural information processing systems, Long Beach, United States, Dec. 2017. [Online]. Available: https://hal.archives-ouvertes.fr/hal-01590988

    work page 2017

  17. [17]

    Multivariate con- volutional sparse coding for electromagnetic brain signals,

    T. D. La Tour, T. Moreau, M. Jas, and A. Gramfort, “Multivariate con- volutional sparse coding for electromagnetic brain signals,” in Advances in Neural Information Processing Systems , 2018, pp. 3292–3302

    work page 2018

  18. [18]

    Least angle regression,

    B. Efron, T. Hastie, I. Johnstone, R. Tibshirani et al. , “Least angle regression,” The Annals of statistics , vol. 32, no. 2, pp. 407–499, 2004

    work page 2004

  19. [19]

    Coordinate descent algorithms for lasso penalized regression,

    T. T. Wu, K. Lange et al. , “Coordinate descent algorithms for lasso penalized regression,” The Annals of Applied Statistics , vol. 2, no. 1, pp. 224–244, 2008

    work page 2008

  20. [20]

    Proximal splitting methods in signal processing,

    P. L. Combettes and J.-C. Pesquet, “Proximal splitting methods in signal processing,” in Fixed-point algorithms for inverse problems in science and engineering. Springer, 2011, pp. 185–212

    work page 2011

  21. [21]

    Proximal algorithms,

    N. Parikh, S. Boyd et al. , “Proximal algorithms,” Foundations and Trends® in Optimization, vol. 1, no. 3, pp. 127–239, 2014

    work page 2014

  22. [22]

    A fast iterative shrinkage-thresholding algo- rithm for linear inverse problems,

    A. Beck and M. Teboulle, “A fast iterative shrinkage-thresholding algo- rithm for linear inverse problems,” SIAM journal on imaging sciences , vol. 2, no. 1, pp. 183–202, 2009

    work page 2009

  23. [23]

    Large Scale Sparse Optimization for Object Detection in High Resolution Images,

    A. Boisbunon, R. Flamary, A. Rakotomamonjy, A. Giros, and J. Zerubia, “Large Scale Sparse Optimization for Object Detection in High Resolution Images,” in MLSP - 24th IEEE Workshop on Machine Learning for Signal Processing , Reims, France, Sep. 2014. [Online]. Available: https://hal.inria.fr/hal-01066235

    work page 2014

  24. [24]

    From safe screening rules to working sets for faster Lasso-type solvers

    M. Massias, A. Gramfort, and J. Salmon, “From safe screening rules to working sets for faster lasso-type solvers,” arXiv preprint arXiv:1703.07285, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  25. [25]

    Celer: a fast solver for the lasso with dual extrapolation,

    ——, “Celer: a fast solver for the lasso with dual extrapolation,” in International Conference in Machine Learning , 2018

    work page 2018

  26. [26]

    Convex optimiza- tion with sparsity-inducing norms,

    F. Bach, R. Jenatton, J. Mairal, G. Obozinski et al., “Convex optimiza- tion with sparsity-inducing norms,” Optimization for Machine Learning, vol. 5, pp. 19–53

    work page

  27. [27]

    Algorithm 447: efficient algorithms for graph manipulation,

    J. Hopcroft and R. Tarjan, “Algorithm 447: efficient algorithms for graph manipulation,” Communications of the ACM, vol. 16, no. 6, pp. 372–378, 1973

    work page 1973

  28. [28]

    Multiple tests based on a gaussian approximation of the unitary events method with delayed coincidence count,

    C. Tuleau-Malot, A. Rouis, F. Grammont, and P. Reynaud-Bouret, “Multiple tests based on a gaussian approximation of the unitary events method with delayed coincidence count,” Neural Computation, vol. 26, no. 7, 2014

    work page 2014

  29. [29]

    Spams: A sparse modeling software,

    J. Mairal, F. Bach, J. Ponce, G. Sapiro, R. Jenatton, and G. Obozinski, “Spams: A sparse modeling software,” URL http://spams-devel. gforge. inria. fr/downloads. html , 2014

    work page 2014

  30. [30]

    A quantitative description of membrane current and its application to conduction and excitation in nerve,

    A. L. Hodgkin and A. F. Huxley, “A quantitative description of membrane current and its application to conduction and excitation in nerve,” The Journal of Physiology , vol. 117, no. 4, pp. 500–544,

    work page

  31. [31]

    Available: https://physoc.onlinelibrary.wiley.com/doi/ abs/10.1113/jphysiol.1952.sp004764

    [Online]. Available: https://physoc.onlinelibrary.wiley.com/doi/ abs/10.1113/jphysiol.1952.sp004764

    work page doi:10.1113/jphysiol.1952.sp004764 1952

  32. [32]

    Origin of the (high frequency) extra-cellular sig- nal,

    “Origin of the (high frequency) extra-cellular sig- nal,” http://christophe-pouzat.github.io/LASCON2016/ OriginOfTheHighFrequencyExtraCellularSignal.html, 2016

    work page 2016

  33. [33]

    Neural correlates of goal-directed spatial navigation in the rat dorsal striatum

    I. Bethus, B. Poucet, and F. Sargolini, “Neural correlates of goal-directed spatial navigation in the rat dorsal striatum.” in Forum of European Neuroscience, Barcelona, Spain, 2012

    work page 2012

  34. [34]

    Online dictionary learning for sparse coding,

    J. Mairal, F. Bach, J. Ponce, and G. Sapiro, “Online dictionary learning for sparse coding,” in Proceedings of the 26th annual international conference on machine learning . ACM, 2009, pp. 689–696

    work page 2009