Refine
Year of publication
Language
- English (93)
Has Fulltext
- yes (93)
Is part of the Bibliography
- no (93)
Keywords
- dendrite (3)
- Visual cortex (2)
- morphology (2)
- natural scenes (2)
- neuronal populations (2)
- primary visual cortex (2)
- stimulus encoding (2)
- visual attention (2)
- Alpha rhythm (1)
- CNNs (1)
Institute
- Ernst Strüngmann Institut (93) (remove)
Residual connections have been proposed as an architecture-based inductive bias to mitigate the problem of exploding and vanishing gradients and increased task performance in both feed-forward and recurrent networks (RNNs) when trained with the backpropagation algorithm. Yet, little is known about how residual connections in RNNs influence their dynamics and fading memory properties. Here, we introduce weakly coupled residual recurrent networks (WCRNNs) in which residual connections result in well-defined Lyapunov exponents and allow for studying properties of fading memory. We investigate how the residual connections of WCRNNs influence their performance, network dynamics, and memory properties on a set of benchmark tasks. We show that several distinct forms of residual connections yield effective inductive biases that result in increased network expressivity. In particular, those are residual connections that (i) result in network dynamics at the proximity of the edge of chaos, (ii) allow networks to capitalize on characteristic spectral properties of the data, and (iii) result in heterogeneous memory properties. In addition, we demonstrate how our results can be extended to non-linear residuals and introduce a weakly coupled residual initialization scheme that can be used for Elman RNNs.
An important question concerning inter-areal communication in the cortex is whether these interactions are synergistic, i.e. convey information beyond what can be performed by isolated signals. In other words, any two signals can either share common information (redundancy) or they can encode complementary information that is only available when both signals are considered together (synergy). Here, we dissociated cortical interactions sharing common information from those encoding complementary information during prediction error processing. To this end, we computed co-information, an information-theoretical measure that distinguishes redundant from synergistic information among brain signals. We analyzed auditory and frontal electrocorticography (ECoG) signals in five common awake marmosets performing two distinct auditory oddball tasks, and investigated to what extent event-related potentials (ERP) and broadband (BB) dynamics exhibit redundancy and synergy for auditory prediction error signals. We observed multiple patterns of redundancy and synergy across the entire cortical hierarchy with distinct dynamics. The information conveyed by ERPs and BB signals was highly synergistic even at lower stages of the hierarchy in the auditory cortex, as well as between lower and higher areas in the frontal cortex. These results indicate that the distributed representations of prediction error signals across the cortical hierarchy can be highly synergistic.
We explore the potential of optically-pumped magnetometers (OPMs) to infer the laminar origins of neural activity non-invasively. OPM sensors can be positioned closer to the scalp than conventional cryogenic MEG sensors, opening an avenue to higher spatial resolution when combined with high-precision forward modelling. By simulating the forward model projection of single dipole sources onto OPM sensor arrays with varying sensor densities and measurement axes, and employing sparse source reconstruction approaches, we find that laminar inference with OPM arrays is possible at relatively low sensor counts at moderate to high signal-to-noise ratios (SNR). We observe improvements in laminar inference with increasing spatial sampling densities and number of measurement axes. Surprisingly, moving sensors closer to the scalp is less advantageous than anticipated - and even detrimental at high SNRs. Biases towards both the superficial and deep surfaces at very low SNRs and a notable bias towards the deep surface when combining empirical Bayesian beamformer (EBB) source reconstruction with a whole-brain analysis pose further challenges. Adequate SNR through appropriate trial numbers and shielding, as well as precise co-registration, is crucial for reliable laminar inference with OPMs.
