Explorations of Sensorimotor Learning and Individuality in Mathematical Models of Behavioral Data Público

Abstract

One of the most obvious properties of a living organism is its highly coordinated behavioral patterns. These patterns reflect the orderly activities of the underlying nervous system that controls the behavior, and also may contain information about this organism's individuality. This thesis is devoted to the studies of behavior in diverse animal systems, and includes two distinct themes: computational rules of sensorimotor control and species classification. First, we try to build a full Bayesian theory of sensorimotor learning by investigating the singing behavior of songbirds and its response to imposed perturbations. Unlike previous models, our normative Bayesian filter model with multiple temporal scales and non-Gaussian sensory feedback can describe the dynamics of the entire probability distribution of the pitch of the song produced by the animals. We will show that our model not only accounts for salient features of the adaptation dynamics, such as the non-monotonic dependence of the behavioral compensation on the perturbation size, but also makes predictions about the behavioral patterns in the classical three-phase perturbation experiments, about the effects of pharmacological silencing of certain nuclei in the bird brain, and about the neural dynamics that is required to implement the model. After this, we move from a single behavior to the whole behavioral repertoire, which is automatically extracted from naturally moving flies. We can show that this behavioral repertoire contains information about the individuality of young flies, and that it can be used to classify fly species in high accuracy. This result suggests that animal behavior patterns might be used to study evolution without the need of any genetic information.

Table of Contents

1 Introduction................................ 1

2 Sensorimotor learning as a 1d Bayesian filter. . . . . . . 5

2.1 Introduction ................................ 5

2.2 Experimental data............................. 8

2.3 Mathematical model............................ 10

2.3.1 1d Bayesianfilter.......................... 10

2.3.2 Non-Gaussianstatistics...................... 14

2.3.3 Fits and predictions........................ 15

2.4 Discussion.................................. 18

2.5 Materials and methods .......................... 21

2.5.1 Experiments ............................ 21

2.5.2 Stable distributions ........................ 22

2.5.3 Fitting................................ 23

2.5.4 Choice of the shape of distributions . . . . . . . . . . . . . . . 25

2.5.5 Linear dependence on pitch shift in a Kalman filter with multiple timescales ............................. 26

3 Sensorimotor learning as a 2d Bayesian filter. . . . . . . 30

3.1 Introduction ................................ 30

3.2 Mathematical model............................ 33

3.2.1 2d Bayesian filter.......................... 33

3.2.2 Fits and predictions........................ 39

3.2.3 Biological interpretations ..................... 44

3.3 Discussion.................................. 48

3.4 Materials and methods .......................... 51

3.4.1 Experimental data......................... 51

3.4.2 Data bootstrapping ........................ 53

3.4.3 Simplification of representations ................. 53

3.4.4 Powerlaw-like distributions.................... 54

3.4.5 Model fitting and parameters................... 54

3.5 Additional figures ............................. 55

4 Behavioral map of a fly predicts species identity . . . . 68

4.1 Introduction ................................ 68

4.2 Behavior maps of fly............................ 68

4.3 Embedding by t-SNE ........................... 70

4.4 Classification by logistic regression.................... 71

4.5 Discussion.................................. 72

5 Conclusion ................................. 73

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 

About this Dissertation

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Laney Graduate School
Department	Physics
Degree	Ph.D.
Submission	Dissertation
Language	English
Research Field	Biophysics, General Physics, Theory
Palavra-chave	Animal behavior Vocal control Bayesian filter Behavioral map Sensorimotor learning
Committee Chair / Thesis Advisor	Nemenman, Ilya, Emory University
Committee Members	Sober, Sam, Emory University Boettcher, Stefan, Emory University Kim, Minsu, Emory University Berman, Gordon, Emory University

Última modificação

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Explorations of Sensorimotor Learning and Individuality in Mathematical Models of Behavioral Data ()	2019-04-05 18:26:22 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions