This document provides an overview of the Deep Learning for Computer Vision course for Fall 2021. It outlines that for the first three weeks, lectures will be online only due to COVID regulations, but starting week 4, lectures will be offered in a hybrid format with physical and online options. Students are encouraged to attend physical lectures biweekly. The course will cover fundamentals of machine learning and deep learning techniques for computer vision tasks. Students will complete assignments and a final project involving real-world visual data. Computing resources are required but not provided. The teaching team, schedule, grading policy, academic integrity expectations, and enrollment process are also summarized.
Teaching Continuity: Supporting staff teaching online when face-to-face class...Samantha Lee Pan
Have you ever been tasked with supporting staff in teaching online when face-to-face classes have been cancelled? In difficult circumstances, campus-based activities could be suspended indefinitely due to a significant environmental, health or socio-political impact. In these extreme cases, online teaching can provide a form of emergency management and continuity of teaching and learning. This type of support was needed during the campus shutdowns of 2016 and 2017 caused by student protests that affected South African higher education institutions nationwide. In this session, colleagues from the University of Cape Town (UCT) will share issues, tools and solutions provided to support academic staff required to teach online during those difficult times. The session is designed to provide an example case, but also learn from others. This session will take the form of a birds of a feather discussion, so we welcome others with similar experiences and institutional or individual stories to join.This session is based on the Sakai Virtual Conference 2017 presentation - Under pressure: Supporting staff teaching online in uncertain times (https://youtu.be/50m4skkITeo) but incorporates further questions to help find a solution in your context.
Teaching and Learning Support Activities at Osaka UniversityHaruo Takemura
This presentation describes activities of Teaching and Learning Support Center, Osaka University. These includes faculty development, pre-faculty development, IT systems for education etc.
TechLogic 2014 Keynote on Inverting an Algorithms Class (Extended Version)suthers
Discussion of the inversion of an Algorithms course: how it is motivated by learning theory; how the activities are organized; outcomes. This is an expanded version of an invited keynote talk for the "TechLogic" conference at the University of Hawaii at Manoa.
Show drafts
volume_up
Empowering the Data Analytics Ecosystem: A Laser Focus on Value
The data analytics ecosystem thrives when every component functions at its peak, unlocking the true potential of data. Here's a laser focus on key areas for an empowered ecosystem:
1. Democratize Access, Not Data:
Granular Access Controls: Provide users with self-service tools tailored to their specific needs, preventing data overload and misuse.
Data Catalogs: Implement robust data catalogs for easy discovery and understanding of available data sources.
2. Foster Collaboration with Clear Roles:
Data Mesh Architecture: Break down data silos by creating a distributed data ownership model with clear ownership and responsibilities.
Collaborative Workspaces: Utilize interactive platforms where data scientists, analysts, and domain experts can work seamlessly together.
3. Leverage Advanced Analytics Strategically:
AI-powered Automation: Automate repetitive tasks like data cleaning and feature engineering, freeing up data talent for higher-level analysis.
Right-Tool Selection: Strategically choose the most effective advanced analytics techniques (e.g., AI, ML) based on specific business problems.
4. Prioritize Data Quality with Automation:
Automated Data Validation: Implement automated data quality checks to identify and rectify errors at the source, minimizing downstream issues.
Data Lineage Tracking: Track the flow of data throughout the ecosystem, ensuring transparency and facilitating root cause analysis for errors.
5. Cultivate a Data-Driven Mindset:
Metrics-Driven Performance Management: Align KPIs and performance metrics with data-driven insights to ensure actionable decision making.
Data Storytelling Workshops: Equip stakeholders with the skills to translate complex data findings into compelling narratives that drive action.
Benefits of a Precise Ecosystem:
Sharpened Focus: Precise access and clear roles ensure everyone works with the most relevant data, maximizing efficiency.
Actionable Insights: Strategic analytics and automated quality checks lead to more reliable and actionable data insights.
Continuous Improvement: Data-driven performance management fosters a culture of learning and continuous improvement.
Sustainable Growth: Empowered by data, organizations can make informed decisions to drive sustainable growth and innovation.
By focusing on these precise actions, organizations can create an empowered data analytics ecosystem that delivers real value by driving data-driven decisions and maximizing the return on their data investment.
