Efficient Neural Coding Research Articles

Signatures of hierarchical temporal processing in the mouse visual system.

A core challenge for the brain is to process information across various timescales. This could be achieved by a hierarchical organization of temporal processing through intrinsic mechanisms (e.g., recurrent coupling or adaptation), but recent evidence from spike recordings of the rodent visual system seems to conflict with this hypothesis. Here, we used an optimized information-theoretic and classical autocorrelation analysis to show that information- and correlation timescales of spiking activity increase along the anatomical hierarchy of the mouse visual system under visual stimulation, while information-theoretic predictability decreases. Moreover, intrinsic timescales for spontaneous activity displayed a similar hierarchy, whereas the hierarchy of predictability was stimulus-dependent. We could reproduce these observations in a basic recurrent network model with correlated sensory input. Our findings suggest that the rodent visual system employs intrinsic mechanisms to achieve longer integration for higher cortical areas, while simultaneously reducing predictability for an efficient neural code.

Plasticity in inhibitory networks improves pattern separation in early olfactory processing.

By combining computational modeling, machine learning, and analysis of calcium imaging data, we demonstrate that associative and non-associative plasticity in the honeybee antennal lobe (AL) - first relay of the insect olfactory system - work together to enhance the contrast between rewarded and unrewarded odors. Training the AL's inhibitory network within specific odor environments enables the suppression of neural responses to common odor components, while amplifying responses to distinctive ones. This study sheds light on the olfactory system's ability to adapt and efficiently learn new odor-reward associations across varying environments, and it proposes innovative, energy-efficient principles applicable to artificial intelligence.

Two stages of bandwidth scaling drives efficient neural coding of natural sounds.

Theories of efficient coding propose that the auditory system is optimized for the statistical structure of natural sounds, yet the transformations underlying optimal acoustic representations are not well understood. Using a database of natural sounds including human speech and a physiologically-inspired auditory model, we explore the consequences of peripheral (cochlear) and mid-level (auditory midbrain) filter tuning transformations on the representation of natural sound spectra and modulation statistics. Whereas Fourier-based sound decompositions have constant time-frequency resolution at all frequencies, cochlear and auditory midbrain filters bandwidths increase proportional to the filter center frequency. This form of bandwidth scaling produces a systematic decrease in spectral resolution and increase in temporal resolution with increasing frequency. Here we demonstrate that cochlear bandwidth scaling produces a frequency-dependent gain that counteracts the tendency of natural sound power to decrease with frequency, resulting in a whitened output representation. Similarly, bandwidth scaling in mid-level auditory filters further enhances the representation of natural sounds by producing a whitened modulation power spectrum (MPS) with higher modulation entropy than both the cochlear outputs and the conventional Fourier MPS. These findings suggest that the tuning characteristics of the peripheral and mid-level auditory system together produce a whitened output representation in three dimensions (frequency, temporal and spectral modulation) that reduces redundancies and allows for a more efficient use of neural resources. This hierarchical multi-stage tuning strategy is thus likely optimized to extract available information and may underlies perceptual sensitivity to natural sounds.

Distinct early and late neural mechanisms regulate feature-specific sensory adaptation in the human visual system

A canonical feature of sensory systems is that they adapt to prolonged or repeated inputs, suggesting the brain encodes the temporal context in which stimuli are embedded. Sensory adaptation has been observed in the central nervous systems of many animal species, using techniques sensitive to a broad range of spatiotemporal scales of neural activity. Two competing models have been proposed to account for the phenomenon. One assumes that adaptation reflects reduced neuronal sensitivity to sensory inputs over time (the "fatigue" account); the other posits that adaptation arises due to increased neuronal selectivity (the "sharpening" account). To adjudicate between these accounts, we exploited the well-known "tilt aftereffect", which reflects adaptation to orientation information in visual stimuli. We recorded whole-brain activity with millisecond precision from human observers as they viewed oriented gratings before and after adaptation, and used inverted encoding modeling to characterize feature-specific neural responses. We found that both fatigue and sharpening mechanisms contribute to the tilt aftereffect, but that they operate at different points in the sensory processing cascade to produce qualitatively distinct outcomes. Specifically, fatigue operates during the initial stages of processing, consistent with tonic inhibition of feedforward responses, whereas sharpening occurs ~200 ms later, consistent with feedback or local recurrent activity. Our findings reconcile two major accounts of sensory adaptation, and reveal how this canonical process optimizes the detection of change in sensory inputs through efficient neural coding.

