Ilya Kuzovkin
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Ilya Kuzovkin has a total of 37 co-authors. Here are the top co-authors by number of shared publications:
- Raul Vicente
University of Tartu
6 shared publications - Jaan Aru
University of Tartu
4 shared publications - Sylvain Rheims
Department of Functional Neurology and Epileptology Hospices Civils de Lyon and Lyon 1 University Lyon France
4 shared publications - Juan R. Vidal
UMRS 449, Université Catholique de Lyon/Ecole Pratique des Hautes Etudes, Lyon, France
3 shared publications Jean-Philippe Lachaux
INSERM U1028, CNRS UMR5292, Lyon Neuroscience Research Center, Lyon, France
2 shared publications- Jean-Philippe Lachaux
Centre de Recherche en Neurosciences de Lyon, INSERM U1028, CNRS UMR-5292, UCB Lyon 1
2 shared publications
Latest Publications
- Open Access
Anterior default mode brain state dynamics predict depressive symptom severity before and during TMS treatment
Transcranial magnetic stimulation (TMS) targeting the left dorsolateral prefrontal cortex is known to progressively reduce symptoms of depression. However, the neural mechanisms supporting this effect are poorly understood. To address this gap, we analysed longitudinal EEG recordings from 70 people undergoing TMS therapy and fitted an established dynamic network model of resting-state activity. Greater baseline symptom severity was associated with reduced occupancy of and fewer transitions into an anterior default mode brain state, alongside increased activity in a posterior default mode state. During treatment, decreases in anterior default mode state engagement following TMS predicted symptom improvement in the latter half of the intervention. Brain state activity exhibited structured, cyclical dynamics, with slower cycles linked to greater baseline severity. These findings suggest that symptoms of depression are characterised by gradual alterations in brain state dynamics, highlighting a central and dissociable role of default mode brain states in the persistence and remission of symptoms.
Open AccessAnterior default mode brain state dynamics predict depressive symptom severity before and during TMS treatment
Transcranial magnetic stimulation (TMS) targeting the left dorsolateral prefrontal cortex is known to progressively reduce symptoms of depression. However, the neural mechanisms supporting this effect are poorly understood. To address this gap, we analysed longitudinal EEG recordings from 70 people undergoing TMS therapy and fitted an established dynamic network model of resting-state activity. Greater baseline symptom severity was associated with reduced occupancy of and fewer transitions into an anterior default mode brain state, alongside increased activity in a posterior default mode state. During treatment, decreases in anterior default mode state engagement following TMS predicted symptom improvement in the latter half of the intervention. Brain state activity exhibited structured, cyclical dynamics, with slower cycles linked to greater baseline severity. These findings suggest that symptoms of depression are characterised by gradual alterations in brain state dynamics, highlighting a central and dissociable role of default mode brain states in the persistence and remission of symptoms.
- Open Access
- 12 May 2020
- Published by Springer Nature in Scientific Reports
- Vol. 10 (1) , 1-17
Identifying task-relevant spectral signatures of perceptual categorization in the human cortex
Human brain has developed mechanisms to efficiently decode sensory information according to perceptual categories of high prevalence in the environment, such as faces, symbols, objects. Neural activity produced within localized brain networks has been associated with the process that integrates both sensory bottom-up and cognitive top-down information processing. Yet, how specifically the different types and components of neural responses reflect the local networks’ selectivity for categorical information processing is still unknown. In this work we train Random Forest classification models to decode eight perceptual categories from broad spectrum of human intracranial signals (4–150 Hz, 100 subjects) obtained during a visual perception task. We then analyze which of the spectral features the algorithm deemed relevant to the perceptual decoding and gain the insights into which parts of the recorded activity are actually characteristic of the visual categorization process in the human brain. We show that network selectivity for a single or multiple categories in sensory and non-sensory cortices is related to specific patterns of power increases and decreases in both low (4–50 Hz) and high (50–150 Hz) frequency bands. By focusing on task-relevant neural activity and separating it into dissociated anatomical and spectrotemporal groups we uncover spectral signatures that characterize neural mechanisms of visual category perception in human brain that have not yet been reported in the literature.