Teaching Continuity: Supporting staff teaching online when face-to-face class...Samantha Lee Pan
Have you ever been tasked with supporting staff in teaching online when face-to-face classes have been cancelled? In difficult circumstances, campus-based activities could be suspended indefinitely due to a significant environmental, health or socio-political impact. In these extreme cases, online teaching can provide a form of emergency management and continuity of teaching and learning. This type of support was needed during the campus shutdowns of 2016 and 2017 caused by student protests that affected South African higher education institutions nationwide. In this session, colleagues from the University of Cape Town (UCT) will share issues, tools and solutions provided to support academic staff required to teach online during those difficult times. The session is designed to provide an example case, but also learn from others. This session will take the form of a birds of a feather discussion, so we welcome others with similar experiences and institutional or individual stories to join.This session is based on the Sakai Virtual Conference 2017 presentation - Under pressure: Supporting staff teaching online in uncertain times (https://youtu.be/50m4skkITeo) but incorporates further questions to help find a solution in your context.
Teaching and Learning Support Activities at Osaka UniversityHaruo Takemura
This presentation describes activities of Teaching and Learning Support Center, Osaka University. These includes faculty development, pre-faculty development, IT systems for education etc.
TechLogic 2014 Keynote on Inverting an Algorithms Class (Extended Version)suthers
Discussion of the inversion of an Algorithms course: how it is motivated by learning theory; how the activities are organized; outcomes. This is an expanded version of an invited keynote talk for the "TechLogic" conference at the University of Hawaii at Manoa.
Show drafts
volume_up
Empowering the Data Analytics Ecosystem: A Laser Focus on Value
The data analytics ecosystem thrives when every component functions at its peak, unlocking the true potential of data. Here's a laser focus on key areas for an empowered ecosystem:
1. Democratize Access, Not Data:
Granular Access Controls: Provide users with self-service tools tailored to their specific needs, preventing data overload and misuse.
Data Catalogs: Implement robust data catalogs for easy discovery and understanding of available data sources.
2. Foster Collaboration with Clear Roles:
Data Mesh Architecture: Break down data silos by creating a distributed data ownership model with clear ownership and responsibilities.
Collaborative Workspaces: Utilize interactive platforms where data scientists, analysts, and domain experts can work seamlessly together.
3. Leverage Advanced Analytics Strategically:
AI-powered Automation: Automate repetitive tasks like data cleaning and feature engineering, freeing up data talent for higher-level analysis.
Right-Tool Selection: Strategically choose the most effective advanced analytics techniques (e.g., AI, ML) based on specific business problems.
4. Prioritize Data Quality with Automation:
Automated Data Validation: Implement automated data quality checks to identify and rectify errors at the source, minimizing downstream issues.
Data Lineage Tracking: Track the flow of data throughout the ecosystem, ensuring transparency and facilitating root cause analysis for errors.
5. Cultivate a Data-Driven Mindset:
Metrics-Driven Performance Management: Align KPIs and performance metrics with data-driven insights to ensure actionable decision making.
Data Storytelling Workshops: Equip stakeholders with the skills to translate complex data findings into compelling narratives that drive action.
Benefits of a Precise Ecosystem:
Sharpened Focus: Precise access and clear roles ensure everyone works with the most relevant data, maximizing efficiency.
Actionable Insights: Strategic analytics and automated quality checks lead to more reliable and actionable data insights.
Continuous Improvement: Data-driven performance management fosters a culture of learning and continuous improvement.
Sustainable Growth: Empowered by data, organizations can make informed decisions to drive sustainable growth and innovation.
By focusing on these precise actions, organizations can create an empowered data analytics ecosystem that delivers real value by driving data-driven decisions and maximizing the return on their data investment.
