|We published the final grades. We are uploading the feedback and it will be available on the h.w section in few hours.|
In your feedback you'll find the pdf with comments in it, and a file "result.txt" with the results of your algorithm and a grade breakdown.
We gave a factor to the performance part, it is calculated as a squared root of the 0-1 accuracy (the 0-1 accuracy is calculated out of 268 patches as there was two blank patches in the test set).
Few project still failed to run, those students will see a comment in the feedback file and should contact the course staff.
Appeals should be submitted via the web-course (in the h.w section), due date is 20/9.
We wish you all Hatima Tova,
the course staff.
We finished checking all your projects, but some of you sent code that doesn't run.
If you got 0 (or don't have a grade) in the project column you are probably one of those.
We open a new submission in the HW section called "project resubmission" please fix your project and upload it as soon as possible.
We will recheck the resubmitted projects by the end of Sunday 9/9.
The failures was mostly due to hard coded directories in your computer, missing files, missing packages or syntax errors. We remind you that we check your projects by script so make sure you follow all the submission instructions published before.
For whom of you that have a grade, we will publish your full feedback in few days as soon as we will publish all grades.
The course staff.
Final Project submission instructions
We postponed the deadline for the final project by 4 days, it is now 4.8.18 23:59.
We opened a new electronic submission in the website HW section. Students who submitted a project to the challenge need to submit a final version to the new submission link.
Submission instructions for your projects:
1. Your submission must contain the (modified) "evaluate.py" file in its main directory.
2. Weights download - If your weights files are too large for the submission system, you can add a python file "weights_download.py". Our evaluation algorithm will first run this file, and then evaluate your results.
3. Run time - You will have 30 minutes for weights download and reconstruction of 3 images and 3 documents (it will be run on NVIDIA Titan Xp).
4. Compatibility - Your algorithm will be evaluated with python 3.6 make sure your code is compatible with this version.
5. Dependencies - Please list all the repositories your algorithm uses as requested.
The course Stuff.
HW1 - grades
We published the grades and the comment for HW1, you can get your comments under the h.w section in the website.
Students who want to appeal on their grade should email Izik no later than 18/7/18.
We fixed the bug we had in the evaluation of the challenge submissions. The updated, now correct, list of three winners are:
1. Yair Feldman and Aviram Magen - 92.48% - 8 point bonus
2. Niv Giladi and Dor Zohar - 85.55% - 5 points bonus
3. Vladimir Polosohin and Natan Begrov - 85.18% - 3 points bonus
As stated before all bonuses announced in class will be respected.
All other challenge participants can recieve their updated evaluation scores by email from Yonatan.
Apologies for the confusion, and congratulations for the winners!
The course Stuff.
Unfortunately, we found an error in our challenge evaluation script.
We now further investigate this and will send an updated evaluation of the submissions tomorrow.
It is likely that the winners ordering will be changed. However, the bonus points that were announced in class today will be respected (every entry that was announced will receive at least the bonus points that were announced).
The course team.
Challenge submission instructions
We will close the submission for the challenge at Saturday 23:59.
Your submission must contain the (modified) "evaluate.py" file in its main directory.
Weights download - If your weights files are too large for the submission system, you can add a python file "weights_download.py". Our evaluation algorithm will first run this file, and then evaluate your results.
Run time - You will have 30 minutes for weights download and reconstruction of 3 images and 3 documents (it will be run on NVIDIA Titan Xp).
Solution description - Please add short description for your solution.
Dependencies - Please list all the repositories your algorithm uses as requested. If you have something that cannot be installed by pip, please email Yonatan in advance.
H.W 4 is published
Please find HW 4 in the HW section.
Due date is July 1st.
AWS Computation Resources
Due to technical/bureaucratic difficulties at the Amazon Web Services side, it turned out to be impossible for us to create all student accounts under one umbrella account of the course (despite huge efforts on our side).
Therefore, we must change the strategy and let you guys register to AWS on your own. You will each get $100 credit, directly from AWS, to run their GPU machines.
The process is very simple and is thoroughly described in a document we created for you. Under "Course Material"=>"General info", you will find detailed instructions guiding you how to do this, step-by-step.
Overall, now you have multiple options to access GPU computation resources:
* Google Colaboratory
* Faculty Servers
* AWS Instances
|EE students can connect to the faculty servers by doing these steps:|
1. Change your TX (Technion single sign on) password as described in
2. Wait few hours
3. A CS account will be opened automatically for you with the same password as the TX (assuming you are registered to the course).
