An efficient coding hypothesis links sparsity and selectivity of neural responses

Dendrites and efficiency: Optimizing performance and resource utilization

The brain is a highly efficient system that has evolved to optimize performance under limited resources. In this review, we highlight recent theoretical and experimental studies that support the view that dendrites make information processing and storage in the brain more efficient. This is achieved through the dynamic modulation of integration versus segregation of inputs and activity within a neuron. We argue that under conditions of limited energy and space, dendrites help biological networks to implement complex functions such as processing natural stimuli on behavioral timescales, performing the inference process on those stimuli in a context-specific manner, and storing the information in overlapping populations of neurons. A global picture starts to emerge, in which dendrites help the brain achieve efficiency through a combination of optimization strategies that balance the tradeoff between performance and resource utilization.

Neural heterogeneity promotes robust learning

The brain is a hugely diverse, heterogeneous structure. Whether or not heterogeneity at the neural level plays a functional role remains unclear, and has been relatively little explored in models which are often highly homogeneous. We compared the performance of spiking neural networks trained to carry out tasks of real-world difficulty, with varying degrees of heterogeneity, and found that heterogeneity substantially improved task performance. Learning with heterogeneity was more stable and robust, particularly for tasks with a rich temporal structure. In addition, the distribution of neuronal parameters in the trained networks is similar to those observed experimentally. We suggest that the heterogeneity observed in the brain may be more than just the byproduct of noisy processes, but rather may serve an active and important role in allowing animals to learn in changing environments.

Clustering of Neural Activity: A Design Principle for Population Codes

We propose that correlations among neurons are generically strong enough to organize neural activity patterns into a discrete set of clusters, which can each be viewed as a population codeword. Our reasoning starts with the analysis of retinal ganglion cell data using maximum entropy models, showing that the population is robustly in a frustrated, marginally sub-critical, or glassy, state. This leads to an argument that neural populations in many other brain areas might share this structure. Next, we use latent variable models to show that this glassy state possesses well-defined clusters of neural activity. Clusters have three appealing properties: (i) clusters exhibit error correction, i.e., they are reproducibly elicited by the same stimulus despite variability at the level of constituent neurons; (ii) clusters encode qualitatively different visual features than their constituent neurons; and (iii) clusters can be learned by downstream neural circuits in an unsupervised fashion. We hypothesize that these properties give rise to a "learnable" neural code which the cortical hierarchy uses to extract increasingly complex features without supervision or reinforcement.

STRFs in primary auditory cortex emerge from masking-based statistics of natural sounds

We investigate how the neural processing in auditory cortex is shaped by the statistics of natural sounds. Hypothesising that auditory cortex (A1) represents the structural primitives out of which sounds are composed, we employ a statistical model to extract such components. The input to the model are cochleagrams which approximate the non-linear transformations a sound undergoes from the outer ear, through the cochlea to the auditory nerve. Cochleagram components do not superimpose linearly, but rather according to a rule which can be approximated using the max function. This is a consequence of the compression inherent in the cochleagram and the sparsity of natural sounds. Furthermore, cochleagrams do not have negative values. Cochleagrams are therefore not matched well by the assumptions of standard linear approaches such as sparse coding or ICA. We therefore consider a new encoding approach for natural sounds, which combines a model of early auditory processing with maximal causes analysis (MCA), a sparse coding model which captures both the non-linear combination rule and non-negativity of the data. An efficient truncated EM algorithm is used to fit the MCA model to cochleagram data. We characterize the generative fields (GFs) inferred by MCA with respect to in vivo neural responses in A1 by applying reverse correlation to estimate spectro-temporal receptive fields (STRFs) implied by the learned GFs. Despite the GFs being non-negative, the STRF estimates are found to contain both positive and negative subfields, where the negative subfields can be attributed to explaining away effects as captured by the applied inference method. A direct comparison with ferret A1 shows many similar forms, and the spectral and temporal modulation tuning of both ferret and model STRFs show similar ranges over the population. In summary, our model represents an alternative to linear approaches for biological auditory encoding while it captures salient data properties and links inhibitory subfields to explaining away effects.