Efficient neural codes naturally emerge through gradient descent learning

Human sensory systems are more sensitive to common features in the environment than uncommon features. For example, small deviations from the more frequently encountered horizontal orientations can be more easily detected than small deviations from the less frequent diagonal ones. Here we find that artificial neural networks trained to recognize objects also have patterns of sensitivity that match the statistics of features in images. To interpret these findings, we show mathematically that learning with gradient descent in neural networks preferentially creates representations that are more sensitive to common features, a hallmark of efficient coding. This effect occurs in systems with otherwise unconstrained coding resources, and additionally when learning towards both supervised and unsupervised objectives. This result demonstrates that efficient codes can naturally emerge from gradient-like learning.

Scalable and Efficient Neural Speech Coding: A Hybrid Design

We present a scalable and efficient neural waveform coding system for speech compression. We formulate the speech coding problem as an autoencoding task, where a convolutional neural network (CNN) performs encoding and decoding as a neural waveform codec (NWC) during its feedforward routine. The proposed NWC also defines quantization and entropy coding as a trainable module, so the coding artifacts and bitrate control are handled during the optimization process. We achieve efficiency by introducing compact model components to NWC, such as gated residual networks and depthwise separable convolution. Furthermore, the proposed models are with a scalable architecture, cross-module residual learning (CMRL), to cover a wide range of bitrates. To this end, we employ the residual coding concept to concatenate multiple NWC autoencoding modules, where each NWC module performs residual coding to restore any reconstruction loss that its preceding modules have created. CMRL can scale down to cover lower bitrates as well, for which it employs linear predictive coding (LPC) module as its first autoencoder. The hybrid design integrates LPC and NWC by redefining LPC's quantization as a differentiable process, making the system training an end-to-end manner. The decoder of proposed system is with either one NWC (0.12 million parameters) in low to medium bitrate ranges (12 to 20 kbps) or two NWCs in the high bitrate (32 kbps). Although the decoding complexity is not yet as low as that of conventional speech codecs, it is significantly reduced from that of other neural speech coders, such as a WaveNet-based vocoder. For wide-band speech coding quality, our system yields comparable or superior performance to AMR-WB and Opus on TIMIT test utterances at low and medium bitrates. The proposed system can scale up to higher bitrates to achieve near transparent performance.

Unsupervised learning in images and audio to produce neural receptive fields: a primer and accessible notebook

Sensory processing relies on efficient computation driven by a combination of low-level unsupervised, statistical structural learning, and high-level task-dependent learning. In the earliest stages of sensory processing, sparse and independent coding strategies are capable of modeling neural processing using the same coding strategy with only a change in the input (e.g., grayscale images, color images, and audio). We present a consolidated review of Independent Component Analysis (ICA) as an efficient neural coding scheme with the ability to model early visual and auditory neural processing. We created a self-contained, accessible Jupyter notebook using Python to demonstrate the efficient coding principle for different modalities following a consistent five-step strategy. For each modality, derived receptive field models from natural and non-natural inputs are contrasted, demonstrating how neural codes are not produced when the inputs sufficiently deviate from those animals were evolved to process. Additionally, the demonstration shows that ICA produces more neurally-appropriate receptive field models than those based on common compression strategies, such as Principal Component Analysis. The five-step strategy not only produces neural-like models but also promotes reuse of code to emphasize the input-agnostic nature where each modality can be modeled with only a change in inputs. This notebook can be used to readily observe the links between unsupervised machine learning strategies and early sensory neuroscience, improving our understanding of flexible data-driven neural development in nature and future applications.

Durable memories and efficient neural coding through mnemonic training using the method of loci

Mnemonic techniques, such as the method of loci, can powerfully boost memory. We compared memory athletes ranked among the world's top 50 in memory sports to mnemonics-naïve controls. In a second study, participants completed a 6-week memory training, working memory training, or no intervention. Behaviorally, memory training enhanced durable, longer-lasting memories. Functional magnetic resonance imaging during encoding and recognition revealed task-based activation decreases in lateral prefrontal, as well as in parahippocampal and retrosplenial cortices in both memory athletes and participants after memory training, partly associated with better performance after 4 months. This was complemented by hippocampal-neocortical coupling during consolidation, which was stronger the more durable memories participants formed. Our findings advance knowledge on how mnemonic training boosts durable memory formation through decreased task-based activation and increased consolidation thereafter. This is in line with conceptual accounts of neural efficiency and highlights a complex interplay of neural processes critical for extraordinary memory.

