Cosyne 2009 Workshops
March 2-3, 2009
Snow Bird, Utah
Workshop Title
Calculations for communication: neural coding of songs in the bird brain
Organizer(s)
Sarah Woolley, Columbia University
Abstract
The songbird serves as a model for understanding how natural sensory signals are encoded and decoded by neural circuits. Songbirds learn to produce and recognize the complex vocalizations that they use to communicate. The songbird neural circuits that code sound lend themselves to combined experimental and theoretical approaches. In this workshop, computational and systems neuroscientists will discuss the sensory tuning, neural discrimination, circuit dynamics and effects of experience on neuronal responses that result in song learning and perception.
Schedule
Morning session (8:30 – 11:00AM)
8:30 – 8:40 Introductory remarks
8:40 – 9:00 Sarah Woolley Auditory coding properties that predict dynamic and static receptive fields
9:10 – 9:30 Liam Paninski Statistical models for neural encoding, decoding, and optimal stimulus design
9:40 – 10:00 Break for coffee and discussion
10:00 - 10:20 Frederic Theunissen Coding of sound features in primary and secondary auditory areas
10:30 – 10:50 Tatyana Sharpee Advantages of multidimensional encoding of sounds
The morning session ends at 11:00AM.
Afternoon session (4:30 – 7:30PM)
4:30 – 4:50 Kamal Sen Neural discrimination of songs in field L
5:00 – 5:20 Tim Gentner Suppression and song recognition learning in NCM
5:30 - 6:00 Break for coffee and discussion
6:00 – 6:20 Richard Hahnloser How the songbird brain listens to its own songs
6:30 – 6:50 Richard Mooney Visualizing and manipulating neural mechanisms for song learning
7:00 – 7:30 General Discussion
Speakers and Talk Abstracts
1. Sarah Woolley, Columbia University, Auditory coding properties that predict dynamic and static receptive fields
The responses of songbird auditory midbrain neurons to songs and other complex sounds are well characterized by spectrotemporal receptive fields. Some midbrain neurons show changes in spectral and temporal tuning as measured from spectrotemporal receptive fields when processing complex sounds with different statistical properties. These cells have “dynamic” receptive fields. Other midbrain cells maintain the same receptive fields regardless of what sounds they process. These cells show “static” receptive fields. The neural mechanisms that subserve differences in spectrotemporal tuning during the processing of different complex sounds are unknown. I will discuss how dynamic and static spectrotemporal tuning can be predicted by other coding properties such as firing rate, responses to pure tones and temporal response patterns, and suggest neural mechanisms that may lead to dynamic versus static receptive fields.
2. Liam Paninski, Columbia University, Statistical models for neural encoding, decoding, and optimal stimulus design
We describe statistical model-based techniques that in some cases provide a unified solution to the neural encoding and decoding problems. These models can capture stimulus dependencies as well as spike history and interneuronal interaction effects in population spike trains, and are intimately related to biophysically-based models of integrate-and-fire type. We describe flexible, powerful likelihood-based methods for fitting these encoding models and then for using the models to perform optimal decoding. Each of these tasks turn out to be computationally tractable, due to a key concavity property of the model likelihood. Finally, we return to the encoding problem to describe how to use these models to adaptively optimize the stimuli presented to the cell on a trial-by-trial basis, in order that we may infer the optimal model parameters as efficiently as possible.
3. Frederic Theunissen, UC Berkeley, Coding of sound features in primary and secondary auditory areas
I will argue that, although decomposing sound into narrow frequency channels as it is done by the ear yields an efficient representation for transmission, such a decomposition is a poor representation for extracting and memorizing behaviorally relevant features from complex sounds. An auditory system interested in vocal recognition or more generally in mediating percepts such as timbre and pitch must therefore resynthesize acoustical features by cross-channel computations. I will illustrate this principle by showing the types of receptive fields that are found in the ascending auditory system of songbirds. I will show how these receptive fields code song features and could be involved in mediating distinct perceptual attributes. I will finish by speculating on how such specialized receptive fields could emerge during development.