The information carried by natural sounds enters the cortex of mammals in a specific format: the cochleagram. Instead of representing the original pressure waveforms, the inner ear represents how the energy in a sound is distributed across frequency bands and how the energy distribution evolves over time. The generation of cochleagrams is highly non-linear resulting in the dominance of one sound source per time-frequency bin under natural conditions (masking). Auditory cortex is believed to decompose cochleagrams into structural primitives, i.e., reappearing regular spectro-temporal subpatterns that make up cochleagram patterns (similar to edges in images). However, such a decomposition has so far only been modeled without considering masking and non-negativity. Here we apply a novel non-linear sparse coding model that can capture masking non-linearities and non-negativities. When trained on cochleagrams of natural sounds, the model gives rise to an encoding primarily based-on spectro-temporally localized components. If stimulated by a sound, the encoding units compete to explain its contents. The competition is a direct consequence of the statistical sound model, and it results in neural responses being best described by spectro-temporal receptive fields (STRFs) with positive and negative subfields. The emerging STRFs show a higher similarity to experimentally measured STRFs than a model without masking, which provides evidence for cortical encoding being consistent with the masking based sound statistics of cochleagrams. Furthermore, and more generally, our study suggests for the first time that negative subfields of STRFs may be direct evidence for explaining away effects resulting from performing inference in an underlying statistical model.

Sensory cortex is optimized for prediction of future input

Neurons in sensory cortex are tuned to diverse features in natural scenes. But what determines which features neurons become selective to? Here we explore the idea that neuronal selectivity is optimized to represent features in the recent sensory past that best predict immediate future inputs. We tested this hypothesis using simple feedforward neural networks, which were trained to predict the next few moments of video or audio in clips of natural scenes. The networks developed receptive fields that closely matched those of real cortical neurons in different mammalian species, including the oriented spatial tuning of primary visual cortex, the frequency selectivity of primary auditory cortex and, most notably, their temporal tuning properties. Furthermore, the better a network predicted future inputs the more closely its receptive fields resembled those in the brain. This suggests that sensory processing is optimized to extract those features with the most capacity to predict future input.

A large part of our brain is devoted to processing the sensory inputs that we receive from the world. This allows us to tell, for example, whether we are looking at a cat or a dog, and if we are hearing a bark or a meow. Neurons in the sensory cortex respond to these stimuli by generating spikes of activity. Within each sensory area, neurons respond best to stimuli with precise properties: those in the primary visual cortex prefer edge-like structures that move in a certain direction at a given speed, while neurons in the primary auditory cortex favour sounds that change in loudness over a particular range of frequencies.

Singer et al. sought to understand why neurons respond to the particular features of stimuli that they do. Why do visual neurons react more to moving edges than to, say, rotating hexagons? And why do auditory neurons respond more to certain changing sounds than to, say, constant tones? One leading idea is that the brain tries to use as few spikes as possible to represent real-world stimuli. Known as sparse coding, this principle can account for much of the behaviour of sensory neurons.

Another possibility is that sensory areas respond the way they do because it enables them to best predict future sensory input. To test this idea, Singer et al. used a computer to simulate a network of neurons and trained this network to predict the next few frames of video clips using the previous few frames. When the network had learned this task, Singer et al. examined the neurons’ preferred stimuli. Like neurons in primary visual cortex, the simulated neurons typically responded most to edges that moved over time.

The same network was also trained in a similar way, but this time using sound. As for neurons in primary auditory cortex, the simulated neurons preferred sounds that changed in loudness at particular frequencies. Notably, for both vision and audition, the simulated neurons favoured recent inputs over those further into the past. In this way and others, they were more similar to real neurons than simulated neurons that used sparse coding.

Both artificial networks trained to foretell sensory input and the brain therefore favour the same types of stimuli: the ones that are good at helping to grasp future information. This suggests that the brain represents the sensory world so as to be able to best predict the future.