Adjusting primitives for graph : SHORT REPORT / NOTESSubhajit Sahu
Graph algorithms, like PageRank Compressed Sparse Row (CSR) is an adjacency-list based graph representation that is
Multiply with different modes (map)
1. Performance of sequential execution based vs OpenMP based vector multiply.
2. Comparing various launch configs for CUDA based vector multiply.
Sum with different storage types (reduce)
1. Performance of vector element sum using float vs bfloat16 as the storage type.
Sum with different modes (reduce)
1. Performance of sequential execution based vs OpenMP based vector element sum.
2. Performance of memcpy vs in-place based CUDA based vector element sum.
3. Comparing various launch configs for CUDA based vector element sum (memcpy).
4. Comparing various launch configs for CUDA based vector element sum (in-place).
Sum with in-place strategies of CUDA mode (reduce)
1. Comparing various launch configs for CUDA based vector element sum (in-place).
Techniques to optimize the pagerank algorithm usually fall in two categories. One is to try reducing the work per iteration, and the other is to try reducing the number of iterations. These goals are often at odds with one another. Skipping computation on vertices which have already converged has the potential to save iteration time. Skipping in-identical vertices, with the same in-links, helps reduce duplicate computations and thus could help reduce iteration time. Road networks often have chains which can be short-circuited before pagerank computation to improve performance. Final ranks of chain nodes can be easily calculated. This could reduce both the iteration time, and the number of iterations. If a graph has no dangling nodes, pagerank of each strongly connected component can be computed in topological order. This could help reduce the iteration time, no. of iterations, and also enable multi-iteration concurrency in pagerank computation. The combination of all of the above methods is the STICD algorithm. [sticd] For dynamic graphs, unchanged components whose ranks are unaffected can be skipped altogether.
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...Subhajit Sahu
Abstract — Levelwise PageRank is an alternative method of PageRank computation which decomposes the input graph into a directed acyclic block-graph of strongly connected components, and processes them in topological order, one level at a time. This enables calculation for ranks in a distributed fashion without per-iteration communication, unlike the standard method where all vertices are processed in each iteration. It however comes with a precondition of the absence of dead ends in the input graph. Here, the native non-distributed performance of Levelwise PageRank was compared against Monolithic PageRank on a CPU as well as a GPU. To ensure a fair comparison, Monolithic PageRank was also performed on a graph where vertices were split by components. Results indicate that Levelwise PageRank is about as fast as Monolithic PageRank on the CPU, but quite a bit slower on the GPU. Slowdown on the GPU is likely caused by a large submission of small workloads, and expected to be non-issue when the computation is performed on massive graphs.
Explore our comprehensive data analysis project presentation on predicting product ad campaign performance. Learn how data-driven insights can optimize your marketing strategies and enhance campaign effectiveness. Perfect for professionals and students looking to understand the power of data analysis in advertising. for more details visit: https://bostoninstituteofanalytics.org/data-science-and-artificial-intelligence/
1. Deep Learning for Computer Vision
Fall 2021
http://vllab.ee.ntu.edu.tw/dlcv.html (Public website)
https://cool.ntu.edu.tw/courses/8854 (NTU COOL)
Yu-Chiang Frank Wang 王鈺強, Professor
Dept. Electrical Engineering, National Taiwan University
2021/09/28
2. Before We Start…
• Following Taiwan CDC & NTU Regulations…
• We offer online lectures only for the first three weeks (i.e., 9/28, 10/5, 10/12).
• Starting Week #4, we plan to offer lectures in hybrid forms.
That is, physical lectures will be given each week, & the recorded lectures
(presentation video w/ audio & lecture notes) will be uploaded to NTU COOL afterwards.
• We strongly suggest you to attend physical lectures in bi-weekly patterns.
That is, if your ID #s are even, you can attend the lectures of Weeks #4, #6, #8, etc.
If you ID #s are odd, you can attend those of Weeks #5, #7, #9, etc.
• Please wear a mask to protect yourself & those around you if you come to the class.
• It’s totally OK if you choose to take the online course instead.
• However, # of enrollments of is still based on the classroom capacity (180).
2
3. Deep Learning for Computer Vision
• Time/Location: Tuesday 7, 8, 9 (14:20-17:30) @ BL-113 (starting Week #4)
• Website:
http://vllab.ee.ntu.edu.tw/dlcv.html (public)
https://cool.ntu.edu.tw/courses/8854 (NTU COOL)
https://www.facebook.com/groups/544627580166401 (Facebook: DLCV Fall 2021)
• Required Knowledge & Skills
• Knowledge of linear algebra, vector calculus, and probability (required)
• Programming skills (required in Python)
• Backgrounds in machine learning (strongly preferable)
• This course is offered in English (lecture part).
