Author name cluster

Philippe Ciuciu

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

8 papers

2 author rows

ICLR Conference 2022 Conference Paper

SHINE: SHaring the INverse Estimate from the forward pass for bi-level optimization and implicit models

Zaccharie Ramzi
Florian Mannel
Shaojie Bai
Jean-Luc Starck
Philippe Ciuciu
Thomas Moreau 0001

In recent years, implicit deep learning has emerged as a method to increase the depth of deep neural networks. While their training is memory-efficient, they are still significantly slower to train than their explicit counterparts. In Deep Equilibrium Models~(DEQs), the training is performed as a bi-level problem, and its computational complexity is partially driven by the iterative inversion of a huge Jacobian matrix. In this paper, we propose a novel strategy to tackle this computational bottleneck from which many bi-level problems suffer. The main idea is to use the quasi-Newton matrices from the forward pass to efficiently approximate the inverse Jacobian matrix in the direction needed for the gradient computation. We provide a theorem that motivates using our method with the original forward algorithms. In addition, by modifying these forward algorithms, we further provide theoretical guarantees that our method asymptotically estimates the true implicit gradient. We empirically study this approach in many settings, ranging from hyperparameter optimization to large Multiscale DEQs applied to CIFAR and ImageNet. We show that it reduces the computational cost of the backward pass by up to two orders of magnitude. All this is achieved while retaining the excellent performance of the original models in hyperparameter optimization and on CIFAR, and giving encouraging and competitive results on ImageNet.

Details

YNIMG Journal 2021 Journal Article

Multivariate semi-blind deconvolution of fMRI time series

Hamza Cherkaoui
Thomas Moreau
Abderrahim Halimi
Claire Leroy
Philippe Ciuciu

Whole brain estimation of the haemodynamic response function (HRF) in functional magnetic resonance imaging (fMRI) is critical to get insight on the global status of the neurovascular coupling of an individual in healthy or pathological condition. Most of existing approaches in the literature works on task-fMRI data and relies on the experimental paradigm as a surrogate of neural activity, hence remaining inoperative on resting-stage fMRI (rs-fMRI) data. To cope with this issue, recent works have performed either a two-step analysis to detect large neural events and then characterize the HRF shape or a joint estimation of both the neural and haemodynamic components in an univariate fashion. In this work, we express the neural activity signals as a combination of piece-wise constant temporal atoms associated with sparse spatial maps and introduce an haemodynamic parcellation of the brain featuring a temporally dilated version of a given HRF model in each parcel with unknown dilation parameters. We formulate the joint estimation of the HRF shapes and spatio-temporal neural representations as a multivariate semi-blind deconvolution problem in a paradigm-free setting and introduce constraints inspired from the dictionary learning literature to ease its identifiability. A fast alternating minimization algorithm, along with its efficient implementation, is proposed and validated on both synthetic and real rs-fMRI data at the subject level. To demonstrate its significance at the population level, we apply this new framework to the UK Biobank data set, first for the discrimination of haemodynamic territories between balanced groups (n=24 individuals in each) patients with an history of stroke and healthy controls and second, for the analysis of normal aging on the neurovascular coupling. Overall, we statistically demonstrate that a pathology like stroke or a condition like normal brain aging induce longer haemodynamic delays in certain brain areas (e.g. Willis polygon, occipital, temporal and frontal cortices) and that this haemodynamic feature may be predictive with an accuracy of 74 % of the individual's age in a supervised classification task performed on n=459 subjects.

Details DOI

YNIMG Journal 2020 Journal Article

Emergence of β and γ networks following multisensory training

Daria La Rocca
Philippe Ciuciu
Denis-Alexander Engemann
Virginie van Wassenhove

Our perceptual reality relies on inferences about the causal structure of the world given by multiple sensory inputs. In ecological settings, multisensory events that cohere in time and space benefit inferential processes: hearing and seeing a speaker enhances speech comprehension, and the acoustic changes of flapping wings naturally pace the motion of a flock of birds. Here, we asked how a few minutes of (multi)sensory training could shape cortical interactions in a subsequent unisensory perceptual task. For this, we investigated oscillatory activity and functional connectivity as a function of individuals’ sensory history during training. Human participants performed a visual motion coherence discrimination task while being recorded with magnetoencephalography. Three groups of participants performed the same task with visual stimuli only, while listening to acoustic textures temporally comodulated with the strength of visual motion coherence, or with auditory noise uncorrelated with visual motion. The functional connectivity patterns before and after training were contrasted to resting-state networks to assess the variability of common task-relevant networks, and the emergence of new functional interactions as a function of sensory history. One major finding is the emergence of a large-scale synchronization in the high γ (gamma: 60 − 120 H z ) and β (beta: 15 − 30 H z ) bands for individuals who underwent comodulated multisensory training. The post-training network involved prefrontal, parietal, and visual cortices. Our results suggest that the integration of evidence and decision-making strategies become more efficient following congruent multisensory training through plasticity in network routing and oscillatory regimes.