Variation in the “coefficient of variation”: Rethinking the violation of the scalar property in time-duration judgments

The coefficient of variation (CV), also known as relative standard deviation, has been used to measure the constancy of the Weber fraction, a key signature of efficient neural coding in time perception. It has long been debated whether or not duration judgments follow Weber's law, with arguments based on examinations of the CV. However, what has been largely ignored in this debate is that the observed CVs may be modulated by temporal context and decision uncertainty, thus questioning conclusions based on this measure. Here, we used a temporal reproduction paradigm to examine the variation of the CV with two types of temporal context: full-range mixed vs. sub-range blocked intervals, separately for intervals presented in the visual and auditory modalities. We found a strong contextual modulation of both interval-duration reproductions and the observed CVs. We then applied a two-stage Bayesian model to predict those variations. Without assuming a violation of the constancy of the Weber fraction, our model successfully predicted the central-tendency effect and the variation in the CV. Our findings and modeling results indicate that both the accuracy and precision of our timing behavior are highly dependent on the temporal context and decision uncertainty. And, critically, they advise caution with using variations of the CV to reject the constancy of the Weber fraction of duration estimation.

Adaptive Efficient Coding of Correlated Acoustic Properties.

Natural sounds such as vocalizations often have covarying acoustic attributes, resulting in redundancy in neural coding. The efficient coding hypothesis proposes that sensory systems are able to detect such covariation and adapt to reduce redundancy, leading to more efficient neural coding. Recent psychoacoustic studies have shown the auditory system can rapidly adapt to efficiently encode two covarying dimensions as a single dimension, following passive exposure to sounds in which temporal and spectral attributes covaried in a correlated fashion. However, these studies observed a cost to this adaptation, which was a loss of sensitivity to the orthogonal dimension. Here we explore the neural basis of this psychophysical phenomenon by recording single-unit responses from the primary auditory cortex in awake ferrets exposed passively to stimuli with two correlated attributes, similar in stimulus design to the psychoacoustic experiments in humans. We found: (1) the signal-to-noise ratio of spike-rate coding of cortical responses driven by sounds with correlated attributes remained unchanged along the exposure dimension, but was reduced along the orthogonal dimension; (2) performance of a decoder trained with spike data to discriminate stimuli along the orthogonal dimension was equally reduced; (3) correlations between neurons tuned to the two covarying attributes decreased after exposure; and (4) these exposure effects still occurred if sounds were correlated along two acoustic dimensions, but varied randomly along a third dimension. These neurophysiological results are consistent with the efficient coding hypothesis and may help deepen our understanding of how the auditory system encodes and represents acoustic regularities and covariance.SIGNIFICANCE STATEMENT The efficient coding (EC) hypothesis (Attneave, 1954; Barlow, 1961) proposes that the neural code in sensory systems efficiently encodes natural stimuli by minimizing the number of spikes to transmit a sensory signal. Results of recent psychoacoustic studies in humans are consistent with the EC hypothesis in that, following passive exposure to stimuli with correlated attributes, the auditory system rapidly adapts so as to more efficiently encode the two covarying dimensions as a single dimension. In the current neurophysiological experiments, using a similar stimulus design and the experimental paradigm to the psychoacoustic studies of Stilp et al. (2010) and Stilp and Kluender (2011, 2012, 2016), we recorded responses from single neurons in the auditory cortex of the awake ferret, showing adaptive efficient neural coding of two correlated acoustic attributes.

Active dendrites regulate the spatiotemporal spread of signaling microdomains.