Q&As/discussions can be in any languages communicable…
3
4. What to Expect from this Course?
• (Deep) Learning-Based Computer Vision
• Fundamentals of machine learning
• Deep learning technologies for
visual representation, classification, synthesis, and beyond
• Differ from the course of Computer Vision in Spring 2020/2021
(more geometry-based computer vision)
• Practical Experiences
• Assignments and projects dealing with real-world visual data
• Final projects possibly supported by industries with prizes
• Lots of work and fast paces, but hopefully helpful with lots of fun!
4
5. Disclaimer
• Currently, you are responsible for your own computing resources (i.e., GPUs),
which are absolutely required for your HWs & the final project.
• Syllabus, course policy, and HW/project details might change over time.
(DL/CV/AI are fast growing/changing research topics.)
• This course will be very demanding and challenging!
• Yes, we did fail students in the past semesters (~5% each semester).
5
6. authentication free quota comment
Azure
(Microsoft Azure)
student ID
credit card
+ USD $ 100
+ USD $ 200
1*K80 =
USD $ 0.9 / hour
GCP
(Google Cloud Platform)
credit card USD $300 / 3 months
AWS
(Amazon Web Service)
credit card -
1 GPU =
about US$ 571.01 / year
Colab
(Google Colaboratory)
no limit -
●session may be interrupted
if running time > certain hours
(exact time unknown)
●suggest only for testing code
Computing Resources (TBD)
Not offered (as of Sept 28th)
However, you do need them for your HWs/final projects.
6
8. Teaching Team & Office Hours
• Instructor: Yu-Chiang Frank Wang (王鈺強)
• Research Areas
• Computer Vision, Machine Learning, Deep Learning, & Artificial Intelligence
• Education
• PhD, ECE, Carnegie Mellon University, 2009
• MS, ECE, Carnegie Mellon University, 2004
• BS, EE, National Taiwan University, 2001
• Contact Info
• Email: ycwang@ntu.edu.tw
• Office Hour for DLCV Fall 2021:
After class; otherwise, appointment by email is required.
8
9. TAs & Office Hours: (all at MK-514)
You should contact your TA by email at ntudlcv@gmail.com .
You can also discuss with TAs & other classmates at the Facebook Page (not via messenger)
12
許元譯 (Yuan-Yi Hsu)
MK-514
TA Hours: Mon 14:20-15:20
范萬泉 (Wan-Cyuan Fan)
MK-514
TA Hours: Fri 12:00-13:00
紀彥仰 (Yan-Yang Ji)
MK-514
TA Hours: Tue 11:20-12:20
陳宇軒 (Yu-Hsuan Chen)
MK-514
TA Hours: Wed 12:00-13:00
周子庭 (Zi-Ting Chou)
MK-514
TA Hours: Tue 10:20-11:20
賴永玄 (Yung-Hsuan Lai)
MK-514
TA Hours: Thu 14:00-15:00
林毓宸 (Yu-Chen Lin)
MK-514
TA Hours: Wed 15:00-16:00
張芷榕 (Chih-Jung Chang)
EE BL-系K
TA Hours: Thu 14:00-15:00
10. Tight (yet tentative) Schedule
10
Week Date Topic Remarks
1 9/28 Course logistics & registration; Machine Learning 101 HW #0 out
2 10/05 Introduction to Convolutional Neural Networks (I) HW #0 (10/08 23:59 due)
3 10/12
Introduction to Convolutional Neural Networks (II)
Tutorials on Python, Github, etc. (by TAs)
HW #1 out
4 10/19 Object Detection & Segmentation
5 10/26 Generative Models & Generative Adversarial Networks HW #1 due
6 11/02 Transfer Learning for Visual Classification & Synthesis (I) HW #2 out
7 11/09 Transfer Learning for Visual Classification & Synthesis (II)
8 11/16 TBD (CVPR week) HW #2 due
9 11/23 Recurrent Neural Networks & Transformer (I) HW #3 out
10 11/30 Recurrent Neural Networks & Transformer (II)
11 12/07 Meta-Learning for Visual Analysis (I) HW #3 due;Final Project Announcement
12 12/14 Meta-Learning for Visual Analysis (II) HW #4 out
13 12/21 From Zero-Shot Learning to Explainable AI
14 12/28 Beyond 2D Vision (3D and Depth) HW #4 due; HW #5-1 out (bonus)
15 01/04 Vision and Language HW #5-2 out (bonus)
16 01/11 Guest Lecture (TBD) HW #5 due
17 01/18 Presentation for Final Projects
11. Textbook (Optional)
• Deep Learning, MIT Press
