cse 5559 visualization meets aiweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfit also entails...

18
CSE 5559 Visualization Meets AI Spring 2020 Han-Wei Shen http://www.cse.ohio-state.edu/~hwshen/5559

Upload: others

Post on 06-Jul-2020

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

CSE 5559 Visualization Meets AI

Spring 2020 Han-Wei Shen

http://www.cse.ohio-state.edu/~hwshen/5559

Page 2: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

What is this course about?

• A seminar course • A course with different speakers at each class • A small, discussion based class• Students do assigned reading and grapple/discuss aloud the ideas they have

read • Learning activities

• Mini-lectures • Paper reading and summaries • Presentations• Projects

• State-of-the-art in certain topics

Page 3: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

What is this course about?

• Visualization Meets AI• AI: Machine Learning and Deep Learning

• ML and DL algorithms (e.g. Decision Trees, Random Forest, CNN, RNN, LSTM, GAN, AEs)• Visualization: analytics of data through pictures, animation, and user

interaction • Analytics? Discovery, interpretation, and communication of meaningful patterns in data.

It also entails applying data patterns towards effective decision making. • Principle of Data Visualization (Visual encoding, design, layout, and data representations

and reductions )• Visualization Meets AI?

• Visualization for AI: Explain ML and DL methods using visualization • AI for Visualization: Using ML and DL to facilitate Data Analytics

Page 4: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

What is this course NOT about?

• Detailed explanation of ML/DL algorithms • Mini-lectures on deep learning is provided

• Software tools for ML/DL such as scikit-learn, tensorflow, pytorch etc.• Prior-knowledge or self-education

• Detailed explanation of visualization algorithms • Mini-lectures on principles of visualization• My CSE 5544 website: http://www.cse.ohio-state.edu/~shen.94/5544

• Software tools of visualization such as D3, Matplotlib, etc• Resources on D3/Python provided on class website

Page 5: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Why am I taking this course?

• Give you a flavor of what is research• Self-teach a topic that you always want to learn• Discuss with your classmates and learn• Survey a cutting-edge research topic(s)• Force yourself to try something out (tensorflow, pytorch, D3,

matplotlib) • Identify your future research directions • ________ (fill the blank)

Page 6: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Structure of this course

• Mini-lectures (Every Friday of Week 1-5) • Deep Learning (Guest Instructor: Prof. Herman Shen)

• Introduction to Deep Learning, Why deep?• Deep learning II: Theory of ANN training, Backpropagation and Stochastic Gradient

Descent.• Theory behind Convolutional Network (CNN)• Deep learning IV: Generative Adversarial Network (GAN), WGAN, CGAN, and crCGAN• Deep learning V: Other state-of-the-art networks and future directions

Page 7: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Structure of this course

• Mini-lectures (Every Wednesday of Week 1-5) • Visualization

• Principle of Visualization : Channels for visual encoding • Data and Task Abstraction• Multivariate Visualization and Dimensionality Reduction• common visualization algorithms

Page 8: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Structure of this course • Teamwork (Starting from Week 2)

• 3-4 Students each team • Each team decides a common topic of interest in AI Meets VIS• Each team identifies 12-15 papers to read (List due 1/15)• Each team member picks 3 papers to read • Each team member creates three paper summaries, in the form of power point slides for

a 20 mins presentation/paper (Due 1/24, 2/7, 2/21) • Each team gives two presentations, 40 mins each, of selected topics (between Week 6

and 13) • Each team will discuss the presentation in each class and submit answers to the

discussion questions prepared by the presenting team (Due the following class )• Each team will do a final project together (Due 4/20)

• Between Week 6-13: Each class will have a 40 mins presentation and a 15 minutes discussion

Page 9: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Class Schedule

Page 10: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Class Schedule

Page 11: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

How to read a paper? Topic of interest: Vis for AI or AI for Vis

• What is the research question? • What has been done previously? • What is the authors’ solution to the research question?• What is the main method proposed/used by the authors? What are

the key characteristics and novelty of the proposed method?• What data does the paper use?• How is the proposed method(s) evaluated?

Page 12: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Reference Books (Deep Learning)

Page 13: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Reference Books (Deep Learning)

Page 14: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Paper List

• Http://www.cse.ohio-state.edu/~hwshen/5559• Your contributions are welcome !!

Page 15: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Paper List

• Http://www.cse.ohio-state.edu/~hwshen/5559• Your contributions are welcome !!

Page 16: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Paper List

• Http://www.cse.ohio-state.edu/~hwshen/5559• Your contributions are welcome !!

Page 17: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Blog Articles

• General reading for understanding a subject. Research papers are always harder to read becausethey are more advanced and typically lack sufficient background information

• https://docs.google.com/spreadsheets/d/19vLBLrt3DY_RG9nnwp808-SBIL5npj-ufV9KmcUQdhI/edit#gid=0

• This links contain 100+ blog articles, selected from my own collection in the past year.

Page 18: CSE 5559 Visualization Meets AIweb.cse.ohio-state.edu/~shen.94/5559/intro.pdfIt also entails applying data patterns towards effective decision making. •Principle of Data Visualization

Grading • Paper summary (individual): 30%

• Every individual student will submit three paper summaries in the form of power point slides as if you were to present the papers, each for 20 mins. The papers are chosen from the list of papers jointly decided by the group. Students in each group should choose different papers.

• Presentation (group): 20%• Each group will present 1 or 2 times on state-of-the art (i.e. research papers) of selected

topics. Each presentation will be 40 minutes.• Paper discussion and answers (group): 20%

• Students will discuss the presented paper with other group members immediately after the presentation and answer questions designed by the presenters. Each group will submit the answers in the next class. Absent students will not receive credits for the submission.

• Final project (group): 30% • Each group chooses a topic related to Vis for AI or AI for Vis.