Details DOI

YNIMG Journal 2015 Journal Article

Data-driven HRF estimation for encoding and decoding models

Fabian Pedregosa
Michael Eickenberg
Philippe Ciuciu
Bertrand Thirion
Alexandre Gramfort

Despite the common usage of a canonical, data-independent, hemodynamic response function (HRF), it is known that the shape of the HRF varies across brain regions and subjects. This suggests that a data-driven estimation of this function could lead to more statistical power when modeling BOLD fMRI data. However, unconstrained estimation of the HRF can yield highly unstable results when the number of free parameters is large. We develop a method for the joint estimation of activation and HRF by means of a rank constraint, forcing the estimated HRF to be equal across events or experimental conditions, yet permitting it to differ across voxels. Model estimation leads to an optimization problem that we propose to solve with an efficient quasi-Newton method, exploiting fast gradient computations. This model, called GLM with Rank-1 constraint (R1-GLM), can be extended to the setting of GLM with separate designs which has been shown to improve decoding accuracy in brain activity decoding experiments. We compare 10 different HRF modeling methods in terms of encoding and decoding scores on two different datasets. Our results show that the R1-GLM model outperforms competing methods in both encoding and decoding settings, positioning it as an attractive method both from the points of view of accuracy and computational efficiency.

Details DOI

YNIMG Journal 2014 Journal Article

Interplay between functional connectivity and scale-free dynamics in intrinsic fMRI networks

Philippe Ciuciu
Patrice Abry
Biyu J. He

Studies employing functional connectivity-type analyses have established that spontaneous fluctuations in functional magnetic resonance imaging (fMRI) signals are organized within large-scale brain networks. Meanwhile, fMRI signals have been shown to exhibit 1/f-type power spectra — a hallmark of scale-free dynamics. We studied the interplay between functional connectivity and scale-free dynamics in fMRI signals, utilizing the fractal connectivity framework — a multivariate extension of the univariate fractional Gaussian noise model, which relies on a wavelet formulation for robust parameter estimation. We applied this framework to fMRI data acquired from healthy young adults at rest and while performing a visual detection task. First, we found that scale-invariance existed beyond univariate dynamics, being present also in bivariate cross-temporal dynamics. Second, we observed that frequencies within the scale-free range do not contribute evenly to inter-regional connectivity, with a systematically stronger contribution of the lowest frequencies, both at rest and during task. Third, in addition to a decrease of the Hurst exponent and inter-regional correlations, task performance modified cross-temporal dynamics, inducing a larger contribution of the highest frequencies within the scale-free range to global correlation. Lastly, we found that across individuals, a weaker task modulation of the frequency contribution to inter-regional connectivity was associated with better task performance manifesting as shorter and less variable reaction times. These findings bring together two related fields that have hitherto been studied separately — resting-state networks and scale-free dynamics, and show that scale-free dynamics of human brain activity manifest in cross-regional interactions as well.

Details DOI

YNIMG Journal 2014 Journal Article

Supramodal processing optimizes visual perceptual learning and plasticity

Nicolas Zilber
Philippe Ciuciu
Alexandre Gramfort
Leila Azizi
Virginie van Wassenhove

