INFORMS Journal on Applied Analytics (practice of operations research and management science and the impact this practice has on organizations throughout the world)
It is seldom to present the codes in a lecture/seminar, so my presentation should not be taken as a reference and the codes are used to help the students. Please refer to the following examples for your reference. You might include more detailed results/codes as backup slides in case the audients ask any related questions.
If you use certain material from the Internet without citation, it is treated as plagiarism (抄襲).
Problem statement, background knowledge, the math models to solve the problems, and the (potential) real-world applications or academic contributions
If you read a paper with Python codes, 3 steps to learn:
Reproduce the results, e.g., a table or picture, in the paper
Tune the hyper-parameters or parameters, e.g., the mutation rate of genetic algorithm, learning rate in machine learning algorithms
Comparison: Computational time, solution quality (performance) by using a baseline or an oracle
Rewrite some modules or functions, e.g., replace the crossover module with another approach
If operations research
Problem to be solved?
Mathematical model: Decision variables, objective function and its interpretation of each term, the constraints and their types/meanings
Software solvers and algorithm development: Python codeor Excel solution, What is the optimal/suboptimal policy? Intuitive or greedy policies, duality gap, interpretation, sensitivity analysis and interpretation
Conclusion, future works and research
If machine learning
Data exploration: Mention any special things in preprocessing, Important features (correlation coefficients, Lasso, or even better by using best subset or holistic approach, feature_importances_ from random forests), insights
Model structure (feedforward or the others, and node numbers or size the convolution filter if neural network), feature engineering, hyperparameters in the training algorithm (besides the default values), and the validation process (or cross-validation)
Please use at least ensemble trees and neural networks, the most winning methods in the Kaggle competitions.
We only teach the most fundamental way by using grid search. Please search "hyperparameter optimization algorithms" under Google Scholar for more advanced research results.
The improvement always exists. The amazing paper Stable Regression demonstrated that we could even improve the performance by using optimization-based splitting instead of random splitting.
Performance comparison: Different algorithms (with the optimal hyperparameters), training and testing performance indexes (ROC or R^2), insights and your observations
Conclusion, future works and research
Work assignment for each team member
Summary of your learning (小組心得)
Please indicate the assignment of the final report of the team members. For students who have not done anything, the final project will be given zero points for the written report and presentation. 請註明小組成員的期末報告工作分配，沒有做事的同學，期末專題將會零分(書面報告和上台報告)
If requested, the written report including the cover should not exceed 15 pages, each page is a single column, the font of the content is 12 points, 1.5 times for the line spacing, and the title is not limited in size. 書面報告含封面請勿超過15頁，每頁為單欄，內文為12級字，標楷體和 Times New Roman，1.5 倍行距，標題不限大小。
Please upload your pdf file to cycu ilearn one day before the last class meeting time.
沒有留言:
張貼留言