Limits...
Efficient temporal processing of naturalistic sounds.

Lesica NA, Grothe B - PLoS ONE (2008)

Bottom Line: We find that the onset of ambient noise evokes a change in receptive field dynamics that corresponds to a change from bandpass to lowpass temporal filtering.We show that these changes occur within a few hundred milliseconds of the onset of the noise and are evident across a range of overall stimulus intensities.Using a simple model, we illustrate how these changes in temporal processing exploit differences in the statistical properties of vocalizations and ambient noises to increase the information in the neural response in a manner consistent with the principles of efficient coding.

View Article: PubMed Central - PubMed

Affiliation: Department of Biology II, Ludwig-Maximilians-University Munich, Martinsried, Germany. lesica@zi.biologie.uni-muenchen.de

ABSTRACT
In this study, we investigate the ability of the mammalian auditory pathway to adapt its strategy for temporal processing under natural stimulus conditions. We derive temporal receptive fields from the responses of neurons in the inferior colliculus to vocalization stimuli with and without additional ambient noise. We find that the onset of ambient noise evokes a change in receptive field dynamics that corresponds to a change from bandpass to lowpass temporal filtering. We show that these changes occur within a few hundred milliseconds of the onset of the noise and are evident across a range of overall stimulus intensities. Using a simple model, we illustrate how these changes in temporal processing exploit differences in the statistical properties of vocalizations and ambient noises to increase the information in the neural response in a manner consistent with the principles of efficient coding.

Show MeSH
Context-dependent changes in temporal processing promote efficient coding.a) The power spectrum of the signal (thick lines) and noise (thin lines) in the responses of the model with the V (black) and VN (red) RFs to the vocalization stimulus alone for a typical cell (the same cell for which RFs and MTFs are shown in figures 3a and b). The gray band denotes the 20–120 Hz frequency range. The calculation of the signal and noise components of the responses is described in the Materials and Methods. b) The mutual information rate between the stimulus and the responses of the model with the V and VN RFs. The blue circle corresponds to the cell for which response spectra are shown in b. The crosses denote the sample mean±one standard deviation. c) A histogram showing the percent changes between the information rates shown in b (change from the responses of the VN model to the responses of the V model). d,e) Power spectra and information rates presented as in a and b for responses to the vocalization stimulus with ambient noise. f) A histogram showing the percent changes between the information rates shown in e (change from the responses of the V model to the responses of the VN model).
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC2249929&req=5

pone-0001655-g007: Context-dependent changes in temporal processing promote efficient coding.a) The power spectrum of the signal (thick lines) and noise (thin lines) in the responses of the model with the V (black) and VN (red) RFs to the vocalization stimulus alone for a typical cell (the same cell for which RFs and MTFs are shown in figures 3a and b). The gray band denotes the 20–120 Hz frequency range. The calculation of the signal and noise components of the responses is described in the Materials and Methods. b) The mutual information rate between the stimulus and the responses of the model with the V and VN RFs. The blue circle corresponds to the cell for which response spectra are shown in b. The crosses denote the sample mean±one standard deviation. c) A histogram showing the percent changes between the information rates shown in b (change from the responses of the VN model to the responses of the V model). d,e) Power spectra and information rates presented as in a and b for responses to the vocalization stimulus with ambient noise. f) A histogram showing the percent changes between the information rates shown in e (change from the responses of the V model to the responses of the VN model).

Mentions: To quantify the efficiency of these responses, we computed the mutual information between the stimulus and response using the direct method, as shown in figure 7b. Across the sample of cells, the information rates resulting from processing in the V RF are significantly higher than those resulting from processing in the VN RF (paired t-test, p<0.01, n = 23), with increases as large as 33.2% and an average increase of 10.4±9.4% (see histogram in figure 7c). This result indicates that, in the absence of ambient noise, redundancy reduction through bandpass filtering in the V RF is, indeed, the preferred strategy. When the vocalization stimulus is combined with ambient noise, the shapes of the signal and noise power resulting from processing in the V and VN RFs remain the same, but the overall SNR in the response decreases, as shown in figure 7d. Because of this decrease in SNR, the information rates resulting from processing in the VN RF are now significantly higher than those resulting from processing in the V RF (paired t-test, p<0.01, n = 23, see figure 7e), with increases as large as 39.6% and an average increase of 16.4±10.5% (see histogram in figure 7f). This result suggests that, under noisy conditions, increasing SNR through lowpass filtering in the VN RF is the preferred strategy. Thus, in processing vocalization stimuli with and without noise, the responses of the model have a higher information rate when the RF in the model is matched to the stimulus condition, suggesting that the context-dependent changes in temporal processing that we observe may serve to promote efficient coding by increasing the information in the neural response.