Knowing how the brain handles information from our senses may help to understand disorders associated with sensory processing, such as dyslexia and tinnitus. It may also inspire approaches for training machines to process sensory inputs, improving artificial intelligence.

A Neural Code That Is Isometric to Vocal Output and Correlates with Its Sensory Consequences

What cortical inputs are provided to motor control areas while they drive complex learned behaviors? We study this question in the nucleus interface of the nidopallium (NIf), which is required for normal birdsong production and provides the main source of auditory input to HVC, the driver of adult song. In juvenile and adult zebra finches, we find that spikes in NIf projection neurons precede vocalizations by several tens of milliseconds and are insensitive to distortions of auditory feedback. We identify a local isometry between NIf output and vocalizations: quasi-identical notes produced in different syllables are preceded by highly similar NIf spike patterns. NIf multiunit firing during song precedes responses in auditory cortical neurons by about 50 ms, revealing delayed congruence between NIf spiking and a neural representation of auditory feedback. Our findings suggest that NIf codes for imminent acoustic events within vocal performance.

Transmission of birdsong across generations requires tight interactions between auditory and vocal systems. However, how these interactions take place is poorly understood. We studied neuronal activity in the brain area located at the intersection between auditory and song motor areas, which is known as the nucleus interface of the nidopallium. By recording during singing from neurons in the nucleus interface of the nidopallium that project to motor areas, we found that their spiking precedes peaks in vocal amplitudes by about 50 ms. Notably, quasi-identical notes produced at different times in the song motif were preceded by highly similar spike patterns in these projection neurons. Such local isometry between output from the nucleus interface of the nidopallium and vocalizations suggests that projection neurons in this brain area code for imminent acoustic events within vocal performance. In support of this conclusion, during singing, projection neurons do not respond to playback of white noise sound stimuli, and activity in the nucleus interface of the nidopallium precedes by about 50 ms neural activity in the avian analogue of auditory cortex. Therefore, we conclude that the role of neuronal activity in the nucleus interface of the nidopallium could be to link desired auditory targets to suitable motor commands required for hitting these targets.

Relating observability and compressed sensing of time-varying signals in recurrent linear networks

In this paper, we study how the dynamics of recurrent networks, formulated as general dynamical systems, mediate the recovery of sparse, time-varying signals. Our formulation resembles the well-described problem of compressed sensing, but in a dynamic setting. We specifically consider the problem of recovering a high-dimensional network input, over time, from observation of only a subset of the network states (i.e., the network output). Our goal is to ascertain how the network dynamics may enable recovery, even if classical methods fail at each time instant. We are particularly interested in understanding performance in scenarios where both the input and output are corrupted by disturbance and noise, respectively. Our main results consist of the development of analytical conditions, including a generalized observability criterion, that ensure exact and stable input recovery in a dynamic, recurrent network setting.

Young, active and well-connected: adult-born neurons in the zebra finch are activated during singing

Neuronal replacement in the pallial song control nucleus HVC of adult zebra finches constitutes an interesting case of homeostatic plasticity; in spite of continuous addition and attrition of neurons in ensembles that code song elements, adult song remains remarkably invariant. New neurons migrate into HVC and later synapse with their target, arcopallial song nucleus RA (HVCRA). New HVCRA neurons respond to auditory stimuli (in anaesthetised animals), but whether and when they become functionally active during singing is unknown. We studied this, using 5-bromo-2'-deoxyuridine to birth-date neurons, combined with immunohistochemical detection of immediate-early gene (IEG) expression and retrograde tracer injections into RA to track connectivity. Interestingly, singing was followed by IEG expression in a substantial fraction of new neurons that were not retrogradely labelled from RA, suggesting a possible role in HVC-intrinsic network function. As new HVC neurons matured, the proportion of HVCRA neurons that expressed IEGs after singing increased significantly. Since it was previously shown that singing induces IEG expression in HVC also in deaf birds and that hearing song does not induce IEG expression in HVC, our data provide the first direct evidence that new HVC neurons are engaged in song motor behaviour.

