Patrick Mineault
Research Subjects
Patrick Mineault has a total of 231 co-authors. Here are the top co-authors by number of shared publications:
- Jakob H. Macke
Machine Learning in Science, Excellence Cluster ‘Machine Learning’, Tübingen University, Tübingen, Germany
3 shared publications - Marius Pachitariu
HHMI Janelia Research Campus, Ashburn, USA
3 shared publications - Alexandre Hyafil
Centre de Recerca Matemàtica
2 shared publications - Andreas Savas Tolias
Department of Ophthalmology, Byers Eye Institute, Stanford University School of Medicine
2 shared publications - Arvind Kumar
Department of Computational Science and Technology, School of Electrical Engineering and Computer Science, KTH Royal Institute of Technology, 11428 Stockholm, Sweden
2 shared publications - Athena Akrami
Sainsbury Wellcome Centre, UCL
2 shared publications
Latest Publications
- 11 November 2026
- Published by Society for Neuroscience in The Journal of Neuroscience
- Vol. 46 (45)
NeuroAIs Next Steps: Balancing Computation and Embodiment
Deep neural networks trained at scale now predict neural activity across sensory cortices, raising a question that divides neuroscience: does the convergence between artificial and biological representations constitute understanding, or does understanding require models of biological processes grounded in the embodied, evolved character of natural intelligence? In this Dual Perspectives article, Bing Brunton and Patrick Mineault articulate two research programsProcess and Convergenceto engage this question about the future of NeuroAI research. The conversation is organized around four questions that probe what scaled modelbrain alignment reveals, what the brain is coupled to and evolved to do, what success beyond the lab looks like, and where the field should invest over the coming decade. Their exchanges reveal the distance between the two programs to be primarily about the level of analysis and priorities for what to build next. The conversation converges on shared ground: measure broadly, then distill to causal principles.
- 11 November 2026
- Published by Society for Neuroscience in The Journal of Neuroscience
- Vol. 46 (45)
NeuroAIs Next Steps: Balancing Computation and Embodiment
Deep neural networks trained at scale now predict neural activity across sensory cortices, raising a question that divides neuroscience: does the convergence between artificial and biological representations constitute understanding, or does understanding require models of biological processes grounded in the embodied, evolved character of natural intelligence? In this Dual Perspectives article, Bing Brunton and Patrick Mineault articulate two research programsProcess and Convergenceto engage this question about the future of NeuroAI research. The conversation is organized around four questions that probe what scaled modelbrain alignment reveals, what the brain is coupled to and evolved to do, what success beyond the lab looks like, and where the field should invest over the coming decade. Their exchanges reveal the distance between the two programs to be primarily about the level of analysis and priorities for what to build next. The conversation converges on shared ground: measure broadly, then distill to causal principles.
- 13 July 2023
- Published by Springer Nature in Nature Human Behaviour
- Vol. 7 (7) , 1021-1026
The future of academic publishing
- 13 July 2023
- Published by Springer Nature in Nature Human Behaviour
- Vol. 7 (7) , 1021-1026
The future of academic publishing
- Open Access
- 31 March 2022
- Published by The Open Journal in Journal of Open Source Education
- Vol. 5 (49) , 118
Neuromatch Academy: a 3-week, online summer school in computational neuroscience
- 31 March 2022
- Published by The Open Journal in Journal of Open Source Education
- Vol. 5 (49) , 118
Open AccessNeuromatch Academy: a 3-week, online summer school in computational neuroscience
- Open Access
Your head is there to move you around: Goal-driven models of the primate dorsal pathway
Neurons in the dorsal visual pathway of the mammalian brain are selective for motion stimuli, with the complexity of stimulus representations increasing along the hierarchy. This progression is similar to that of the ventral visual pathway, which is well characterized by artificial neural networks (ANNs) optimized for object recognition. In contrast, there are no image-computable models of the dorsal stream with comparable explanatory power. We hypothesized that the properties of dorsal stream neurons could be explained by a simple learning objective: the need for an organism to orient itself during self-motion. To test this hypothesis, we trained a 3D ResNet to predict an agent’s self-motion parameters from visual stimuli in a simulated environment. We found that the responses in this network accounted well for the selectivity of neurons in a large database of single-neuron recordings from the dorsal visual stream of non-human primates. In contrast, ANNs trained on an action recognition dataset through supervised or self-supervised learning could not explain responses in the dorsal stream, despite also being trained on naturalistic videos with moving objects. These results demonstrate that an ecologically relevant cost function can account for dorsal stream properties in the primate brain.