Multisensory interactions are ubiquitous in cortex and it has been suggested that sensory cortices may be supramodal i. e. capable of functional selectivity irrespective of the sensory modality of inputs (Pascual-Leone and Hamilton, 2001; Renier et al. , 2013; Ricciardi and Pietrini, 2011; Voss and Zatorre, 2012). Here, we asked whether learning to discriminate visual coherence could benefit from supramodal processing. To this end, three groups of participants were briefly trained to discriminate which of a red or green intermixed population of random-dot-kinematograms (RDKs) was most coherent in a visual display while being recorded with magnetoencephalography (MEG). During training, participants heard no sound (V), congruent acoustic textures (AV) or auditory noise (AVn); importantly, congruent acoustic textures shared the temporal statistics – i. e. coherence – of visual RDKs. After training, the AV group significantly outperformed participants trained in V and AVn although they were not aware of their progress. In pre- and post-training blocks, all participants were tested without sound and with the same set of RDKs. When contrasting MEG data collected in these experimental blocks, selective differences were observed in the dynamic pattern and the cortical loci responsive to visual RDKs. First and common to all three groups, vlPFC showed selectivity to the learned coherence levels whereas selectivity in visual motion area hMT+ was only seen for the AV group. Second and solely for the AV group, activity in multisensory cortices (mSTS, pSTS) correlated with post-training performances; additionally, the latencies of these effects suggested feedback from vlPFC to hMT+ possibly mediated by temporal cortices in AV and AVn groups. Altogether, we interpret our results in the context of the Reverse Hierarchy Theory of learning (Ahissar and Hochstein, 2004) in which supramodal processing optimizes visual perceptual learning by capitalizing on sensory-invariant representations — here, global coherence levels across sensory modalities.

Details DOI

YNIMG Journal 2013 Journal Article

Group-level impacts of within- and between-subject hemodynamic variability in fMRI

Solveig Badillo
Thomas Vincent
Philippe Ciuciu

Inter-subject fMRI analyses have specific issues regarding the reliability of the results concerning both the detection of brain activation patterns and the estimation of the underlying dynamics. Among these issues lies the variability of the hemodynamic response function (HRF), that is usually accounted for using functional basis sets in the general linear model context. Here, we use the joint detection-estimation approach (JDE) (Makni et al. , 2008; Vincent et al. , 2010) which combines regional nonparametric HRF inference with spatially adaptive regularization of activation clusters to avoid global smoothing of fMRI images. We show that the JDE-based inference brings a significant improvement in statistical sensitivity for detecting evoked activity in parietal regions. In contrast, the canonical HRF associated with spatially adaptive regularization is more sensitive in other regions, such as motor cortex. This different regional behavior is shown to reflect a larger discrepancy of HRF with the canonical model. By varying parallel imaging acceleration factor, SNR-specific region-based hemodynamic parameters (activation delay and duration) were extracted from the JDE inference. Complementary analyses highlighted their significant departure from the canonical parameters and the strongest between-subject variability that occurs in the parietal region, irrespective of the SNR value. Finally, statistical evidence that the fluctuation of the HRF shape is responsible for the significant change in activation detection performance is demonstrated using paired t-tests between hemodynamic parameters inferred by GLM and JDE.

Details DOI

YNIMG Journal 2008 Journal Article

A fully Bayesian approach to the parcel-based detection-estimation of brain activity in fMRI

Salima Makni
Jérôme Idier
Thomas Vincent
Bertrand Thirion
Ghislaine Dehaene-Lambertz
Philippe Ciuciu

Within-subject analysis in fMRI essentially addresses two problems, i. e. , the detection of activated brain regions in response to an experimental task and the estimation of the underlying dynamics, also known as the characterisation of Hemodynamic response function (HRF). So far, both issues have been treated sequentially while it is known that the HRF model has a dramatic impact on the localisation of activations and that the HRF shape may vary from one region to another. In this paper, we conciliate both issues in a region-based joint detection-estimation framework that we develop in the Bayesian formalism. Instead of considering function basis to account for spatial variability, spatially adaptive General Linear Models are built upon region-based non-parametric estimation of brain dynamics. Regions are first identified as functionally homogeneous parcels in the mask of the grey matter using a specific procedure [Thirion, B. , Flandin, G. , Pinel, P. , Roche, A. , Ciuciu, P. , Poline, J. -B. , August 2006. Dealing with the shortcomings of spatial normalization: Multi-subject parcellation of fMRI datasets. Hum. Brain Mapp. 27 (8), 678–693. ]. Then, in each parcel, prior information is embedded to constrain this estimation. Detection is achieved by modelling activating, deactivating and non-activating voxels through mixture models within each parcel. From the posterior distribution, we infer upon the model parameters using Markov Chain Monte Carlo (MCMC) techniques. Bayesian model comparison allows us to emphasize on artificial datasets first that inhomogeneous gamma-Gaussian mixture models outperform Gaussian mixtures in terms of sensitivity/specificity trade-off and second that it is worthwhile modelling serial correlation through an AR(1) noise process at low signal-to-noise (SNR) ratio. Our approach is then validated on an fMRI experiment that studies habituation to auditory sentence repetition. This phenomenon is clearly recovered as well as the hierarchical temporal organisation of the superior temporal sulcus, which is directly derived from the parcel-based HRF estimates.

Details DOI