Activity in a premotor cortical nucleus of zebra finches is locally organized and exhibits auditory selectivity in neurons but not in glia

Motor functions are often guided by sensory experience, most convincingly illustrated by complex learned behaviors. Key to sensory guidance in motor areas may be the structural and functional organization of sensory inputs and their evoked responses. We study sensory responses in large populations of neurons and neuron-assistive cells in the songbird motor area HVC, an auditory-vocal brain area involved in sensory learning and in adult song production. HVC spike responses to auditory stimulation display remarkable preference for the bird's own song (BOS) compared to other stimuli. Using two-photon calcium imaging in anesthetized zebra finches we measure the spatio-temporal structure of baseline activity and of auditory evoked responses in identified populations of HVC cells. We find strong correlations between calcium signal fluctuations in nearby cells of a given type, both in identified neurons and in astroglia. In identified HVC neurons only, auditory stimulation decorrelates ongoing calcium signals, less for BOS than for other sound stimuli. Overall, calcium transients show strong preference for BOS in identified HVC neurons but not in astroglia, showing diversity in local functional organization among identified neuron and astroglia populations.

Neural representation of a target auditory memory in a cortico-basal ganglia pathway

Vocal learning in songbirds, like speech acquisition in humans, entails a period of sensorimotor integration during which vocalizations are evaluated via auditory feedback and progressively refined to achieve an imitation of memorized vocal sounds. This process requires the brain to compare feedback of current vocal behavior to a memory of target vocal sounds. We report the discovery of two distinct populations of neurons in a cortico-basal ganglia circuit of juvenile songbirds (zebra finches, Taeniopygia guttata) during vocal learning: (1) one in which neurons are selectively tuned to memorized sounds and (2) another in which neurons are selectively tuned to self-produced vocalizations. These results suggest that neurons tuned to learned vocal sounds encode a memory of those target sounds, whereas neurons tuned to self-produced vocalizations encode a representation of current vocal sounds. The presence of neurons tuned to memorized sounds is limited to early stages of sensorimotor integration: after learning, the incidence of neurons encoding memorized vocal sounds was greatly diminished. In contrast to this circuit, neurons known to drive vocal behavior through a parallel cortico-basal ganglia pathway show little selective tuning until late in learning. One interpretation of these data is that representations of current and target vocal sounds in the shell circuit are used to compare ongoing patterns of vocal feedback to memorized sounds, whereas the parallel core circuit has a motor-related role in learning. Such a functional subdivision is similar to mammalian cortico-basal ganglia pathways in which associative-limbic circuits mediate goal-directed responses, whereas sensorimotor circuits support motor aspects of learning.

Selective and efficient neural coding of communication signals depends on early acoustic and social environment

Previous research has shown that postnatal exposure to simple, synthetic sounds can affect the sound representation in the auditory cortex as reflected by changes in the tonotopic map or other relatively simple tuning properties, such as AM tuning. However, their functional implications for neural processing in the generation of ethologically-based perception remain unexplored. Here we examined the effects of noise-rearing and social isolation on the neural processing of communication sounds such as species-specific song, in the primary auditory cortex analog of adult zebra finches. Our electrophysiological recordings reveal that neural tuning to simple frequency-based synthetic sounds is initially established in all the laminae independent of patterned acoustic experience; however, we provide the first evidence that early exposure to patterned sound statistics, such as those found in native sounds, is required for the subsequent emergence of neural selectivity for complex vocalizations and for shaping neural spiking precision in superficial and deep cortical laminae, and for creating efficient neural representations of song and a less redundant ensemble code in all the laminae. Our study also provides the first causal evidence for ‘sparse coding’, such that when the statistics of the stimuli were changed during rearing, as in noise-rearing, that the sparse or optimal representation for species-specific vocalizations disappeared. Taken together, these results imply that a layer-specific differential development of the auditory cortex requires patterned acoustic input, and a specialized and robust sensory representation of complex communication sounds in the auditory cortex requires a rich acoustic and social environment.