- 12 May 2020
- Published by Springer Nature in Scientific Reports
- Vol. 10 (1) , 1-17
Open AccessIdentifying task-relevant spectral signatures of perceptual categorization in the human cortex
Human brain has developed mechanisms to efficiently decode sensory information according to perceptual categories of high prevalence in the environment, such as faces, symbols, objects. Neural activity produced within localized brain networks has been associated with the process that integrates both sensory bottom-up and cognitive top-down information processing. Yet, how specifically the different types and components of neural responses reflect the local networks’ selectivity for categorical information processing is still unknown. In this work we train Random Forest classification models to decode eight perceptual categories from broad spectrum of human intracranial signals (4–150 Hz, 100 subjects) obtained during a visual perception task. We then analyze which of the spectral features the algorithm deemed relevant to the perceptual decoding and gain the insights into which parts of the recorded activity are actually characteristic of the visual categorization process in the human brain. We show that network selectivity for a single or multiple categories in sensory and non-sensory cortices is related to specific patterns of power increases and decreases in both low (4–50 Hz) and high (50–150 Hz) frequency bands. By focusing on task-relevant neural activity and separating it into dissociated anatomical and spectrotemporal groups we uncover spectral signatures that characterize neural mechanisms of visual category perception in human brain that have not yet been reported in the literature.
- 1 February 2020
- Published by IOP Publishing in Journal of Neural Engineering
- Vol. 17 (1) , 016059
Mental state space visualization for interactive modeling of personalized BCI control strategies
Objective . Numerous studies in the area of BCI are focused on the search for a better experimental paradigm—a set of mental actions that a user can evoke consistently and a machine can discriminate reliably. Examples of such mental activities are motor imagery, mental computations, etc. We propose a technique that instead allows the user to try different mental actions in the search for the ones that will work best. Approach . The system is based on a modification of the self-organizing map (SOM) algorithm and enables interactive communication between the user and the learning system through a visualization of user’s mental state space. During the interaction with the system the user converges on the paradigm that is most efficient and intuitive for that particular user. Main results . Results of the two experiments, one allowing muscular activity, another permitting mental activity only, demonstrate soundness of the proposed method and offer preliminary validation of the performance improvement over the traditional closed-loop feedback approach. Significance . The proposed method allows a user to visually explore their mental state space in real time, opening new opportunities for scientific inquiry. The application of this method to the area of brain–computer interfaces enables more efficient search for the mental states that will allow a user to reliably control a BCI system.
- 1 February 2020
- Published by IOP Publishing in Journal of Neural Engineering
- Vol. 17 (1) , 016059
Mental state space visualization for interactive modeling of personalized BCI control strategies
Objective . Numerous studies in the area of BCI are focused on the search for a better experimental paradigm—a set of mental actions that a user can evoke consistently and a machine can discriminate reliably. Examples of such mental activities are motor imagery, mental computations, etc. We propose a technique that instead allows the user to try different mental actions in the search for the ones that will work best. Approach . The system is based on a modification of the self-organizing map (SOM) algorithm and enables interactive communication between the user and the learning system through a visualization of user’s mental state space. During the interaction with the system the user converges on the paradigm that is most efficient and intuitive for that particular user. Main results . Results of the two experiments, one allowing muscular activity, another permitting mental activity only, demonstrate soundness of the proposed method and offer preliminary validation of the performance improvement over the traditional closed-loop feedback approach. Significance . The proposed method allows a user to visually explore their mental state space in real time, opening new opportunities for scientific inquiry. The application of this method to the area of brain–computer interfaces enables more efficient search for the mental states that will allow a user to reliably control a BCI system.
- Open Access
- 6 December 2019
- Published by Cold Spring Harbor Laboratory in bioRxiv
- p. 867119
Mental state space visualization for interactive modeling of personalized BCI control strategies
Objective: Numerous studies in the area of BCI are focused on the search for a better experimental paradigm – a set of mental actions that a user can evoke consistently and a machine can discriminate reliably. Examples of such mental activities are motor imagery, mental computations, etc. We propose a technique that instead allows the user to try different mental actions in the search for the ones that will work best.Approach: The system is based on a modification of the self-organizing map (SOM) algorithm and enables interactive communication between the user and the learning system through a visualization of user’s mental state space. During the interaction with the system the user converges on the paradigm that is most efficient and intuitive for that particular user.Main results: Results of the two experiments, one allowing muscular activity, another permitting mental activity only, demonstrate soundness of the proposed method and offer preliminary validation of the performance improvement over the traditional closed-loop feedback approach.Significance: The proposed method allows a user to visually explore their mental state space in real time, opening new opportunities for scientific inquiry. The application of this method to the area of brain-computer interfaces enables more efficient search for the mental states that will allow a user to reliably control a BCI system.
