A primary goal in systems neuroscience is to understand how neural spike responses encode information about the external world. A popular approach to this problem is to build an explicit probabilistic model that characterizes the encoding relationship in terms of a cascade of stages: (1) linear dimensionality reduction of a high-dimensional stimulus space using a bank of filters or receptive fields (RFs); (2) a nonlinear function from filter outputs to spike rate; and (3) a stochastic spiking process with recurrent feedback. These models have described single- and multi-neuron spike responses in a wide variety of brain areas. This dissertation addresses Bayesian methods to efficiently estimate the linear and non-linear stages of the cascade encoding model. In the first part, the dissertation describes a novel Bayesian receptive field estimator based on a hierarchical prior that flexibly incorporates knowledge about the shapes of neural receptive fields. This estimator achieves error rates several times lower than existing methods, and can be applied to a variety of other neural inference problems such as extracting structure in fMRI data. The dissertation also presents active learning frameworks developed for receptive field estimation incorporating a hierarchical prior in real-time neurophysiology experiments. In addition, the dissertation describes a novel low-rank model for the high dimensional receptive field, combined with a hierarchical prior for more efficient receptive field estimation. In the second part, the dissertation describes new models for neural nonlinearities using Gaussian processes (GPs) and Bayesian active learning algorithms in closed-loop neurophysiology experiments to rapidly estimate neural nonlinearities. The dissertation also presents several stimulus selection criteria and compare their performance in neural nonlinearity estimation. Furthermore, the dissertation presents a variation of the new models by including an additional latent Gaussian noise source, to infer the degree of over-dispersion in neural spike responses. The proposed model successfully captures various mean-variance relationships in neural spike responses and achieves higher prediction accuracy than previous models. / text

Vocal communication sounds are an important class of signals due to their role in social interaction, reproduction, and survival. The higher-order mechanisms by which our auditory system detects and discriminates these sounds to generate perception is still poorly understood. The auditory cortex is thought to play an important role in this process, and our current work provides new evidence that the auditory cortex changes its neural representation of sounds that are acquired in natural social contexts. We use a mouse ultrasonic communication system between pups and adult females to elucidate this. We record single neurons in the auditory cortex of awake mice, and assess the cortical differences between animals that either do (mothers) or do not (naïve virgins) recognize the pup ultrasounds as behaviorally relevant. We then evaluate the role that pup experience and the maternal physiological state play in this cortical plasticity. Finally, we develop a model to predict the responses to pup vocalizations as a way to segregate the diversity of cortical neuronal responses in the hope of more clearly assessing their roles in processing acoustic features. Our results demonstrate the detailed nature by which the core auditory cortex processes natural vocalizations, showing how it changes to represent behavioral relevance.

Many insects have auditory systems capable of detecting the ultrasonic calls of insectivorous bats and use these cues to evade capture. I tested the hypothesis that bats can decrease the conspicuousness of their echolocation calls by varying three call features: duration, repetition rate and ramp times. This was done by examining the AN2 command interneuron’s response to these features in the cricket, Teleogryllus oceanicus, after describing the firing pattern necessary for evasive behaviour. Past studies on duration and repetition rate suggest increased thresholds for short durations and low repetition rates. Measurements of the AN2 response, which controls evasive behaviour, indicated that increased thresholds were a result of a decrease in bursting, raw spike numbers and an increase in latencies in the AN2. Results suggest that there is pressure on bats to evade early detection and that this can be done by employing large ramp times in search phase echolocation calls.

Many insects have auditory systems capable of detecting the ultrasonic calls of insectivorous bats and use these cues to evade capture. I tested the hypothesis that bats can decrease the conspicuousness of their echolocation calls by varying three call features: duration, repetition rate and ramp times. This was done by examining the AN2 command interneuron’s response to these features in the cricket, Teleogryllus oceanicus, after describing the firing pattern necessary for evasive behaviour. Past studies on duration and repetition rate suggest increased thresholds for short durations and low repetition rates. Measurements of the AN2 response, which controls evasive behaviour, indicated that increased thresholds were a result of a decrease in bursting, raw spike numbers and an increase in latencies in the AN2. Results suggest that there is pressure on bats to evade early detection and that this can be done by employing large ramp times in search phase echolocation calls.