Open AccessYour head is there to move you around: Goal-driven models of the primate dorsal pathway
Neurons in the dorsal visual pathway of the mammalian brain are selective for motion stimuli, with the complexity of stimulus representations increasing along the hierarchy. This progression is similar to that of the ventral visual pathway, which is well characterized by artificial neural networks (ANNs) optimized for object recognition. In contrast, there are no image-computable models of the dorsal stream with comparable explanatory power. We hypothesized that the properties of dorsal stream neurons could be explained by a simple learning objective: the need for an organism to orient itself during self-motion. To test this hypothesis, we trained a 3D ResNet to predict an agent’s self-motion parameters from visual stimuli in a simulated environment. We found that the responses in this network accounted well for the selectivity of neurons in a large database of single-neuron recordings from the dorsal visual stream of non-human primates. In contrast, ANNs trained on an action recognition dataset through supervised or self-supervised learning could not explain responses in the dorsal stream, despite also being trained on naturalistic videos with moving objects. These results demonstrate that an ecologically relevant cost function can account for dorsal stream properties in the primate brain.
- Open Access
- 15 February 2021
- Published by Center for Open Science in OSF Preprints
Neuromatch Academy: a 3-week, online summer school in computational neuroscience
Neuromatch Academy (https://neuromatch.io/academy) was designed as an online summer school to cover the basics of computational neuroscience in three weeks. The materials cover dominant and emerging computational neuroscience tools, how they complement one another, and specifically focus on how they can help us to better understand how the brain functions. An original component of the materials is its focus on modeling choices, i.e. how do we choose the right approach, how do we build models, and how can we evaluate models to determine if they provide real (meaningful) insight. This meta-modeling component of the instructional materials asks what questions can be answered by different techniques, and how to apply them meaningfully to get insight about brain function.
- 15 February 2021
- Published by Center for Open Science in OSF Preprints
Open AccessNeuromatch Academy: a 3-week, online summer school in computational neuroscience
Neuromatch Academy (https://neuromatch.io/academy) was designed as an online summer school to cover the basics of computational neuroscience in three weeks. The materials cover dominant and emerging computational neuroscience tools, how they complement one another, and specifically focus on how they can help us to better understand how the brain functions. An original component of the materials is its focus on modeling choices, i.e. how do we choose the right approach, how do we build models, and how can we evaluate models to determine if they provide real (meaningful) insight. This meta-modeling component of the instructional materials asks what questions can be answered by different techniques, and how to apply them meaningfully to get insight about brain function.
- Open Access
- 21 May 2018
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 14 (5) , e1006157
Community-based benchmarking improves spike rate inference from two-photon calcium imaging data
In recent years, two-photon calcium imaging has become a standard tool to probe the function of neural circuits and to study computations in neuronal populations. However, the acquired signal is only an indirect measurement of neural activity due to the comparatively slow dynamics of fluorescent calcium indicators. Different algorithms for estimating spike rates from noisy calcium measurements have been proposed in the past, but it is an open question how far performance can be improved. Here, we report the results of the spikefinder challenge, launched to catalyze the development of new spike rate inference algorithms through crowd-sourcing. We present ten of the submitted algorithms which show improved performance compared to previously evaluated methods. Interestingly, the top-performing algorithms are based on a wide range of principles from deep neural networks to generative models, yet provide highly correlated estimates of the neural activity. The competition shows that benchmark challenges can drive algorithmic developments in neuroscience. Two-photon calcium imaging is one of the major tools to study the activity of large populations of neurons in the brain. In this technique, a fluorescent calcium indicator changes its brightness when a neuron fires an action potential due to an associated increase in intracellular calcium. However, while a number of algorithms have been proposed for estimating spike rates from the measured signal, the problem is far from solved. We organized a public competition using a data set for which ground truth data was available. Participants were given a training set to develop new algorithms, and the performance of the algorithms was evaluated on a hidden test set. Here we report on the results of this competition and discuss the progress made towards better algorithms to infer spiking activity from imaging data.