- 6 December 2019
- Published by Cold Spring Harbor Laboratory in bioRxiv
- p. 867119
Open AccessMental state space visualization for interactive modeling of personalized BCI control strategies
Objective: Numerous studies in the area of BCI are focused on the search for a better experimental paradigm – a set of mental actions that a user can evoke consistently and a machine can discriminate reliably. Examples of such mental activities are motor imagery, mental computations, etc. We propose a technique that instead allows the user to try different mental actions in the search for the ones that will work best.Approach: The system is based on a modification of the self-organizing map (SOM) algorithm and enables interactive communication between the user and the learning system through a visualization of user’s mental state space. During the interaction with the system the user converges on the paradigm that is most efficient and intuitive for that particular user.Main results: Results of the two experiments, one allowing muscular activity, another permitting mental activity only, demonstrate soundness of the proposed method and offer preliminary validation of the performance improvement over the traditional closed-loop feedback approach.Significance: The proposed method allows a user to visually explore their mental state space in real time, opening new opportunities for scientific inquiry. The application of this method to the area of brain-computer interfaces enables more efficient search for the mental states that will allow a user to reliably control a BCI system.
- 12 December 2018
- Published by IOP Publishing in Journal of Neural Engineering
- Vol. 16 (1)
Direct information transfer rate optimisation for SSVEP-based BCI
In this work, a classification method for SSVEP-based BCI is proposed. The classification method uses features extracted by traditional SSVEP-based BCI methods and finds optimal discrimination thresholds for each feature to classify the targets. Optimising the thresholds is formalised as a maximisation task of a performance measure of BCIs called information transfer rate (ITR). However, instead of the standard method of calculating ITR, which makes certain assumptions about the data, a more general formula is derived to avoid incorrect ITR calculation when the standard assumptions are not met. This allows the optimal discrimination thresholds to be automatically calculated and thus eliminates the need for manual parameter selection or performing computationally expensive grid searches. The proposed method shows good performance in classifying targets of a BCI, outperforming previously reported results on the same dataset by a factor of 2 in terms of ITR. The highest achieved ITR on the used dataset was 62 bit/min. The proposed method also provides a way to reduce false classifications, which is important in real-world applications.
- 12 December 2018
- Published by IOP Publishing in Journal of Neural Engineering
- Vol. 16 (1)
Direct information transfer rate optimisation for SSVEP-based BCI
In this work, a classification method for SSVEP-based BCI is proposed. The classification method uses features extracted by traditional SSVEP-based BCI methods and finds optimal discrimination thresholds for each feature to classify the targets. Optimising the thresholds is formalised as a maximisation task of a performance measure of BCIs called information transfer rate (ITR). However, instead of the standard method of calculating ITR, which makes certain assumptions about the data, a more general formula is derived to avoid incorrect ITR calculation when the standard assumptions are not met. This allows the optimal discrimination thresholds to be automatically calculated and thus eliminates the need for manual parameter selection or performing computationally expensive grid searches. The proposed method shows good performance in classifying targets of a BCI, outperforming previously reported results on the same dataset by a factor of 2 in terms of ITR. The highest achieved ITR on the used dataset was 62 bit/min. The proposed method also provides a way to reduce false classifications, which is important in real-world applications.
- Open Access
Identifying task-relevant spectral signatures of perceptual categorization in the human cortex
Human brain has developed mechanisms to efficiently decode sensory information according to perceptual categories of high prevalence in the environment, such as faces, symbols, objects. Neural activity produced within localized brain networks has been associated with the process that integrates both sensory bottom-up and cognitive top-down information processing. Yet, how specifically the different types and components of neural responses reflect the local networks' selectivity for categorical information processing is still unknown. By mimicking the decoding of the sensory information with machine learning we can obtain accurate artificial decoding models. Having the artificial system functionally on par with the biological one we can analyze the mechanics of the artificial system to gain insights into the inner workings of its biological counterpart. In this work we train a Random Forest classification model to decode eight perceptual categories from visual stimuli given a broad spectrum of human intracranial signals 4-150 Hz obtained during a visual perception task, and analyze which of the spectral features the algorithm deemed relevant to the perceptual decoding. We show that network selectivity for a single or multiple categories in sensory and non-sensory cortices is related to specific patterns of power increases and decreases in both low 4-50 Hz and high 50-150 Hz frequency bands. We demonstrate that the locations and patterns of activity that are identified by the algorithm not only coincide with the known spectro-spatial signatures, but extend our knowledge by uncovering additional spectral signatures describing neural mechanisms of visual category perception in human brain.