• Ian Goodfellow, Yoshua Benjio, and Aaron Courville
• Free online versions available at www.deeplearningbook.org
11
12. Textbook (Optional)
• Computer Vision: Algorithms and Applications, Springer
• Richard Szeliski
• Free online versions available at http://szeliski.org/Book/
12
13. About Grading
13
Bonus points available for HWs 1-4. HW 5 (bonus & optional, up to 5%)
HW 0
0% HW 1
15%
HW 2
16%
HW 3
18%
HW 4
18%
HW 5
0%
Final Project
33%
percentage
HW 0 HW 1 HW 2 HW 3 HW 4 HW 5 Final Project
14. • HW assignments (67%)
• HW 0 (non-DL HW, 0% but required)
• HWs #1~4 (15~18% each with possible bonus points)
• HW #5 (optional yet with bonus points)
• Final project + poster presentation + code, etc. (33%)
• Bonus points
• Course participation (e.g., interaction, Q&A, etc.)
• Extra challenges & points in HWs #2-5
• Excellent performance for final project
(e.g., competitions, publications, etc.)
• Cash Prize (optional)
• Selected final projects might be sponsored by industries.
• Details to be announced.
14
About Grading (cont’d)
16. About Course HWs/Projects
• About HW late policy
• In case you have dates, midterms, HW dues for other course, etc.
如期中考、專題、社團、約會、找不到人抄
• All students need to finish HW #0 (see slide/video for DLCV_W1-2)
by 10/8 Fri 23:59pm.
• We will offer free late days based on your HW #0 score.
• For HWs 1~4, up to THREE free late days this semester.
(e.g., 1min ~ 23hr 59min all count as ONE late day.)
• After HW due day, a penalty of 30% per day.
• Once late days are used, the decisions are final.
We’ll maximize your final score
based on HWs score and the late days used.
• No late submission for the final project
16
HW0 Score
(full score: 100)
Number of
Free Late Days
60 - 100 3
30 - 59 2
1 - 29 1
0 / no submission 0
17. About Course HWs/Projects
• About Final Project
• 4 (max) people per group
• Project proposal/progress/final report* + poster & oral presentation
• Selected topics possibly come with cash prizes.
• Details to be announced after mid semester.
• Evaluated by instructor, TAs, and possibly guest judges
• (Intra/inter-group) peer evaluation will be conducted.
• Snack/drinks will be provided during final presentation (week of 1/18).
17
18. Academic Integrity
• Can discuss HW with peers, but DO NOT copy and/or share code
• Plagiarism is against university policy.
• Violation in ANY form for HWs & final project would result in F.
• Seriously, we gave at least five Fs in previous semesters for the above cases.
• Do not directly use code from Internet unless you have permissions.
• If not sure, ask!
• If so, do specify in your HW/project.
• You CANNOT use your published work as your final project.
• However, you are encouraged to extend your previous work.
• Also, you are encouraged to turn your high-quality projects into publications.
18
19. DOs and DONTs for the TAs (& Instructor)
• Do NOT send private messages to the TAs via Facebook.
• TAs are here to help, but they are not your tutors 24/7.
• TAs will NOT debug for you,
including addressing coding, environmental, library dependency problems.
• TAs do NOT answer questions not related to the course (e.g., research, etc.)
• If you cannot make the TA hours, please email the TAs to schedule an
appointment instead of stopping by the lab directly.
19
20. How to Sign Up If Not Already In?
• Capacity
• Classroom capacity: 180; currently registered: 150
• Only additional 30 students allowed due to capacity
• Priority
• EECS > Engr. w/ related backgrounds > Sci. > others
• Based on seniority, backgrounds, programming skills, etc.
• If you are still interested in this course and plan to enroll,
please fill in the following form by 9/28 Tue 23:59:
https://forms.gle/quFfkKvX7HESAaWH7
• We will announce the enrollment results via email
no later than 10/1 Fri 5pm. All decisions are final.
20