- 21 May 2018
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 14 (5) , e1006157
Open AccessCommunity-based benchmarking improves spike rate inference from two-photon calcium imaging data
In recent years, two-photon calcium imaging has become a standard tool to probe the function of neural circuits and to study computations in neuronal populations. However, the acquired signal is only an indirect measurement of neural activity due to the comparatively slow dynamics of fluorescent calcium indicators. Different algorithms for estimating spike rates from noisy calcium measurements have been proposed in the past, but it is an open question how far performance can be improved. Here, we report the results of the spikefinder challenge, launched to catalyze the development of new spike rate inference algorithms through crowd-sourcing. We present ten of the submitted algorithms which show improved performance compared to previously evaluated methods. Interestingly, the top-performing algorithms are based on a wide range of principles from deep neural networks to generative models, yet provide highly correlated estimates of the neural activity. The competition shows that benchmark challenges can drive algorithmic developments in neuroscience. Two-photon calcium imaging is one of the major tools to study the activity of large populations of neurons in the brain. In this technique, a fluorescent calcium indicator changes its brightness when a neuron fires an action potential due to an associated increase in intracellular calcium. However, while a number of algorithms have been proposed for estimating spike rates from the measured signal, the problem is far from solved. We organized a public competition using a data set for which ground truth data was available. Participants were given a training set to develop new algorithms, and the performance of the algorithms was evaluated on a hidden test set. Here we report on the results of this competition and discuss the progress made towards better algorithms to infer spiking activity from imaging data.
- Open Access
- 18 August 2017
- Published by openRxiv
- p. 177956
Community-based benchmarking improves spike rate inference from two-photon calcium imaging data
In recent years, two-photon calcium imaging has become a standard tool to probe the function of neural circuits and to study computations in neuronal populations. However, the acquired signal is only an indirect measurement of neural activity due to the comparatively slow dynamics of fluorescent calcium indicators. Different algorithms for estimating spike trains from noisy calcium measurements have been proposed in the past, but it is an open question how far performance can be improved. Here, we report the results of the spikefinder challenge, launched to catalyze the development of new spike inference algorithms through crowd-sourcing. We present ten of the submitted algorithms which show improved performance compared to previously evaluated methods. Interestingly, the top-performing algorithms are based on a wide range of principles from deep neural networks to generative models, yet provide highly correlated estimates of the neural activity. The competition shows that benchmark challenges can drive algorithmic developments in neuroscience.
- 18 August 2017
- Published by openRxiv
- p. 177956
Open AccessCommunity-based benchmarking improves spike rate inference from two-photon calcium imaging data
In recent years, two-photon calcium imaging has become a standard tool to probe the function of neural circuits and to study computations in neuronal populations. However, the acquired signal is only an indirect measurement of neural activity due to the comparatively slow dynamics of fluorescent calcium indicators. Different algorithms for estimating spike trains from noisy calcium measurements have been proposed in the past, but it is an open question how far performance can be improved. Here, we report the results of the spikefinder challenge, launched to catalyze the development of new spike inference algorithms through crowd-sourcing. We present ten of the submitted algorithms which show improved performance compared to previously evaluated methods. Interestingly, the top-performing algorithms are based on a wide range of principles from deep neural networks to generative models, yet provide highly correlated estimates of the neural activity. The competition shows that benchmark challenges can drive algorithmic developments in neuroscience.
- Open Access
- 31 August 2016
- Published by Society for Neuroscience in The Journal of Neuroscience
- Vol. 36 (35) , 9227-9239
Mechanisms of Saccadic Suppression in Primate Cortical Area V4
Psychophysical studies have shown that subjects are often unaware of visual stimuli presented around the time of an eye movement. This saccadic suppression is thought to be a mechanism for maintaining perceptual stability. The brain might accomplish saccadic suppression by reducing the gain of visual responses to specific stimuli or by simply suppressing firing uniformly for all stimuli. Moreover, the suppression might be identical across the visual field or concentrated at specific points. To evaluate these possibilities, we recorded from individual neurons in cortical area V4 of nonhuman primates trained to execute saccadic eye movements. We found that both modes of suppression were evident in the visual responses of these neurons and that the two modes showed different spatial and temporal profiles: while gain changes started earlier and were more widely distributed across visual space, nonspecific suppression was found more often in the peripheral visual field, after the completion of the saccade. Peripheral suppression was also associated with increased noise correlations and stronger local field potential oscillations in the α frequency band. This pattern of results suggests that saccadic suppression shares some of the circuitry responsible for allocating voluntary attention. SIGNIFICANCE STATEMENT We explore our surroundings by looking at things, but each eye movement that we make causes an abrupt shift of the visual input. Why doesn't the world look like a film recorded on a shaky camera? The answer in part is a brain mechanism called saccadic suppression, which reduces the responses of visual neurons around the time of each eye movement. Here we reveal several new properties of the underlying mechanisms. First, the suppression operates differently in the central and peripheral visual fields. Second, it appears to be controlled by oscillations in the local field potentials at frequencies traditionally associated with attention. These results suggest that saccadic suppression shares the brain circuits responsible for actively ignoring irrelevant stimuli.