Efficient temporal processing of naturalistic sounds.

Lesica NA, Grothe B - PLoS ONE (2008)

Context-dependent changes in temporal processing promote efficient coding.a) The power spectrum of the signal (thick lines) and noise (thin lines) in the responses of the model with the V (black) and VN (red) RFs to the vocalization stimulus alone for a typical cell (the same cell for which RFs and MTFs are shown in figures 3a and b). The gray band denotes the 20–120 Hz frequency range. The calculation of the signal and noise components of the responses is described in the Materials and Methods. b) The mutual information rate between the stimulus and the responses of the model with the V and VN RFs. The blue circle corresponds to the cell for which response spectra are shown in b. The crosses denote the sample mean±one standard deviation. c) A histogram showing the percent changes between the information rates shown in b (change from the responses of the VN model to the responses of the V model). d,e) Power spectra and information rates presented as in a and b for responses to the vocalization stimulus with ambient noise. f) A histogram showing the percent changes between the information rates shown in e (change from the responses of the V model to the responses of the VN model).
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC2249929&req=5

pone-0001655-g007: Context-dependent changes in temporal processing promote efficient coding.a) The power spectrum of the signal (thick lines) and noise (thin lines) in the responses of the model with the V (black) and VN (red) RFs to the vocalization stimulus alone for a typical cell (the same cell for which RFs and MTFs are shown in figures 3a and b). The gray band denotes the 20–120 Hz frequency range. The calculation of the signal and noise components of the responses is described in the Materials and Methods. b) The mutual information rate between the stimulus and the responses of the model with the V and VN RFs. The blue circle corresponds to the cell for which response spectra are shown in b. The crosses denote the sample mean±one standard deviation. c) A histogram showing the percent changes between the information rates shown in b (change from the responses of the VN model to the responses of the V model). d,e) Power spectra and information rates presented as in a and b for responses to the vocalization stimulus with ambient noise. f) A histogram showing the percent changes between the information rates shown in e (change from the responses of the V model to the responses of the VN model).
Mentions: To quantify the efficiency of these responses, we computed the mutual information between the stimulus and response using the direct method, as shown in figure 7b. Across the sample of cells, the information rates resulting from processing in the V RF are significantly higher than those resulting from processing in the VN RF (paired t-test, p<0.01, n = 23), with increases as large as 33.2% and an average increase of 10.4±9.4% (see histogram in figure 7c). This result indicates that, in the absence of ambient noise, redundancy reduction through bandpass filtering in the V RF is, indeed, the preferred strategy. When the vocalization stimulus is combined with ambient noise, the shapes of the signal and noise power resulting from processing in the V and VN RFs remain the same, but the overall SNR in the response decreases, as shown in figure 7d. Because of this decrease in SNR, the information rates resulting from processing in the VN RF are now significantly higher than those resulting from processing in the V RF (paired t-test, p<0.01, n = 23, see figure 7e), with increases as large as 39.6% and an average increase of 16.4±10.5% (see histogram in figure 7f). This result suggests that, under noisy conditions, increasing SNR through lowpass filtering in the VN RF is the preferred strategy. Thus, in processing vocalization stimuli with and without noise, the responses of the model have a higher information rate when the RF in the model is matched to the stimulus condition, suggesting that the context-dependent changes in temporal processing that we observe may serve to promote efficient coding by increasing the information in the neural response.

Bottom Line: We find that the onset of ambient noise evokes a change in receptive field dynamics that corresponds to a change from bandpass to lowpass temporal filtering.We show that these changes occur within a few hundred milliseconds of the onset of the noise and are evident across a range of overall stimulus intensities.Using a simple model, we illustrate how these changes in temporal processing exploit differences in the statistical properties of vocalizations and ambient noises to increase the information in the neural response in a manner consistent with the principles of efficient coding.

View Article: PubMed Central - PubMed

Affiliation: Department of Biology II, Ludwig-Maximilians-University Munich, Martinsried, Germany. lesica@zi.biologie.uni-muenchen.de

ABSTRACT
In this study, we investigate the ability of the mammalian auditory pathway to adapt its strategy for temporal processing under natural stimulus conditions. We derive temporal receptive fields from the responses of neurons in the inferior colliculus to vocalization stimuli with and without additional ambient noise. We find that the onset of ambient noise evokes a change in receptive field dynamics that corresponds to a change from bandpass to lowpass temporal filtering. We show that these changes occur within a few hundred milliseconds of the onset of the noise and are evident across a range of overall stimulus intensities. Using a simple model, we illustrate how these changes in temporal processing exploit differences in the statistical properties of vocalizations and ambient noises to increase the information in the neural response in a manner consistent with the principles of efficient coding.

Show MeSH