( Print View )
[Texts & Links]
HW#5 --- last modified December 06 2021 13:25:20.
Due date: Dec 6
Files to be submitted:
Purpose: To gain experience with recurrent networks, network tuning, and neural network applications.
Related Course Outcomes:
The main course outcomes covered by this assignment are:
CLO4 -- Be able to select neural network layers type to build a network suitable for various learning tasks such as object classification, object detection, language processing, planning, policy selection, etc.
CLO5 -- Be able to select an appropriate regularization technique for a given learning task.
CLO6 -- Be able to code and train with a library such as Caffe, Theano, Tensorflow a multi-layer neural network.
CLO7 -- Be able to measure the performance of a model, determine if more data in needed, as well as how to tune the model.
For the last homework I want you to build a neural network to solve some application you are interested in. I am going to be flexible on what this is, but your final homework needs to satisfy the following requirements:
- You should write up in a file Hw5.pdf (included in Hw5.zip) a clearly stated problem that you hope to be a able to solve with your neural network.
- You should in this file have a diagram with your proposed Neural Net architecture.
- Your architecture must make use of some kind of recurrent layer such as an LSTM layer, or should be doing some kind of embedding, or involve an adversarial network.
- You must train using SGD with some kind of regularization technique.
- You should include both your training code, testing code, and code to use your trained model in the Hw5.zip.
- Your Hw5.pdf file should have careful instructions on how to use each program you include. You are allowed to use libraries of your choice, scikit,
etc.; however, if you use something we haven't described so far in class, you also need to write down installation instructions.
- Your goal is to get your program to solve the desired problem as accurately as possible, so you should conduct experiments to tune your architecture and hyperparameters to achieve this. You need to write up all such experiments (do this as you conduct them) in Hw5.pdf
- You should write up some conclusions in Hw5.pdf about your program which should include final accuracy achieved on never before seen data.
|Items 1-8 above each worth 1pt except 5 (grading on code quality, does it work, etc) and 7 (grading on how well written up your experiments are) which are worth 2pts