Microdomains that emerge from spatially constricted spread of biochemical signaling components play a central role in several neuronal computations. Although dendrites, endowed with several voltage-gated ion channels, form a prominent structural substrate for microdomain physiology, it is not known if these channels regulate the spatiotemporal spread of signaling microdomains. Here, we employed a multiscale, morphologically realistic, conductance-based model of the hippocampal pyramidal neuron that accounted for experimental details of electrical and calcium-dependent biochemical signaling. We activated synaptic N-Methyl-d-Aspartate receptors through theta-burst stimulation (TBS) or pairing (TBP) and assessed microdomain propagation along a signaling pathway that included calmodulin, calcium/calmodulin-dependent protein kinase II (CaMKII) and protein phosphatase 1. We found that the spatiotemporal spread of the TBS-evoked microdomain in phosphorylated CaMKII (pCaMKII) was amplified in comparison to that of the corresponding calcium microdomain. Next, we assessed the role of two dendritically expressed inactivating channels, one restorative (A-type potassium) and another regenerative (T-type calcium), by systematically varying their conductances. Whereas A-type potassium channels suppressed the spread of pCaMKII microdomains by altering the voltage response to TBS, T-type calcium channels enhanced this spread by modulating TBS-induced calcium influx without changing the voltage. Finally, we explored cross-dependencies of these channels with other model components, and demonstrated the heavy mutual interdependence of several biophysical and biochemical properties in regulating microdomains and their spread. Our conclusions unveil a pivotal role for dendritic voltage-gated ion channels in actively amplifying or suppressing biochemical signals and their spatiotemporal spread, with critical implications for clustered synaptic plasticity, robust information transfer and efficient neural coding.

Efficient coding matters in the organization of the early visual system

Individual areas in the brain are organized into a hierarchical network as a result of evolution. Previous work indicated that the receptive fields (RFs) of individual areas have been evolved to favor metabolically efficient neural codes. In this paper, we propose that not only the RFs of individual areas, but also the organization of adjacent neurons and the hierarchical structure composed of these areas have been evolved to support efficient coding. To verify this hypothesis, we introduce a feed-forward three-layer network to simulate the early stages of human visual system. We emphasize that the network is not a purely feed-forward one since it also includes intra-layer connections, which are essential but usually ignored in the literature. Simulation results strongly reveal that (1) the obtained RFs of the simulated retinal ganglion cells (RGCs) or neurons in the lateral geniculate nucleus (LGN) and V1 simple neurons are consistent to the neurophysiological data; (2) the responses of closer RGCs are more correlated, and V1 simple neurons with similar orientations prefer to cluster together; (3) the hierarchical organization of the early visual system is beneficial for saving energy, which accords with the requirement of metabolically efficient neural coding in the process of human brain evolution.

Behavioral state modulates the ON visual motion pathway of Drosophila

The behavioral state of an animal can dynamically modulate visual processing. In flies, the behavioral state is known to alter the temporal tuning of neurons that carry visual motion information into the central brain. However, where this modulation occurs and how it tunes the properties of this neural circuit are not well understood. Here, we show that the behavioral state alters the baseline activity levels and the temporal tuning of the first directionally selective neuron in the ON motion pathway (T4) as well as its primary input neurons (Mi1, Tm3, Mi4, Mi9). These effects are especially prominent in the inhibitory neuron Mi4, and we show that central octopaminergic neurons provide input to Mi4 and increase its excitability. We further show that octopamine neurons are required for sustained behavioral responses to fast-moving, but not slow-moving, visual stimuli in walking flies. These results indicate that behavioral-state modulation acts directly on the inputs to the directionally selective neurons and supports efficient neural coding of motion stimuli.

Inputs to prefrontal cortex support visual recognition in the aging brain.

Predictive coding models of brain function propose that top-down cortical signals promote efficient neural codes by carrying predictions of upcoming sensory events. We hypothesized that older brains would employ these codes more prominently given their longer repertoire of sensory experience. We measured the connectivity underlying stimulus-evoked responses in cortical visual networks using electroencephalography and dynamic causal modeling and found that in young adults with reported normal or corrected-to-normal vision, signals propagated from early visual regions and reverberated along reciprocal connections to temporal, parietal and frontal cortices, while in contrast, the network was driven by both early visual and prefrontal inputs in older adults with reported normal or corrected-to-normal vision. Previously thought of as exceptions to the rule of bottom-up signal propagation, our results demonstrate a prominent role for prefrontal inputs in driving vision in aged brains in line with lifespan-dependent predictive neural codes.

Temporal decorrelation by SK channels enables efficient neural coding and perception of natural stimuli.

It is commonly assumed that neural systems efficiently process natural sensory input. However, the mechanisms by which such efficient processing is achieved, and the consequences for perception and behaviour remain poorly understood. Here we show that small conductance calcium-activated potassium (SK) channels enable efficient neural processing and perception of natural stimuli. Specifically, these channels allow for the high-pass filtering of sensory input, thereby removing temporal correlations or, equivalently, whitening frequency response power. Varying the degree of adaptation through pharmacological manipulation of SK channels reduced efficiency of coding of natural stimuli, which in turn gave rise to predictable changes in behavioural responses that were no longer matched to natural stimulus statistics. Our results thus demonstrate a novel mechanism by which the nervous system can implement efficient processing and perception of natural sensory input that is likely to be shared across systems and species.

