Limits...
Discriminative learning of receptive fields from responses to non-Gaussian stimulus ensembles.

Meyer AF, Diepenbrock JP, Happel MF, Ohl FW, Anemüller J - PLoS ONE (2014)

Bottom Line: Results demonstrate that the proposed approach successfully identifies correct underlying STRFs, even in cases where second-order methods based on the spike-triggered average (STA) do not.Applied to small data samples, the method is shown to converge on smaller amounts of experimental recordings and with lower estimation variance than the generalized linear model and recent information theoretic methods.Thus, CbRF estimation may prove useful for investigation of neuronal processes in response to natural stimuli and in settings where rapid adaptation is induced by experimental design.

View Article: PubMed Central - PubMed

Affiliation: Department of Medical Physics and Acoustics and Cluster of Excellence ''Hearing4all'', University of Oldenburg, Oldenburg, Germany.

ABSTRACT
Analysis of sensory neurons' processing characteristics requires simultaneous measurement of presented stimuli and concurrent spike responses. The functional transformation from high-dimensional stimulus space to the binary space of spike and non-spike responses is commonly described with linear-nonlinear models, whose linear filter component describes the neuron's receptive field. From a machine learning perspective, this corresponds to the binary classification problem of discriminating spike-eliciting from non-spike-eliciting stimulus examples. The classification-based receptive field (CbRF) estimation method proposed here adapts a linear large-margin classifier to optimally predict experimental stimulus-response data and subsequently interprets learned classifier weights as the neuron's receptive field filter. Computational learning theory provides a theoretical framework for learning from data and guarantees optimality in the sense that the risk of erroneously assigning a spike-eliciting stimulus example to the non-spike class (and vice versa) is minimized. Efficacy of the CbRF method is validated with simulations and for auditory spectro-temporal receptive field (STRF) estimation from experimental recordings in the auditory midbrain of Mongolian gerbils. Acoustic stimulation is performed with frequency-modulated tone complexes that mimic properties of natural stimuli, specifically non-Gaussian amplitude distribution and higher-order correlations. Results demonstrate that the proposed approach successfully identifies correct underlying STRFs, even in cases where second-order methods based on the spike-triggered average (STA) do not. Applied to small data samples, the method is shown to converge on smaller amounts of experimental recordings and with lower estimation variance than the generalized linear model and recent information theoretic methods. Thus, CbRF estimation may prove useful for investigation of neuronal processes in response to natural stimuli and in settings where rapid adaptation is induced by experimental design.

Show MeSH

Related in: MedlinePlus

STRF estimation from gerbil inferior colliculus (IC) responses to frequency-modulated (FM) sweep complex stimuli.(A) Example segment of block-design FM tone complex with length 1 s. Amplitude scaling in decibel (dB), dynamic range limited to 25 dB below maximum for visualization. (B) Stimulus amplitude histogram, shown for each spectral band after centering; red (blue) indicate high (low) probability, respectively. (C) Normalized spectro-temporal auto-correlation function of stimulus ensemble. (D) STRFs estimated from recorded responses of four gerbil IC units (columns) with four inference methods (rows, explanation cf. Table 1). All units had best frequency below 8 kHz and analysis was restricted to the range 0.5 kHz to 8 kHz. The spike waveform density function of each unit is shown on top of each column, verifying single-unit activity [66]. Spectro-temporally transient ("diagonal'') patterns that are exhibited in the ridge regression-based estimates (top row) lack confirmation in the MID-, GLM-, and CbRF-derived STRF estimates (lower three rows). Thus, we hypothesize that these are an artefactual result originating from higher-order correlations and distribution asymmetries within the stimulus ensemble which the ridge regression method is not robust to. In general, MID, GLM, and CbRF produce very similar STRF estimates, with the latter two methods revealing a slightly finer tuning in some cases. (E) Validation experiment with dynamic moving ripple (DMR) stimuli responses recorded from two identical units (units C and D) as shown in experiment panel D. Spectro-temporal transients absent in all methods' STRF estimates, presumably due the absence of higher-order correlations in the DMR stimuli and consistent with the explanation of panel D results.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3974709&req=5