Open AccessIdentifying task-relevant spectral signatures of perceptual categorization in the human cortex
Human brain has developed mechanisms to efficiently decode sensory information according to perceptual categories of high prevalence in the environment, such as faces, symbols, objects. Neural activity produced within localized brain networks has been associated with the process that integrates both sensory bottom-up and cognitive top-down information processing. Yet, how specifically the different types and components of neural responses reflect the local networks' selectivity for categorical information processing is still unknown. By mimicking the decoding of the sensory information with machine learning we can obtain accurate artificial decoding models. Having the artificial system functionally on par with the biological one we can analyze the mechanics of the artificial system to gain insights into the inner workings of its biological counterpart. In this work we train a Random Forest classification model to decode eight perceptual categories from visual stimuli given a broad spectrum of human intracranial signals 4-150 Hz obtained during a visual perception task, and analyze which of the spectral features the algorithm deemed relevant to the perceptual decoding. We show that network selectivity for a single or multiple categories in sensory and non-sensory cortices is related to specific patterns of power increases and decreases in both low 4-50 Hz and high 50-150 Hz frequency bands. We demonstrate that the locations and patterns of activity that are identified by the algorithm not only coincide with the known spectro-spatial signatures, but extend our knowledge by uncovering additional spectral signatures describing neural mechanisms of visual category perception in human brain.
- Open Access
- 8 August 2018
- Published by Springer Nature in Communications Biology
- Vol. 1 (1) , 1-12
Activations of deep convolutional neural networks are aligned with gamma band activity of human visual cortex
Recent advances in the field of artificial intelligence have revealed principles about neural processing, in particular about vision. Previous work demonstrated a direct correspondence between the hierarchy of the human visual areas and layers of deep convolutional neural networks (DCNN) trained on visual object recognition. We use DCNN to investigate which frequency bands correlate with feature transformations of increasing complexity along the ventral visual pathway. By capitalizing on intracranial depth recordings from 100 patients we assess the alignment between the DCNN and signals at different frequency bands. We find that gamma activity (30–70 Hz) matches the increasing complexity of visual feature representations in DCNN. These findings show that the activity of the DCNN captures the essential characteristics of biological object recognition not only in space and time, but also in the frequency domain. These results demonstrate the potential that artificial intelligence algorithms have in advancing our understanding of the brain.
- 8 August 2018
- Published by Springer Nature in Communications Biology
- Vol. 1 (1) , 1-12
Open AccessActivations of deep convolutional neural networks are aligned with gamma band activity of human visual cortex
Recent advances in the field of artificial intelligence have revealed principles about neural processing, in particular about vision. Previous work demonstrated a direct correspondence between the hierarchy of the human visual areas and layers of deep convolutional neural networks (DCNN) trained on visual object recognition. We use DCNN to investigate which frequency bands correlate with feature transformations of increasing complexity along the ventral visual pathway. By capitalizing on intracranial depth recordings from 100 patients we assess the alignment between the DCNN and signals at different frequency bands. We find that gamma activity (30–70 Hz) matches the increasing complexity of visual feature representations in DCNN. These findings show that the activity of the DCNN captures the essential characteristics of biological object recognition not only in space and time, but also in the frequency domain. These results demonstrate the potential that artificial intelligence algorithms have in advancing our understanding of the brain.
- Open Access
Combining Static and Dynamic Features for Multivariate Sequence Classification
Model precision in a classification task is highly dependent on the feature space that is used to train the model. Moreover, whether the features are sequential or static will dictate which classification method can be applied as most of the machine learning algorithms are designed to deal with either one or another type of data. In real-life scenarios, however, it is often the case that both static and dynamic features are present, or can be extracted from the data. In this work, we demonstrate how generative models such as Hidden Markov Models (HMM) and Long Short-Term Memory (LSTM) artificial neural networks can be used to extract temporal information from the dynamic data. We explore how the extracted information can be combined with the static features in order to improve the classification performance. We evaluate the existing techniques and suggest a hybrid approach, which outperforms other methods on several public datasets.
Open AccessCombining Static and Dynamic Features for Multivariate Sequence Classification
Model precision in a classification task is highly dependent on the feature space that is used to train the model. Moreover, whether the features are sequential or static will dictate which classification method can be applied as most of the machine learning algorithms are designed to deal with either one or another type of data. In real-life scenarios, however, it is often the case that both static and dynamic features are present, or can be extracted from the data. In this work, we demonstrate how generative models such as Hidden Markov Models (HMM) and Long Short-Term Memory (LSTM) artificial neural networks can be used to extract temporal information from the dynamic data. We explore how the extracted information can be combined with the static features in order to improve the classification performance. We evaluate the existing techniques and suggest a hybrid approach, which outperforms other methods on several public datasets.