Submitted by Automa??o e Estat?stica (sst@bczm.ufrn.br) on 2016-02-17T22:41:57Z No. of bitstreams: 1 VitorLopesDosSantos_TESE.pdf: 14542830 bytes, checksum: 6152cf87bb56d50e13c1279c3841c19e (MD5) / Approved for entry into archive by Arlan Eloi Leite Silva (eloihistoriador@yahoo.com.br) on 2016-02-19T22:43:00Z (GMT) No. of bitstreams: 1 VitorLopesDosSantos_TESE.pdf: 14542830 bytes, checksum: 6152cf87bb56d50e13c1279c3841c19e (MD5) / Made available in DSpace on 2016-02-19T22:43:00Z (GMT). No. of bitstreams: 1 VitorLopesDosSantos_TESE.pdf: 14542830 bytes, checksum: 6152cf87bb56d50e13c1279c3841c19e (MD5) Previous issue date: 2015-02-25 / Os recentes avan?os t?cnicos das duas ?ltimas d?cadas para o registro de sinais neuroeletrofisiol?gicos foram essenciais para que se testassem hip?teses h? muito propostas acerca de como c?lulas nervosas processam e armazenam informa??o. No entanto, ao permitir maior detalhamento dos dados coletados, as novas tecnologias levam inevitavelmente ao aumento de sua complexidade estat?stica e, consequentemente, ? necessidade de novas ferramentas matem?tico-computacionais para sua an?lise. Nesta tese, apresentamos novos m?todos para a an?lise de dois componentes fundamentais nas atuais teorias da codifica??o neural: (1) assembleias celulares, definidas pela co-ativa??o de subgrupos neuronais; e (2) o padr?o temporal de atividade de neur?nios individuais. Em rela??o a (1), desenvolvemos um m?todo baseado em an?lise de componentes independentes para identificar e rastrear padr?es de co-ativa??o significativos com alta resolu??o temporal. Superamos limita??es de m?todos anteriores, ao efetivamente isolar assembleias e abrir a possibilidade de analisar simultaneamente grandes popula??es neuronais. Em rela??o a (2), apresentamos uma nova t?cnica para a extra??o de padr?es de atividade em trens de disparo baseada na decomposi??o wavelet. Demonstramos, por meio de simula??es e de aplica??o a dados reais, que nossa ferramenta supera as mais utilizadas atualmente para decodificar respostas de neur?nios e estimar a informa??o de Shannon entre trens de disparos e est?mulos externos.

Neurons can encode information using the rate of their action potentials, making the relation between input rate and output rate a prominent feature of neuronal information processing. This relation, known as I{O function, can rapidly change in response to various factors or neuronal processes. Most noticeably, a neuron can undergo a multiplicative operation, resulting in a change of the slope of its I{O curve, also know as gain change. Gain changes represent multiplicative operations, and they are wide- spread. They have been found to play an important role in the encoding of spatial location and coordinate transformation, to signal amplification, and other neuronal functions. One of the factors found to introduce and control neuronal gain is synaptic Short Term Depression (STD). We use both integrate-and- re and conductance based neuron models to identify the effect of STD in excitatory and inhibitory modulatory input. More specifically, we are interested in the effect of STD at the inhibitory synapse from Purkinje cells to cerebellar nucleus neurons. Using a previously published, biologically realistic model, we find that the presence of STD results in a gain change. Most importantly we identify STD at the inhibitory synapse to enable excitation-mediated gain control. To isolate the mechanism that allows excitation to control gain, even though STD is applied at a different synapse, we first show that the overall effect is mediated by average conductance. Having done this, we find that the effect of STD is based on the non-linearity introduced in the relation between input rate and average conductance. We find this effect to vary, depending on the position of the I{O function on the input rate axis. Modulatory input shifts the I{O curve along the input rate axis, consequently shifting it to a position where STD has a different effect. The gain differences in the STD effects between the two positions enable excitation to perform gain control.