Beauty is in the efficient coding of the beholder.

Sexual ornaments are often assumed to be indicators of mate quality. Yet it remains poorly known how certain ornaments are chosen before any coevolutionary race makes them indicative. Perceptual biases have been proposed to play this role, but known biases are mostly restricted to a specific taxon, which precludes evaluating their general importance in sexual selection. Here we identify a potentially universal perceptual bias in mate choice. We used an algorithm that models the sparseness of the activity of simple cells in the primary visual cortex (or V1) of humans when coding images of female faces. Sparseness was found positively correlated with attractiveness as rated by men and explained up to 17% of variance in attractiveness. Because V1 is adapted to process signals from natural scenes, in general, not faces specifically, our results indicate that attractiveness for female faces is influenced by a visual bias. Sparseness and more generally efficient neural coding are ubiquitous, occurring in various animals and sensory modalities, suggesting that the influence of efficient coding on mate choice can be widespread in animals.

The Efficient Coding of Speech: Cross-Linguistic Differences.

Neural coding in the auditory system has been shown to obey the principle of efficient neural coding. The statistical properties of speech appear to be particularly well matched to the auditory neural code. However, only English has so far been analyzed from an efficient coding perspective. It thus remains unknown whether such an approach is able to capture differences between the sound patterns of different languages. Here, we use independent component analysis to derive information theoretically optimal, non-redundant codes (filter populations) for seven typologically distinct languages (Dutch, English, Japanese, Marathi, Polish, Spanish and Turkish) and relate the statistical properties of these filter populations to documented differences in the speech rhythms (Analysis 1) and consonant inventories (Analysis 2) of these languages. We show that consonant class membership plays a particularly important role in shaping the statistical structure of speech in different languages, suggesting that acoustic transience, a property that discriminates consonant classes from one another, is highly relevant for efficient coding.

Multiplexed coding through synchronous and asynchronous spiking

The prodigious capacity of our brain to process information relies on efficient neural coding strategies. In engineered systems, bandwidth is often increased through multiplexing multiple signals are simultaneously, yet independently, transmitted through a single communication channel. We have proposed previously that neural systems might implement the same sort of solution [1]. Here, we tested if/how multiplexed coding could be achieved through combined rate and temporal coding. We hypothesized that a set of neurons could independently encode two signals by using asynchronous spike rate to encode one signal and synchronous spike timing to encode the other. To test our hypothesis, we built a feed-forward neural network comprising Morris-Lecar (ML) model neurons. All neurons received a common input constructed from two distinct signals, slow and fast, plus uncorrelated fast noise. According to our hypothesis, slow and fast signals are independently encoded by different types of spikes; in other words, differentially correlated output spikes, namely, asynchronous (Async) and synchronous (Sync), enable encoding of slow and fast signals, respectively. To assess the feasibility of the multiplexed coding scheme, recorded spikes were classified into two independent classes based on the peristimulus time histogram (PSTH) calculated from the entire set of neurons. Spikes whose instantaneous rates exceeded a threshold were designated “Sync” and all others were designated “Async”. The spike triggered average (STA) was calculated for slow and fast signals using Sync and Async spikes, resulting in four different STAs. The Async-slow and Sync-fast STAs were clearly structured whereas the other two were not (Figure 1A). Using

Wide-field feedback neurons dynamically tune early visual processing.

An important strategy for efficient neural coding is to match the range of cellular responses to the distribution of relevant input signals. However, the structure and relevance of sensory signals depend on behavioral state. Here, we show that behavior modifies neural activity at the earliest stages of fly vision. We describe a class of wide-field neurons that provide feedback to the most peripheral layer of the Drosophila visual system, the lamina. Using in vivo patch-clamp electrophysiology, we found that lamina wide-field neurons respond to low-frequency luminance fluctuations. Recordings in flying flies revealed that the gain and frequency tuning of wide-field neurons change during flight, and that these effects are mimicked by the neuromodulator octopamine. Genetically silencing wide-field neurons increased behavioral responses to slow-motion stimuli. Together, these findings identify a cell type that is gated by behavior to enhance neural coding by subtracting low-frequency signals from the inputs to motion detection circuits.

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.