- Open Access
Activations of Deep Convolutional Neural Network are Aligned with Gamma Band Activity of Human Visual Cortex
Previous work demonstrated a direct correspondence between the hierarchy of the human visual areas and layers of deep convolutional neural networks (DCNN) trained on visual object recognition. We used DCNNs to investigate which frequency bands correlate with feature transformations of increasing complexity along the ventral visual pathway. By capitalizing on intracranial depth recordings from 100 patients and 11293 electrodes we assessed the alignment between the DCNN and signals at different frequency bands in different time windows. We found that gamma activity, especially in the low gamma-band (30 – 70 Hz), matched the increasing complexity of visual feature representations in the DCNN. These findings show that the activity of the DCNN captures the essential characteristics of biological object recognition not only in space and time, but also in the frequency domain. These results also demonstrate the potential that modern artificial intelligence algorithms have in advancing our understanding of the brain. Significance Statement: Recent advances in the field of artificial intelligence have revealed principles about neural processing, in particular about vision. Previous works have demonstrated a direct correspondence between the hierarchy of human visual areas and layers of deep convolutional neural networks (DCNNs), suggesting that DCNN is a good model of visual object recognition in primate brain. Studying intracranial depth recordings allowed us to extend previous works by assessing when and at which frequency bands the activity of the visual system corresponds to the DCNN. Our key finding is that signals in gamma frequencies along the ventral visual pathway are aligned with the layers of DCNN. Gamma frequencies play a major role in transforming visual input to coherent object representations.
Open AccessActivations of Deep Convolutional Neural Network are Aligned with Gamma Band Activity of Human Visual Cortex
Previous work demonstrated a direct correspondence between the hierarchy of the human visual areas and layers of deep convolutional neural networks (DCNN) trained on visual object recognition. We used DCNNs to investigate which frequency bands correlate with feature transformations of increasing complexity along the ventral visual pathway. By capitalizing on intracranial depth recordings from 100 patients and 11293 electrodes we assessed the alignment between the DCNN and signals at different frequency bands in different time windows. We found that gamma activity, especially in the low gamma-band (30 – 70 Hz), matched the increasing complexity of visual feature representations in the DCNN. These findings show that the activity of the DCNN captures the essential characteristics of biological object recognition not only in space and time, but also in the frequency domain. These results also demonstrate the potential that modern artificial intelligence algorithms have in advancing our understanding of the brain. Significance Statement: Recent advances in the field of artificial intelligence have revealed principles about neural processing, in particular about vision. Previous works have demonstrated a direct correspondence between the hierarchy of human visual areas and layers of deep convolutional neural networks (DCNNs), suggesting that DCNN is a good model of visual object recognition in primate brain. Studying intracranial depth recordings allowed us to extend previous works by assessing when and at which frequency bands the activity of the visual system corresponds to the DCNN. Our key finding is that signals in gamma frequencies along the ventral visual pathway are aligned with the layers of DCNN. Gamma frequencies play a major role in transforming visual input to coherent object representations.
- Top Cited PapersOpen Access
- 5 April 2017
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 12 (4) , e0172395
Multiagent cooperation and competition with deep reinforcement learning
Evolution of cooperation and competition can appear when multiple adaptive agents share a biological, social, or technological niche. In the present work we study how cooperation and competition emerge between autonomous agents that learn by reinforcement while using only their raw visual input as the state representation. In particular, we extend the Deep Q-Learning framework to multiagent environments to investigate the interaction between two learning agents in the well-known video game Pong. By manipulating the classical rewarding scheme of Pong we show how competitive and collaborative behaviors emerge. We also describe the progression from competitive to collaborative behavior when the incentive to cooperate is increased. Finally we show how learning by playing against another adaptive agent, instead of against a hard-wired algorithm, results in more robust strategies. The present work shows that Deep Q-Networks can become a useful tool for studying decentralized learning of multiagent systems coping with high-dimensional environments.
- 5 April 2017
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 12 (4) , e0172395
Top Cited PapersOpen AccessMultiagent cooperation and competition with deep reinforcement learning
Evolution of cooperation and competition can appear when multiple adaptive agents share a biological, social, or technological niche. In the present work we study how cooperation and competition emerge between autonomous agents that learn by reinforcement while using only their raw visual input as the state representation. In particular, we extend the Deep Q-Learning framework to multiagent environments to investigate the interaction between two learning agents in the well-known video game Pong. By manipulating the classical rewarding scheme of Pong we show how competitive and collaborative behaviors emerge. We also describe the progression from competitive to collaborative behavior when the incentive to cooperate is increased. Finally we show how learning by playing against another adaptive agent, instead of against a hard-wired algorithm, results in more robust strategies. The present work shows that Deep Q-Networks can become a useful tool for studying decentralized learning of multiagent systems coping with high-dimensional environments.