Dynamic plasticity in coupled avian midbrain maps

Gurinder S. AtwalPhysical 2004 · DOI

TL;DR

This paper models how the barn owl's auditory map adapts to visual input using information theory and statistical mechanics. It reveals that the rate of visual-auditory misalignment is crucial for learning, with gradual shifts enabling adaptation in adult owls where sudden shifts fail.

What Problem Does This Paper Address?

Animals, like the barn owl, build internal maps of their environment by integrating sensory information. In barn owls, the auditory map (how they "hear" space) is initially calibrated by the visual map (how they "see" space). This calibration is a form of neural plasticity. While young owls can adapt to significant visual shifts, adult owls show much less plasticity. This paper aims to understand the underlying dynamics of this adaptive process, specifically how the auditory receptive field (RF) changes in response to visual input, and why adult owls are less adaptable than juveniles. They explore how the degree and speed of visual-auditory misalignment affect this learning.

Methods

The authors employ an information-theoretic approach, modeling the neural system's goal as maximizing the mutual information between sensory input and neural output, subject to biophysical constraints.

  • Information Theory: They use the concept of mutual information ($I[O;\{s\}]$) to quantify how much information the neural output ($O$) carries about the sensory inputs ($\{s\}$).
  • Statistical Mechanics: The "replica trick" is used to handle the complex entropy calculations involved in continuous variables.
  • Biophysical Constraints: To make the model biologically plausible, they introduce constraints on the neural receptive fields (RFs):
  • Gain Control: Limits the overall strength of the RF.
  • Temporal Dynamics: Penalizes rapid changes in the RF, reflecting the slow process of axonal growth and synapse formation.
  • Spatial Extent: Penalizes RFs that are too spread out, reflecting the localized nature of neural connections.
  • Model: They represent the neural output as a linear sum of weighted, filtered sensory inputs plus noise. The core of the model is deriving the differential equation governing the evolution of the RFs under these constraints.

Datasets

Not discussed in the paper. The study is theoretical and uses simulated data based on experimental observations of barn owls.

Key Insights and Ideas

  • Information Maximization as a Driving Principle: The paper frames neural learning as an optimization problem: maximizing information transfer under realistic biological limitations. This provides a powerful theoretical lens.
  • Two Types of Adaptation: For young owls (high plasticity), the model predicts two distinct adaptation patterns depending on the magnitude of visual-auditory misalignment: a smooth shift for small misalignments and a "decay-and-grow" mechanism for large misalignments.
  • The Critical Role of Misalignment Speed: The most striking finding is that for adult owls (low plasticity), the rate at which the visual and auditory maps become misaligned is paramount. Gradual, incremental shifts allow for adaptation, whereas a sudden, large shift leads to failure. This aligns with recent experimental findings.
  • Defining a "Critical Speed": The authors derive a characteristic speed ($v_c$) that demarcates the boundary between adaptive and non-adaptive learning regimes in the time-dependent misalignment scenario.

Key Results

  • Young Owls:
    • Small misalignment ($c=l_A$): The auditory RF smoothly shifts to align with the visual RF (Fig. 1).
  • Large misalignment ($c=2l_A$): The original auditory RF decays, and a new one emerges at the learned location, without the peak traversing the intermediate space (Fig. 2).
  • Adult Owls:
    • Large instantaneous misalignment ($c=2l_A$): Very little adaptation occurs; the RF magnitude decays (Fig. 3a).
  • Time-dependent misalignment ($c=vt$): Even with a large initial misalignment, gradual shifts allow for some, albeit slow, adaptation (Fig. 3b). This is a key result supporting incremental learning.
  • Critical Speed: A characteristic speed ($v_c$) is identified, below which learning occurs and above which it becomes non-adaptive.

Limitations and Future Work

  • Linear RF Approximation: The model uses a linear approximation for neural receptive fields, which is justified at low signal-to-noise ratios (SNR) but may not hold for all conditions.
  • Simplified Noise Model: The noise is assumed to be Gaussian and independent.
  • Simplified Dynamics: The model focuses on the long-time scale dynamics of RF adaptation, abstracting away faster neural processes.
  • Future Work: The authors suggest that further investigation into the role of the characteristic speed ($v_c$) and its relation to physiological parameters could be fruitful. Exploring more complex RF models and noise structures could also enhance the model's realism. The paper also implicitly points towards the need for experimental validation of the derived critical speed.

Summary updated

Abstract

Internal mapping of the external environment is carried out using the receptive fields of topographic neurons in the brain, and in a normal barn owl the aural and visual subcortical maps are aligned from early experiences. However, instantaneous misalignment of the aural and visual stimuli has been observed to result in adaptive behavior, manifested by functional and anatomical changes of the auditory processing system. Using methods of information theory and statistical mechanics a model of the adaptive dynamics of the aural receptive field is presented and analyzed. The dynamics is determined by maximizing the mutual information between the neural output and the weighted sensory neural inputs, admixed with noise, subject to biophysical constraints. The reduced costs of neural rewiring, as in the case of young barn owls, reveal two qualitatively different types of receptive field adaptation depending on the magnitude of the audiovisual misalignment. By letting the misalignment increase with time, it is shown that the ability to adapt can be increased even when neural rewiring costs are high, in agreement with recent experimental reports of the increased plasticity of the auditory space map in adult barn owls due to incremental learning. Finally, a critical speed of misalignment is identified, demarcating the crossover from adaptive to nonadaptive behavior.
BibTeX
@inproceedings{atwal2004dynamic,
  author = {Gurinder S. Atwal},
  title = {{Dynamic plasticity in coupled avian midbrain maps}},
  booktitle = {Physical},
  year = {2004},
  doi = {10.1103/physreve.70.061904},
}

Loading related content...

Figure