4. Follow the steps in the following link to connect using the TX credentials. The Rishon server is accessible from the CS wifi/network only.
H.W 3 due date postponed
|Due to numerous requests we decided to postponed the deadline for HW 3 to Saturday 9/6.|
Final Project is out
|Please find the final project description file in the HW section.|
H.W 3 is published
Please find HW 3 in the HW section. For this HW you will need GPUs. Your first option is to use google colab (as we used for the tutorials), make sure you setup a GPU runtime under runtime settings.
The second option is the faculty server where every student will have 12 hours of GPU for this HW. In the link below you will find a tutorial about the faculty GPU servers.
Students from other faculties will need to setup a faculty user, if this is the case for you please fill your details in the following link:
(This server is available from the technion only)
All questions about the servers should be sent to , with the title "faculty servers"
H.W 2 deadline postponed
|Due to numerous requests we decided to postpone the deadline for H.W 2 to 20/5/18 at 23:59.|
In the meanwhile make sure you frequently check out the F.A.Q page.
Clarifications for H.W 2
|We have uploaded some F.A.Q to the site.|
Clarification on high-order derivatives
We've added an explanation page regarding high-order derivatives in ANNs under Tutorial 4 material.
This may assist you in solving HW2 using a fully-vectorized (hence faster) implementation.
Homework submission policy and HW1 deadline postponement
We received numerous requests for postponement of the HW1 deadline. We would like to emphasize that the course’s submission policy is:
ככלל, לא תינתן דחיית שיעורי בית פרט למילואים
This time (only) we grant a delay of 24 hours in case this policy was not clear. Thus, everybody is allowed to submit (or resubmit) HW1 until tonight (2/5) at 23:59.
Please start working on the next exercises earlier so that you will make the subsequent deadlines.
We also remind you that submission can be done in pairs only.
The Course Staff
Homework 2 has been published
|Due date is 15/5/18.|
Submission in pairs only.
HW 1 corrections
An updated version of homework 1 has been uploaded.
The major change is in question 2(B), where the mistake bound you need to prove is 12/gamma^2, and not 8/gamma^2.
The other changes are the ones written in the F.A.Q.
H.W 1 due date postpone
|כתוצאה מהחגים ושינוי התרגולים הגשת תרגיל בית 1 תהיה עד תאריך 1/5/18.|
שימו לב שאת שאלה 2 ניתן לפתור על בסיס החומר מהרצאה 2.
שאלה 3 בחלקה מתבססת על תרגול 3 שיועבר בשבוע הבא.
Tomorrow's Tutorial is canceled
|לטובת סנכרון בין ההרצאות לתרגולים, התרגול של יונתן מחר מבוטל.|
החומר יושלם בהמשך הסמסטר.
שימו לב שהשבוע גם התרגול של איציק לא יתקיים עקב מתכונת יום ה' ביום ג' הקרוב.
חג עצמאות שמח,
Homework 1 has been published
|You will be able to solve question 2 after lecture 2.|
Due date is 24/4/18
In order to allocate a GPU machine for you to work on,
please fill in your and your partners Technion email addresses in the following form:
Only one student from each pairs needs to fill the form.
Please only fill the form if you are actively participating in the course (not as a free listener).
For those of you who have not yet found a partner, leave the partner email blank.
Today's Lecture is canceled
|עקב מחלה ההרצאה היום מבוטלת.|
שיעור השלמה יקבע בהמשך.
התרגולים השבוע יתנהלו כסדרם.
בברכת שבוע נעים והחלמה לרן.
|In the following link you can find a good reference for scientific programming in python, make sure you are familiar with lectures 1-4 to get ready for the upcoming homework.|
Please ask any of your friends, who intend to register to the course, to do so ASAP.
In order to organize the computing resources required for the course (AWS machines with GPUs), we need to know the total number of students.
The Course Staff
Annotated versions of lecture 1 and tutorial 1 have been uploaded
|They can be found under "course material"|
Welcome to deep learning 236606
In the course website you can find all the relevant course information.
On Tuesday March 20th, singularly, the tutorial is canceled in order to preserve the order of lectures and tutorials along the semester.
Wishing us all a pleasant semester,
the course staff.