4. Tatyana Sharpee, Salk Institute, Advantages of multidimensional encoding of sounds
Sounds arrive at our ears and are perceived by us as continuous time-varying signals. Yet, neurons that encode different aspect of sound waveform can only provide discrete representations of these signals because of finite spike duration and refractory periods. The distortions introduced by these discrete measurements limit the accuracy with which continuous signals can be reconstructed under the best circumstances. I will discuss different possible sampling strategies and how they depend on the structure of the noise in the auditory environment. In a natural auditory environment, noise values are strongly correlated in time. Under these conditions, multi-dimensional encoding of sounds based on sampling the values of sound log-amplitude and its time derivative are shown to be advantageous. These results agree with recent characterization of auditory processing in the field L region of auditory forebrain of songbirds.
5. Kamal Sen, Boston University, Neural discrimination of songs in field L
In order to correctly classify, discriminate and recognize stimuli,sensory systems must be able to compensate for natural variations in the stimuli. This ability to deal with variations in stimulus parameters such as intensity, timing and speed of presentation, is known as response invariance. Previously, we have shown that some neurons in field L (the avian analog of primary auditory cortex) in the zebra finch are capable of intensity invariant discrimination of songs. Here I will discuss the ability of field L neurons to cope with natural variations in timing in multiple renditions of songs, as well as synthetic variations in time-warped songs which are artificially slowed down or sped up by upto a factor of two relative to the normal song.
6. Tim Gentner, UCSD, Suppression and song recognition learning in NCM
Current understanding of experience dependent plasticity in sensory cortices turns on the notion that as signals acquire behavioral relevance, they elicit increasingly robust neural responses in single cells and/or drive increasingly large neural populations. Similar facilitative effects are seen in the songbird auditory forebrain region CMM when birds learn to recognize conspecific songs. Here, I describe a discrete population of neurons in the adjacent auditory region NCM where the opposite occurs. Following song recognition training, learned songs elicit significantly lower extracellular responses from single NCM neurons than do unfamiliar songs. Moreover, this effect is directly tied to associative learning, as control songs presented in the absence of behavioral contingencies also elicit robust responses. These findings are consistent with an active role for (1) broad scale response suppression and (2) contributions from irrelevant signals in shaping the learned sensory representation for natural, spectro-temporally complex, acoustic communication signals.
7. Richard Hahnloser, UZH / ETHZ, How the songbird brain listens to its own songs
We investigate auditory feedback processing in the auditory forebrain of zebra finches that are in a late developmental stage of song learning. Overall, we see similarity of spike responses during singing and during playback of the bird’s own song, with song responses commonly leading by a few milliseconds. However, brief time-locked acoustic perturbations of auditory feedback reveal complex sensitivity that cannot be predicted from passive playback responses. Some neurons that respond to playback perturbations do not respond to song perturbations, which is reminiscent of sensory-motor mirror neurons. By contrast, some neurons are highly feedback sensitive in that they respond vigorously to song perturbations, but not to unperturbed songs or perturbed playback. These findings suggest that a computational function of forebrain auditory areas may be to detect errors between actual feedback and mirrored feedback deriving from an internal model of the bird’s own song or that of its tutor.
8. Richard Mooney, Duke University, Visualizing and manipulating neural mechanisms for song learning
Songbirds learn to sing by using auditory feedback to evaluate vocal performance in reference to a memorized acoustic model, or song template. Part of this process could harness corollary discharge of song motor commands to generate predictions of auditory feedback associated with the resultant vocalization. Comparisons between these predictions and the real time feedback could then be used for error correction. First, I will describe properties of song sensorimotor neurons that appear suited to provide such predictive signals. Then I will discuss results of our efforts to identify and manipulate central representations of auditory feedback. Finally, I will present results of in vivo imaging experiments that examine how experience of the song model and early attempts at imitiation are reflected as structural changes in song sensorimotor neurons.