<p>A considerable amount of our everyday tactile experience requires interactions between textured surfaces and our fingertips. Such interactions elicit complex vibrations on our skin surface, which are encoded by the mechanosensitive afferents and conveyed to the brain where the perception of the textures emerges seemingly effortlessly. Intuitively, a fundamental question that may be asked is: “what features of the vibration stimuli are behaviourally relevant and what are the neural signatures of these features?” The goal of this thesis is to investigate these questions, which we have done using a combination of theoretical and experimental approaches.</p> <p>Our theoretical approach (in Chapter 2) has been to create an ideal Bayesian perceptual observer that utilizes all the information available in a spike-rate based neural code and makes optimal inferences regarding the amplitude and the frequency of vibration stimuli. Our experimental approach has been to estimate the performance of human participants in vibrotactile detection (in Chapter 3), and in amplitude and frequency discrimination (in Chapter 4) tasks by using psychophysical procedures.</p> <p>The results of these approaches suggest that the human perceptual observer, i.e. the human nervous system, probably uses a rate code to represent vibrotactile amplitude, but a non-rate code, such as a spike timing code, to represent vibrotactile frequency. Additionally, we conclude that humans are capable of inferring and separately perceiving the amplitude and frequency of vibrotactile stimuli; however, depending on experimental tasks, humans might also rely on a feature that combines the amplitude and frequency of vibrotactile stimuli.</p> / Doctor of Philosophy (PhD)

A stochastic leaky integrate-and-fire neuron model was implemented in this study to simulate the spiking activity of the electrosensory "P-unit" receptor neurons of the weakly electric fish Apteronotus leptorhynchus. In the context of sensory coding, these cells have been previously shown to respond in experiment to natural random narrowband signals with either a linear or nonlinear coding scheme, depending on the intrinsic firing rate of the cell in the absence of external stimulation. It was hypothesised in this study that this duality is due to the relation of the stimulus to the neuron's excitation threshold. This hypothesis was validated with the model by lowering the threshold of the neuron or increasing its intrinsic noise, or randomness, either of which made the relation between firing rate and input strength more linear. Furthermore, synchronous P-unit firing to a common input also plays a role in decoding the stimulus at deeper levels of the neural pathways. Synchronisation and desynchronisation between multiple model responses for different types of natural communication signals were shown to agree with experimental observations. A novel result of resonance-induced synchrony enhancement of P-units to certain communication frequencies was also found.

A stochastic leaky integrate-and-fire neuron model was implemented in this study to simulate the spiking activity of the electrosensory "P-unit" receptor neurons of the weakly electric fish Apteronotus leptorhynchus. In the context of sensory coding, these cells have been previously shown to respond in experiment to natural random narrowband signals with either a linear or nonlinear coding scheme, depending on the intrinsic firing rate of the cell in the absence of external stimulation. It was hypothesised in this study that this duality is due to the relation of the stimulus to the neuron's excitation threshold. This hypothesis was validated with the model by lowering the threshold of the neuron or increasing its intrinsic noise, or randomness, either of which made the relation between firing rate and input strength more linear. Furthermore, synchronous P-unit firing to a common input also plays a role in decoding the stimulus at deeper levels of the neural pathways. Synchronisation and desynchronisation between multiple model responses for different types of natural communication signals were shown to agree with experimental observations. A novel result of resonance-induced synchrony enhancement of P-units to certain communication frequencies was also found.

A stochastic leaky integrate-and-fire neuron model was implemented in this study to simulate the spiking activity of the electrosensory "P-unit" receptor neurons of the weakly electric fish Apteronotus leptorhynchus. In the context of sensory coding, these cells have been previously shown to respond in experiment to natural random narrowband signals with either a linear or nonlinear coding scheme, depending on the intrinsic firing rate of the cell in the absence of external stimulation. It was hypothesised in this study that this duality is due to the relation of the stimulus to the neuron's excitation threshold. This hypothesis was validated with the model by lowering the threshold of the neuron or increasing its intrinsic noise, or randomness, either of which made the relation between firing rate and input strength more linear. Furthermore, synchronous P-unit firing to a common input also plays a role in decoding the stimulus at deeper levels of the neural pathways. Synchronisation and desynchronisation between multiple model responses for different types of natural communication signals were shown to agree with experimental observations. A novel result of resonance-induced synchrony enhancement of P-units to certain communication frequencies was also found.