An important question concerning inter-areal communication in the cortex is whether these interactions are synergistic, i.e. brain signals can either share common information (redundancy) or they can encode complementary information that is only available when both signals are considered together (synergy). Here, we dissociated cortical interactions sharing common information from those encoding complementary information during prediction error processing. To this end, we computed co-information, an information-theoretical measure that distinguishes redundant from synergistic information among brain signals. We analyzed auditory and frontal electrocorticography (ECoG) signals in five common awake marmosets performing two distinct auditory oddball tasks and investigated to what extent event-related potentials (ERP) and broadband (BB) dynamics encoded redundant and synergistic information during auditory prediction error processing. In both tasks, we observed multiple patterns of synergy across the entire cortical hierarchy with distinct dynamics. The information conveyed by ERPs and BB signals was highly synergistic even at lower stages of the hierarchy in the auditory cortex, as well as between auditory and frontal regions. Using a brain-constrained neural network, we simulated the spatio-temporal patterns of synergy and redundancy observed in the experimental results and further demonstrated that the emergence of synergy between auditory and frontal regions requires the presence of strong, long-distance, feedback and feedforward connections. These results indicate that the distributed representations of prediction error signals across the cortical hierarchy can be highly synergistic.
Convolutional neural networks (CNNs) are one of the most successful computer vision systems to solve object recognition. Furthermore, CNNs have major applications in understanding the nature of visual representations in the human brain. Yet it remains poorly understood how CNNs actually make their decisions, what the nature of their internal representations is, and how their recognition strategies differ from humans. Specifically, there is a major debate about the question of whether CNNs primarily rely on surface regularities of objects, or whether they are capable of exploiting the spatial arrangement of features, similar to humans. Here, we develop a novel feature-scrambling approach to explicitly test whether CNNs use the spatial arrangement of features (i.e. object parts) to classify objects. We combine this approach with a systematic manipulation of effective receptive field sizes of CNNs as well as minimal recognizable configurations (MIRCs) analysis. In contrast to much previous literature, we provide evidence that CNNs are in fact capable of using relatively long-range spatial relationships for object classification. Moreover, the extent to which CNNs use spatial relationships depends heavily on the dataset, e.g. texture vs. sketch. In fact, CNNs even use different strategies for different classes within heterogeneous datasets (ImageNet), suggesting CNNs have a continuous spectrum of classification strategies. Finally, we show that CNNs learn the spatial arrangement of features only up to an intermediate level of granularity, which suggests that intermediate rather than global shape features provide the optimal trade-off between sensitivity and specificity in object classification. These results provide novel insights into the nature of CNN representations and the extent to which they rely on the spatial arrangement of features for object classification.
Anticipating future events is a key computational task for neuronal networks. Experimental evidence suggests that reliable temporal sequences in neural activity play a functional role in the association and anticipation of events in time. However, how neurons can differentiate and anticipate multiple spike sequences remains largely unknown. We implement a learning rule based on predictive processing, where neurons exclusively fire for the initial, unpredictable inputs in a spiking sequence, leading to an efficient representation with reduced post-synaptic firing. Combining this mechanism with inhibitory feedback leads to sparse firing in the network, enabling neurons to selectively anticipate different sequences in the input. We demonstrate that intermediate levels of inhibition are optimal to decorrelate neuronal activity and to enable the prediction of future inputs. Notably, each sequence is independently encoded in the sparse, anticipatory firing of the network. Overall, our results demonstrate that the interplay of self-supervised predictive learning rules and inhibitory feedback enables fast and efficient classification of different input sequences.
Aberrant neurophysiological signaling associated with speech impairments in Parkinson’s disease
(2023)
Difficulty producing intelligible speech is a debilitating symptom of Parkinson’s disease (PD). Yet, both the robust evaluation of speech impairments and the identification of the affected brain systems are challenging. Using task-free magnetoencephalography, we examine the spectral and spatial definitions of the functional neuropathology underlying reduced speech quality in patients with PD using a new approach to characterize speech impairments and a novel brain-imaging marker. We found that the interactive scoring of speech impairments in PD (N = 59) is reliable across non-expert raters, and better related to the hallmark motor and cognitive impairments of PD than automatically-extracted acoustical features. By relating these speech impairment ratings to neurophysiological deviations from healthy adults (N = 65), we show that articulation impairments in patients with PD are associated with aberrant activity in the left inferior frontal cortex, and that functional connectivity of this region with somatomotor cortices mediates the influence of cognitive decline on speech deficits.