pone-0093062-g007: STRF estimation from gerbil inferior colliculus (IC) responses to frequency-modulated (FM) sweep complex stimuli.(A) Example segment of block-design FM tone complex with length 1 s. Amplitude scaling in decibel (dB), dynamic range limited to 25 dB below maximum for visualization. (B) Stimulus amplitude histogram, shown for each spectral band after centering; red (blue) indicate high (low) probability, respectively. (C) Normalized spectro-temporal auto-correlation function of stimulus ensemble. (D) STRFs estimated from recorded responses of four gerbil IC units (columns) with four inference methods (rows, explanation cf. Table 1). All units had best frequency below 8 kHz and analysis was restricted to the range 0.5 kHz to 8 kHz. The spike waveform density function of each unit is shown on top of each column, verifying single-unit activity [66]. Spectro-temporally transient ("diagonal'') patterns that are exhibited in the ridge regression-based estimates (top row) lack confirmation in the MID-, GLM-, and CbRF-derived STRF estimates (lower three rows). Thus, we hypothesize that these are an artefactual result originating from higher-order correlations and distribution asymmetries within the stimulus ensemble which the ridge regression method is not robust to. In general, MID, GLM, and CbRF produce very similar STRF estimates, with the latter two methods revealing a slightly finer tuning in some cases. (E) Validation experiment with dynamic moving ripple (DMR) stimuli responses recorded from two identical units (units C and D) as shown in experiment panel D. Spectro-temporal transients absent in all methods' STRF estimates, presumably due the absence of higher-order correlations in the DMR stimuli and consistent with the explanation of panel D results.

Mentions: Figure 7A shows a 1 s segment of an FM complex stimulus spectrogram. The block length is 0.1 s and each block contains four sweeps with randomly chosen starting and ending frequencies. The stimulus distribution is shown in Figure 7B. Stimulus examples were sampled from the stimulus spectrogram by recasting spectro-temporal patches preceding the response in a 40 ms time window as vectors. Thus, the statistics of the stimulus ensemble is well approximated by the distribution of samples in each frequency channel, which is clearly non-Gaussian in this case (mean skewness ). As indicated in Figure 7C, second-order correlations in the stimulus ensemble are most pronounced in temporal direction spanning the whole patch size. This is a result of the high temporal resolution (2 ms) of the filter bank corresponding to the bin width of the spike trains. All units had a best frequency below 6 kHz. Therefore, we restricted the analysis to the range 0.5 kHz to 8 kHz resulting in 900-dimensional stimulus vectors.


Discriminative learning of receptive fields from responses to non-Gaussian stimulus ensembles.

Meyer AF, Diepenbrock JP, Happel MF, Ohl FW, Anemüller J - PLoS ONE (2014)

STRF estimation from gerbil inferior colliculus (IC) responses to frequency-modulated (FM) sweep complex stimuli.(A) Example segment of block-design FM tone complex with length 1 s. Amplitude scaling in decibel (dB), dynamic range limited to 25 dB below maximum for visualization. (B) Stimulus amplitude histogram, shown for each spectral band after centering; red (blue) indicate high (low) probability, respectively. (C) Normalized spectro-temporal auto-correlation function of stimulus ensemble. (D) STRFs estimated from recorded responses of four gerbil IC units (columns) with four inference methods (rows, explanation cf. Table 1). All units had best frequency below 8 kHz and analysis was restricted to the range 0.5 kHz to 8 kHz. The spike waveform density function of each unit is shown on top of each column, verifying single-unit activity [66]. Spectro-temporally transient ("diagonal'') patterns that are exhibited in the ridge regression-based estimates (top row) lack confirmation in the MID-, GLM-, and CbRF-derived STRF estimates (lower three rows). Thus, we hypothesize that these are an artefactual result originating from higher-order correlations and distribution asymmetries within the stimulus ensemble which the ridge regression method is not robust to. In general, MID, GLM, and CbRF produce very similar STRF estimates, with the latter two methods revealing a slightly finer tuning in some cases. (E) Validation experiment with dynamic moving ripple (DMR) stimuli responses recorded from two identical units (units C and D) as shown in experiment panel D. Spectro-temporal transients absent in all methods' STRF estimates, presumably due the absence of higher-order correlations in the DMR stimuli and consistent with the explanation of panel D results.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3974709&req=5

