Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data Augmentation for Automated Adaptive Rodent Training

Dibyendu Das, Alfredo Fontanini, Joshua F. Kogan, Haibin Ling, C. R. Ramakrishnan, I. V. Ramakrishnan

TL;DR

This work built several artificial rodent models, which in turn would be used to build an efficient and automatic trainer and developed a novel similarity metric based on the action probability distribution to measure the behavioral resemblance of the models to that of real rodents.

Abstract

Fully optimized automation of behavioral training protocols for lab animals like rodents has long been a coveted goal for researchers. It is an otherwise labor-intensive and time-consuming process that demands close interaction between the animal and the researcher. In this work, we used a data-driven approach to optimize the way rodents are trained in labs. In pursuit of our goal, we looked at data augmentation, a technique that scales well in data-poor environments. Using data augmentation, we built several artificial rodent models, which in turn would be used to build an efficient and automatic trainer. Then we developed a novel similarity metric based on the action probability distribution to measure the behavioral resemblance of our models to that of real rodents.

Data Augmentation for Automated Adaptive Rodent Training

TL;DR

This work built several artificial rodent models, which in turn would be used to build an efficient and automatic trainer and developed a novel similarity metric based on the action probability distribution to measure the behavioral resemblance of the models to that of real rodents.

Abstract

Fully optimized automation of behavioral training protocols for lab animals like rodents has long been a coveted goal for researchers. It is an otherwise labor-intensive and time-consuming process that demands close interaction between the animal and the researcher. In this work, we used a data-driven approach to optimize the way rodents are trained in labs. In pursuit of our goal, we looked at data augmentation, a technique that scales well in data-poor environments. Using data augmentation, we built several artificial rodent models, which in turn would be used to build an efficient and automatic trainer. Then we developed a novel similarity metric based on the action probability distribution to measure the behavioral resemblance of our models to that of real rodents.

Paper Structure

This paper contains 12 sections, 11 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Training Setup
  3. Modelling the Behavior of Rodents using Reinforcement Learning
  4. Artificial Rodent Model
  5. States $\mathcal{S}^+$
  6. Actions $\mathcal{A}$
  7. Reward Model $\mathcal{R}$
  8. Similarity Metrics
  9. Similarity between individual behaviors
  10. Similarity between group behaviors
  11. Experimental Results
  12. Conclusions

Figures (3)

  • Figure 1: Left: sliding window mean accuracy of the artificial and the real rodents across all the trials in session 1 (Black dots represent the model). Right: Terminal accuracy of the artificial and the real rodents in all the sessions.
  • Figure 2: (a) Correlation heat-map of the similarity values (the lower the more similar) across multiple executions of the rodent model in each session. (b) Correlation heat-map of the similarity values between trajectories (sequence of trials) in each session. Data is obtained from five real rodents.
  • Figure 3: Correlation heat-map of the similarity values(lower values are more similar) between multiple sessions (average over each sliding window) for the rodent model.