SpikeShip: a method for fast, unsupervised discovery of high-dimensional neural spiking patterns
(2023)
Neural coding and memory formation depend on temporal spiking sequences that span high-dimensional neural ensembles. The unsupervised discovery and characterization of these spiking sequences requires a suitable dissimilarity measure to spiking patterns, which can then be used for clustering and decoding. Here, we present a new dissimilarity measure based on optimal transport theory called SpikeShip, which compares multi-neuron spiking patterns based on all the relative spike-timing relationships among neurons. SpikeShip computes the optimal transport cost to make all the relative spike timing relationships (across neurons) identical between two spiking patterns. We show that this transport cost can be decomposed into a temporal rigid translation term, which captures global latency shifts, and a vector of neuron-specific transport flows, which reflect inter-neuronal spike timing differences. SpikeShip can be effectively computed for high-dimensional neuronal ensembles, has a low (linear) computational cost that has the same order as the spike count, and is sensitive to higher-order correlations. Furthermore SpikeShip is binless, can handle any form of spike time distributions, is not affected by firing rate fluctuations, can detect patterns with a low signal-to-noise ratio, and can be effectively combined with a sliding window approach. We compare the advantages and differences between SpikeShip and other measures like SPIKE and Victor-P urpura distance. We applied SpikeShip to large-scale Neuropixel recordings during spontaneous activity and visual encoding. We show that high-dimensional spiking sequences detected via SpikeShip reliably distinguish between different natural images and different behavioral states. These spiking sequences carried complementary information to conventional firing rate codes. SpikeShip opens new avenues for studying neural coding and memory consolidation by rapid and unsupervised detection of temporal spiking patterns in high-dimensional neural ensembles.
An important question concerning inter-areal communication in the cortex is whether these interactions are synergistic, i.e. convey information beyond what can be performed by isolated signals. Here, we dissociated cortical interactions sharing common information from those encoding complementary information during prediction error processing. To this end, we computed co-information, an information-theoretical measure that distinguishes redundant from synergistic information among brain signals. We analyzed auditory and frontal electrocorticography (ECoG) signals in three common awake marmosets and investigated to what extent event-related-potentials (ERP) and broadband (BB) dynamics exhibit redundancy and synergy for auditory prediction error signals. We observed multiple patterns of redundancy and synergy across the entire cortical hierarchy with distinct dynamics. The information conveyed by ERPs and BB signals was highly synergistic even at lower stages of the hierarchy in the auditory cortex, as well as between lower and higher areas in the frontal cortex. These results indicate that the distributed representations of prediction error signals across the cortical hierarchy can be highly synergistic.
Sensory processing relies on interactions between excitatory and inhibitory neurons, which are often coordinated by 30-80Hz gamma oscillations. However, the specific contributions of distinct interneurons to gamma synchronization remain unclear. We performed high-density recordings from V1 in awake mice and used optogenetics to identify PV+ (Parvalbumin) and Sst+ (Somatostatin) interneurons. PV interneurons were highly phase-locked to visually-induced gamma oscillations. Sst cells were heterogeneous, with only a subset of narrow-waveform cells showing strong gamma phase-locking. Interestingly, PV interneurons consistently fired at an earlier phase in the gamma cycle (≈6ms or 60 degrees) than Sst interneurons. Consequently, PV and Sst activity showed differential temporal relations with excitatory cells. In particular, the 1st and 2nd spikes in burst events, which were strongly gamma phase-locked, shortly preceded PV and Sst activity, respectively. These findings indicate a primary role of PV interneurons in synchronizing excitatory cells and suggest that PV and Sst interneurons control the excitability of somatic and dendritic neural compartments with precise time delays coordinated by gamma oscillations.