Looking for Machine Learning Developer familiar with Dynet Neura

  • Posted:
  • Proposals: 12
  • Remote
  • #2587699
  • OPPORTUNITY
  • Open for Proposals
Baber S.Dominic T.Basile M.Laraib H.Ruyun L. + 7 others have already sent a proposal.
  • 12

Description

Experience Level: Expert
We are in need of ML developer to create custom models to be implemented into the Dynet Neural Network. These text-based models will be primarily written via C++ and Python. Example model types include RNN, XOR, Feed Forward Models, Text Categorization, Reinforcement models.

We want you to create the models or take the base of open source models to then edit / customize. The project will utilize these models and the NN to help create unique visual content in very intuitive and creative input methods
--------------------------------------------------------------------------------
Detailed:
The Dynet Neural Network is part of an application plugin we’ve developed that takes plain sentences and / or paragraphs, parses the phrase, sentence, or paragraph structure, and creates instructional command sets based off of the parsed data. Our goal with the NN portion of this is to:

● Create a model that retains the command sets that are generated, and looks for patterns from incoming parsed verbiage to call up previously used commands faster if applicable. RNN and Text Cat models are applicable here.

● Create a model that looks at unique or newly parsed verbiage, and can help create random, or custom strings of instructional command sets, based off of the existing command sets in its database. (i.e. There are command sets for “walk” and “jump,” but not for “skip.” The model should use either symbolic variables or create new parameters in assistance with the NLP dictionary to form a command set that uses walk and jump to create a skip command. This could be a combination of XOR, RNN, and Feed Forward, using the Text Cat model.

● Create a Reinforcement Model which observes the other two models and the NLP (if applicable) to build training weights that help the models growing data libraries become more efficient over time and with repeated use.

● (Optional) Create a Convolutional NN model, in association with a MNIST model which observes visual data and links visual data to NPL input, creating training weights that are visually linked to the information produced by the first model mentioned above.

Input formats:

Dynet accepts and exports text-based files (.txt), and can read .model files. We prefer all files to be written in C++.
-------------------------------------------------------------------------------------------------------------------
*Pls let us see your reasonable budget and timeline for this project.

New Proposal

Create an account now and send a proposal now to get this project.

Sign up

Clarification Board Ask a Question

    There are no clarification messages.