CS412 PROJECT SOLVED

CS412 PROJECT 
Project Description
The course project is to give the students hands-on experience on applying what you have learned in the class to
the problems of your interest. You may choose one of the default course projects, we will provide you with
the corresponding dataset or links.
Default Projects
Image
Imbalanced Fish Image Classification
Fine-Grained Classification for Salmon Images
Text
Fake News detection
Author Identification
Grade Policy (Report + Presentation)
Final Report (15%)
We will provide a template, which is like a formal conference paper. Each team needs to submit a report for your
project. The instructors and TAs will evaluate the report from the three perspectives: methodology (10%), results
(e.g., test accuracy, 10%), writing (10%). The report is due on 27 May 2021, no late day is allowed.
Project Milestone (5%)
A 2-5 pages report for the progress and preliminary results of your project
Presentation (10%)
A 10-15 minutes short presentation for each team. We may invite other faculties or students to attend the
presentation.
Notice
The trained models and source code should be submitted as well.
Team size
Students may do final projects solo or in teams of up to 5 people. Teamwork is encouraged. Larger teams are
expected to do larger projects.
Contribution: In the final report we ask for a statement of what each team member contributed to the project.
Team members will typically get the same grade, but we may differentiate in extreme cases of unequal
contribution. You can contact us in confidence in the event of unequal contribution.
Resources
 You can use any deep learning/machine learning framework you like (PyTorch, TensorFlow,
Chainer, MXNet, etc.)
 You may use any existing code, libraries, and refer to any papers, books, online references, etc. for
your project. However, you must cite your sources in your report and indicate which parts of the
project are your contribution and which parts were implemented by others.
 Do not look at another CS412 group’s code. Plagiarism is NOT allowed, and you will be graded as 0
for this project part.
CS412 PROJECT SPRING, 2021, ZJUI
Imbalanced Fish Image Classification
Project Introduction
In this project, we aim at designing an image classifier on the Fish2016 dataset. There are 42 classes in the
dataset. The number of each class has a large variant. You need to come up with some algorithms that can train
a robust classifier on the imbalanced data.
Data Preparation
1. Download data from https://zjuintlmy.sharepoint.com/:u:/g/personal/gaoangwang_intl_zju_edu_cn/Ecj6wLTmfLlPhUH5oV7RikBcX9x-z4DQZrCEPPTiimeKw?e=nSMblp.
(Do NOT spread the data. This may contain a license issue.)
2. Split the dataset into training and validation. Specifically, the first 3/4 images of each class are used
in the training. The rest are used as validation.
Classifier
1. Pytorch is recommended for this project. You can use pre-trained models from
https://pytorch.org/vision/0.8/models.html.
2. Try to come up with some ideas for imbalanced data classification. For example, you are free to
design some loss function.
Evaluation
the mAP is the evaluation metrics. Report the overall mAP and AP of each class on the validation set. You can
check https://scikit-learn.org/stable/modules/generated/sklearn.metrics.average_precision_score.html for
computing the score. Compare with the baseline method that does not include any imbalanced data processing
strategy.
Some Image Examples
CS412 PROJECT SPRING, 2021, ZJUI
Fine-Grained Classification for Salmon Images
Project Introduction
In this project, we aim at designing an image classifier on the Salmon images. Different salmon species look
very similar. You are required to design a classifier to distinguish different salmon species.
Data Preparation
1. Download the data from https://zjuintlmy.sharepoint.com/:u:/g/personal/gaoangwang_intl_zju_edu_cn/EdT_p6M4LjxJtLYe1pqUKhIBOK
CkaL0QSrHhZdbD46_dyA?e=SdS21J.
(Do NOT spread the data. This may contain a license issue.)
2. Split the dataset into training and validation. Specifically, the first 3/4 images of each class are used
in the training. The rest are used as validation.
Classifier
1. Pytorch is recommended for this project. You can use pre-trained models from
https://pytorch.org/vision/0.8/models.html.