pone-0093062-g007: STRF estimation from gerbil inferior colliculus (IC) responses to frequency-modulated (FM) sweep complex stimuli.(A) Example segment of block-design FM tone complex with length 1 s. Amplitude scaling in decibel (dB), dynamic range limited to 25 dB below maximum for visualization. (B) Stimulus amplitude histogram, shown for each spectral band after centering; red (blue) indicate high (low) probability, respectively. (C) Normalized spectro-temporal auto-correlation function of stimulus ensemble. (D) STRFs estimated from recorded responses of four gerbil IC units (columns) with four inference methods (rows, explanation cf. Table 1). All units had best frequency below 8 kHz and analysis was restricted to the range 0.5 kHz to 8 kHz. The spike waveform density function of each unit is shown on top of each column, verifying single-unit activity [66]. Spectro-temporally transient ("diagonal'') patterns that are exhibited in the ridge regression-based estimates (top row) lack confirmation in the MID-, GLM-, and CbRF-derived STRF estimates (lower three rows). Thus, we hypothesize that these are an artefactual result originating from higher-order correlations and distribution asymmetries within the stimulus ensemble which the ridge regression method is not robust to. In general, MID, GLM, and CbRF produce very similar STRF estimates, with the latter two methods revealing a slightly finer tuning in some cases. (E) Validation experiment with dynamic moving ripple (DMR) stimuli responses recorded from two identical units (units C and D) as shown in experiment panel D. Spectro-temporal transients absent in all methods' STRF estimates, presumably due the absence of higher-order correlations in the DMR stimuli and consistent with the explanation of panel D results.
Mentions: Figure 7A shows a 1 s segment of an FM complex stimulus spectrogram. The block length is 0.1 s and each block contains four sweeps with randomly chosen starting and ending frequencies. The stimulus distribution is shown in Figure 7B. Stimulus examples were sampled from the stimulus spectrogram by recasting spectro-temporal patches preceding the response in a 40 ms time window as vectors. Thus, the statistics of the stimulus ensemble is well approximated by the distribution of samples in each frequency channel, which is clearly non-Gaussian in this case (mean skewness ). As indicated in Figure 7C, second-order correlations in the stimulus ensemble are most pronounced in temporal direction spanning the whole patch size. This is a result of the high temporal resolution (2 ms) of the filter bank corresponding to the bin width of the spike trains. All units had a best frequency below 6 kHz. Therefore, we restricted the analysis to the range 0.5 kHz to 8 kHz resulting in 900-dimensional stimulus vectors.

Bottom Line: Results demonstrate that the proposed approach successfully identifies correct underlying STRFs, even in cases where second-order methods based on the spike-triggered average (STA) do not.Applied to small data samples, the method is shown to converge on smaller amounts of experimental recordings and with lower estimation variance than the generalized linear model and recent information theoretic methods.Thus, CbRF estimation may prove useful for investigation of neuronal processes in response to natural stimuli and in settings where rapid adaptation is induced by experimental design.

View Article: PubMed Central - PubMed

Affiliation: Department of Medical Physics and Acoustics and Cluster of Excellence ''Hearing4all'', University of Oldenburg, Oldenburg, Germany.

ABSTRACT
Analysis of sensory neurons' processing characteristics requires simultaneous measurement of presented stimuli and concurrent spike responses. The functional transformation from high-dimensional stimulus space to the binary space of spike and non-spike responses is commonly described with linear-nonlinear models, whose linear filter component describes the neuron's receptive field. From a machine learning perspective, this corresponds to the binary classification problem of discriminating spike-eliciting from non-spike-eliciting stimulus examples. The classification-based receptive field (CbRF) estimation method proposed here adapts a linear large-margin classifier to optimally predict experimental stimulus-response data and subsequently interprets learned classifier weights as the neuron's receptive field filter. Computational learning theory provides a theoretical framework for learning from data and guarantees optimality in the sense that the risk of erroneously assigning a spike-eliciting stimulus example to the non-spike class (and vice versa) is minimized. Efficacy of the CbRF method is validated with simulations and for auditory spectro-temporal receptive field (STRF) estimation from experimental recordings in the auditory midbrain of Mongolian gerbils. Acoustic stimulation is performed with frequency-modulated tone complexes that mimic properties of natural stimuli, specifically non-Gaussian amplitude distribution and higher-order correlations. Results demonstrate that the proposed approach successfully identifies correct underlying STRFs, even in cases where second-order methods based on the spike-triggered average (STA) do not. Applied to small data samples, the method is shown to converge on smaller amounts of experimental recordings and with lower estimation variance than the generalized linear model and recent information theoretic methods. Thus, CbRF estimation may prove useful for investigation of neuronal processes in response to natural stimuli and in settings where rapid adaptation is induced by experimental design.

Show MeSH
Related in: MedlinePlus