首页 » Python辅导 » Python辅导 | ECE57000 Course Project

Python辅导 | ECE57000 Course Project

Course Project

(Instructions below adapted from Prof. Jeffrey Siskind’s ECE57000 syllabus from last year.) Logistics and mechanisms for submitting various parts of the project will be finalized later in the semester.

 

The course project will be individual. No group projects.  However, I encourage you to discuss your papers and project with other students.  Explaining your papers or project to someone else can help test your own understanding.

 

Research Paper Selection (at least 3)

Students will be required to select and read at least three recent conference or journal research papers in the fields of AI, computer vision, natural language processing, or machine learning. In general, the papers should have been published within the last 3 years in one of the following venues: AAAI, IJCAI, CVPR, ICCV, ECCV, ACL, NAACL, EMNLP, NIPS/NeurIPS, ICML, ICLR, AISTATS, or JMLR. One paper can be older than 3 years but must have > 500 citations on Google scholar (if older, the citation count must be included in the “note” filed of the BibTeX entry.  Finally, given the implementation requirement below, one of the papers must have material that can be implemented.

 

You are welcome to read more than three papers. Your selection of papers must be approved by me. I may be willing to accept papers even if they do not fit the criteria above, but all paper selections, whether or not they meet the above criteria, must be approved by me. If you intend to select papers that are older than three years old or from a venue other than listed above, I suggest that you discuss this with me prior to the due date.

 

You will be required to submit a BibTeX entry for each paper including a URL for downloading a pdf of the paper.  For a journal paper, this should contain (at least) the paper title, paper abstract, authors, journal, volume, year, pages and pdf URL. For a conference paper, this should contain (at

least) the paper title, paper abstract, authors, conference, year, and pdf URL. Please note that you should not send me the paper itself, just the BibTeX entry.

BibTeX journal example:

@article{inouye2017review,  abstract = {The Poisson distribution has been widely studied … [truncated but you should include full abstract]},  author = {Inouye, D. I. and Yang, E. and Allen, G. I. and Ravikumar, P.},  journal = {Wiley Interdisciplinary Reviews: Computational Statistics},  number = {3},  pages = {e1398},

title = {A Review of Multivariate Distributions for Count Data Derived from the

Poisson Distribution},  volume = {9},  year = {2017},  note = {XXX citations},

url = {https://arxiv.org/pdf/1609.00066.pdf}

}

 

Implementation

You will be required to do at least one of the following:

  1. Reimplement paper idea: Implement and evaluate the ideas from at least one paper. If an implementation of the paper is already available (e.g. from the author’s website), you must state this in your report and compare your implementation to the existing implementation, both in terms of code and performance.
  2. Implement new idea (much harder, research-oriented): Propose, implement, and evaluate an extension or novel idea related to the papers you read. For this option, you can build off of any existing implementation but your implementation must extend or alter the original idea in a significant way.

 

The implementation must be nontrivial. A good guideline is that the implementation should be at least four pages of code. This is not a strict guideline. Ultimately, I will determine whether or not the implementation meets the non-triviality requirement. If you have questions about your implementation, please contact me to discuss.

 

The implementation must be in Python.  (UPDATE) You may use whatever libraries you would like such as PyTorch, TensorFlow, and Keras. However, I am only familiar with PyTorch and we will likely have at least one assignment in PyTorch so if all things are equal, I would suggest using PyTorch.  I encourage you to have your implementation conform to the scikit-learn estimator API

(see the APIs of scikitlearn objects, Rolling your own estimator and other scikit-learn documentation).  This would enable interfacing with other parts of scikit-learn including hyperparameter tuning or pipelining.  (UPDATE) Conforming to the scikit-learn estimator API is no longer required but is encouraged.

 

You must conduct a substantive evaluation of your implementation to determine how well it solves the intended problem. Ideally, you should replicate the experiments presented in the paper but I will not require this.

 

Term Paper

You will be required to write a six page paper in LaTeX meeting the typesetting conventions of ICML (see ICML 2019 author instructions for LaTeX template and instructions). Approximately three pages of this paper should be a substantive critique of the three (or more) papers that you have read. And approximately three pages of this paper should be a description of your implementation and evaluation of the material from one of the papers. (Note: You should not put your name on the term paper to accommodate the reviews. See “Reviews” section below.)

 

5 Minute Spotlight Video

You will also be required to create a 5 minute video presentation of your term paper. The video should be between 4-5 minutes but not longer than 5 minutes. This “spotlight” presentation should cover

  1. A brief summary and critique of the three papers that you have read
  2. A description of your implementation.
  3. A description and discussion of your evaluation.
  4. (Optional) Any concluding thoughts or future directions.

You should only have between 5-10 slides to fit within the 5 minutes.  One minute per slide is usually reasonable.

 

You can see real examples of video abstracts from top machine learning conference at https://nips.cc/Conferences/2018/Schedule?type=Poster (look for “3-min video” links). I do not expect your video to be at the quality level of these videos, but it hopefully gives you some ideas.  Note your video will be 5 minutes instead of 3 minutes so you will have a little more time than the examples above. The video can be in any format you want including animations, slides with narration, video of you presenting, or combination of the above. However, the focus should be on clarity rather than fanciness.

 

Project Presentation

Given the large number of students in the class, we will randomly select students to present during the presentation class periods (see class schedule).  The first presentation class period will be Nov 18 so all students should be ready to present by Nov 18.  We will announce 1-2 hours before class which students will be presenting.  The format will be 5 minutes (like the video) + 2 minutes for questions while the next presenter sets up.

 

The presentation must be given from a laptop or the computer that is built-in to the lectern in the lecture hall. You can use whatever tools you wish to prepare your presentation (i.e. LaTeX/beamer or PowerPoint). If you are scheduled to give your presentation, you should arrive in class early to make sure that your presentation setup works and that you are prepared to give your presentation in the allotted time slot.

 

Reviews

The term papers will be reviewed by other students in the class. Like all conferences, this process will be double blind: reviewers will not know the identity of authors and vice versa. To support this, like all conferences, you should NOT put your name on the term paper submission or on your reviews. In place of your name, you should put your PUID. Also, like conferences, reviews will be confidential. The only person who will be privy to the reviews will be the reviewer, the instructor, and the author.

 

I will assign each term paper to five students to read. Each student will be required to read five student term papers (other than their own) and prepare conference-style reviews, primarily indicating clarity and the quality of the implementation effort. The exact format for the review will be determined later in the semester.


程序辅导定制C/C++/JAVA/安卓/PYTHON/留学生/PHP/APP开发/MATLAB


本网站支持 Alipay WeChatPay PayPal等支付方式

E-mail: vipdue@outlook.com  微信号:vipnxx


如果您使用手机请先保存二维码,微信识别。如果用电脑,直接掏出手机果断扫描。

blank

发表评论

您的电子邮箱地址不会被公开。