- 31 August 2016
- Published by Society for Neuroscience in The Journal of Neuroscience
- Vol. 36 (35) , 9227-9239
Open AccessMechanisms of Saccadic Suppression in Primate Cortical Area V4
Psychophysical studies have shown that subjects are often unaware of visual stimuli presented around the time of an eye movement. This saccadic suppression is thought to be a mechanism for maintaining perceptual stability. The brain might accomplish saccadic suppression by reducing the gain of visual responses to specific stimuli or by simply suppressing firing uniformly for all stimuli. Moreover, the suppression might be identical across the visual field or concentrated at specific points. To evaluate these possibilities, we recorded from individual neurons in cortical area V4 of nonhuman primates trained to execute saccadic eye movements. We found that both modes of suppression were evident in the visual responses of these neurons and that the two modes showed different spatial and temporal profiles: while gain changes started earlier and were more widely distributed across visual space, nonspecific suppression was found more often in the peripheral visual field, after the completion of the saccade. Peripheral suppression was also associated with increased noise correlations and stronger local field potential oscillations in the α frequency band. This pattern of results suggests that saccadic suppression shares some of the circuitry responsible for allocating voluntary attention. SIGNIFICANCE STATEMENT We explore our surroundings by looking at things, but each eye movement that we make causes an abrupt shift of the visual input. Why doesn't the world look like a film recorded on a shaky camera? The answer in part is a brain mechanism called saccadic suppression, which reduces the responses of visual neurons around the time of each eye movement. Here we reveal several new properties of the underlying mechanisms. First, the suppression operates differently in the central and peripheral visual fields. Second, it appears to be controlled by oscillations in the local field potentials at frequencies traditionally associated with attention. These results suggest that saccadic suppression shares the brain circuits responsible for actively ignoring irrelevant stimuli.
- Open Access
- 1 January 2013
- Published by Frontiers Media SA in Frontiers in Computational Neuroscience
- Vol. 7, 41936
Local field potentials reflect multiple spatial scales in V4
Local field potentials (LFP) reflect the properties of neuronal circuits or columns recorded in a volume around a microelectrode (Buzsáki et al., 2012). The extent of this integration volume has been a subject of some debate, with estimates ranging from a few hundred microns (Katzner et al., 2009; Xing et al., 2009) to several millimeters (Kreiman et al., 2006). We estimated receptive fields (RFs) of multi-unit activity (MUA) and LFPs at an intermediate level of visual processing, in area V4 of two macaques. The spatial structure of LFP receptive fields varied greatly as a function of time lag following stimulus onset, with the retinotopy of LFPs matching that of MUAs at a restricted set of time lags. A model-based analysis of the LFPs allowed us to recover two distinct stimulus-triggered components: an MUA-like retinotopic component that originated in a small volume around the microelectrodes (~350 µm), and a second component that was shared across the entire V4 region; this second component had tuning properties unrelated to those of the MUAs. Our results suggest that the LFP reflects neural activity across multiple spatial scales, which both complicates its interpretation and offers new opportunities for investigating the large-scale structure of network processing.
- 1 January 2013
- Published by Frontiers Media SA in Frontiers in Computational Neuroscience
- Vol. 7, 41936
Open AccessLocal field potentials reflect multiple spatial scales in V4
Local field potentials (LFP) reflect the properties of neuronal circuits or columns recorded in a volume around a microelectrode (Buzsáki et al., 2012). The extent of this integration volume has been a subject of some debate, with estimates ranging from a few hundred microns (Katzner et al., 2009; Xing et al., 2009) to several millimeters (Kreiman et al., 2006). We estimated receptive fields (RFs) of multi-unit activity (MUA) and LFPs at an intermediate level of visual processing, in area V4 of two macaques. The spatial structure of LFP receptive fields varied greatly as a function of time lag following stimulus onset, with the retinotopy of LFPs matching that of MUAs at a restricted set of time lags. A model-based analysis of the LFPs allowed us to recover two distinct stimulus-triggered components: an MUA-like retinotopic component that originated in a small volume around the microelectrodes (~350 µm), and a second component that was shared across the entire V4 region; this second component had tuning properties unrelated to those of the MUAs. Our results suggest that the LFP reflects neural activity across multiple spatial scales, which both complicates its interpretation and offers new opportunities for investigating the large-scale structure of network processing.