2. Try to come up with some ideas for fine-grained classification.
Evaluation
the mAP is the evaluation metrics. Report the overall mAP and AP of each class on the validation set. You can
check https://scikit-learn.org/stable/modules/generated/sklearn.metrics.average_precision_score.html for
computing the score. Compare with the baseline method that does not look at fine-grained information.
Some Image Examples
CS412 PROJECT SPRING, 2021, ZJUI
Fake News detection
Project Introduction
In this project, we aim at developing a machine learning program to identify when an article might be fake news.
Please see the details with the following link: https://www.kaggle.com/c/fake-news/overview
Data Preparation
1. Download data from the Blackboard (Fake_News_data.zip)
2. train.csv: A full training dataset with the following attributes:
 id: unique id for a news article
 title: the title of a news article
 author: author of the news article
 text: the text of the article; could be incomplete
 label: a label that marks the article as potentially unreliable
 1: unreliable
 0: reliable
3. test.csv: A testing training dataset with all the same attributes at train.csv without the label.
Evaluation
The evaluation metric for this competition is accuracy, a very straightforward metric.
accuracy = correct predictions
correct predictions + incorrect predictions
Accuracy measures false positives and false negatives equally, and really should only be used in simple cases
and when classes are of generally equal class size.
Notice
1. Pytorch or Tensorflow is recommended for this project.
2. Though a high accuracy or good performance is desired, we are looking forward to more creative
ideas and your problem-solving skills.
3. You may refer to available solutions from the internet, but do not copy all the codes from them. Try
to propose your solution and do cite the codes and references if your implementation is based on
them.
4. Besides solving the tasks in these projects, I do recommend you guys to identify the current
challenges in these tasks and the limitations of your methods or current state of the arts.
CS412 PROJECT SPRING, 2021, ZJUI
Author Identification
Project Introduction
In this project, you are challenged to predict the author of excerpts from horror stories by Edgar Allan Poe, Mary
Shelley, and HP Lovecraft.
Please see the details with the following link: https://www.kaggle.com/c/spooky-author-identification/overview
Data Preparation
The dataset contains text from works of fiction written by spooky authors of the public domain: Edgar Allan Poe,
HP Lovecraft, and Mary Shelley. Your objective is to accurately identify the author of the sentences in the test
set. You can download data from the Blackboard (Author_Identification_data.zip)
File descriptions
 train.csv - the training set
 test.csv - the test set
Data fields
 id - a unique identifier for each sentence
 text - some text written by one of the authors
 author - the author of the sentence (EAP: Edgar Allan Poe, HPL: HP Lovecraft; MWS: Mary
Wollstonecraft Shelley)
Evaluation
Submissions are evaluated using multi-class logarithmic loss. Each id has one true class. For each id, you must
submit a predicted probability for each author. The formula is then:
𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 = − 1
𝑁𝑁��𝑦𝑦𝑖𝑖𝑖𝑖 log�𝑝𝑝𝑖𝑖𝑖𝑖�
𝑀𝑀
𝑗𝑗=1
𝑁𝑁
𝑖𝑖=1
where N is the number of observations in the test set, M is the number of class labels, (log) is the natural logarithm,
(𝑦𝑦𝑖𝑖𝑖𝑖) is 1 if observation (i) belongs to a class (j) and 0 otherwise, and (𝑝𝑝𝑖𝑖𝑖𝑖) is the predicted probability that
observation (i) belongs to a class (j).
The submitted probabilities for given sentences are not required to sum to one because they are rescaled before
being scored. Each row is divided by the row sum. To avoid the extremes of the log function, predicted
probabilities are replaced with 𝑚𝑚𝑚𝑚𝑚𝑚(𝑚𝑚𝑚𝑚𝑚𝑚(𝑝𝑝, 1 − 10−15), 10−15).
Notice
1. Pytorch or Tensorflow is recommended for this project.
2. Though a high accuracy or good performance is desired, we are looking forward to more creative
ideas and your problem-solving skills.
3. You may refer to available solutions from the internet, but do not copy all the codes from them. Try
to propose your solution and do cite the codes and references if your implementation is based on
them.
4. Besides solving the tasks in these projects, I do recommend you guys to identify the current
challenges in these tasks and the limitations of your methods or current state of the arts.