Curious Word Learning

Welcome. This code representation is related to the Active Word Learning through Self-supervision study.

Citation

If you use this work please cite us as:

Gelderloos, L., Mahmoudi Kamelabad, A., & Alishahi, A. (2020). Active Word Learning through Self-supervision. In S. Denison, M. Mack, Y. Xu, & B. C. Armstrong (Eds.), Proceedings for the 42nd Annual Meeting of the Cognitive Science Society (pp. 1050–1056). Cognitive Science Society.

Alternatively you can download the bib file here as well:

Download .bib file

General Variables, Data, and constant definitions

dict_words_boxes

{img_id:(str): {
    obj_id:(str): {
        'bnbox': a list of four coordinations,
        'word':  the linguistic representation of the object
    },
},

Regular data (dictionary with all images and their object ids, corresponding words).

language_input

a tensor of word indexes.

main.py

input

The shape of input to main.py

< Number of Batches: int > < Learning Rate: float > < Curiosity Setting: str > < Seed >

• Number of Batches: int, (default= 40)
• Learning Rate: [.1, .01, .001, .0001, .00001] : float
• Curiosity Setting: [“curious”, “plasticity”, “sn”, “random”], (sn: Subjective Novelty)
• Seed: The ones used to run the experiment: [123, 234, 345, 456, 567, 678, 789, 890, 901, 12, 23, 34, 45, 56, 67, 78, 89, 90, 1, 100]

Code Structure considerations

Constants

In order to manage constants I made the class UniversalConstants for constants in toolbox.py. All the code files should import this class and make an object of it with specific name uc. Here is how you do it:

from toolbox import UniversalConstants
uc = UniversalConstants()

Bug Problems

report bugs either here on GitHub or contact Alireza personally.