Deep Reinforcement Learning

Dear students, 

from now on, you can hand in your non-final version of your write-up for the reviews. As we already announced, the submission is open until Sunday, 26.09.21.

Further, as some of you asked for additional time, we decided to shift the project's deadline. The final version of your project (which we will consider for grading) needs to be handed in together with the final version of your write-up (17.10.21). We hope that this additional amount of time will help you. 

Also, we will from now on evaluate your agents roughly every 6 hours to give you more feedback. 

Best,

The MoSi-DRL Team

Dear students, 

 

we have decided to decrease the average return needed to pass the scenarios S3 and S4. 

To pass, you will need an average return of:

• S3: 70
• S4: 55

As usual, we will test this average reward with the provided Hermes files. 

 

Best,

The MoSi-DRL Team

 

 

Dear students,

as announced in the project presentation, even though you have access to everything you need to carry this out on your own, we still want to provide you with a service that automatically evaluates your submitted agents and reports the results back to you. This service is available starting today.

Note that the fixed random seeds used by our evaluation runs and the exact evaluation parameters are also those used for grading your agents.

Details on how the evaluation service works can be found in our forum, so make sure to register if you have not already done so.

Best,

The MoSi-DRL Team

Dear students, 

 

from now on, the grade of your talk can be found on your Personal Status page. If you want feedback considering your grade, please contact your advisor. 

Further, you can find the slides from the project presentation under Materials.

 

 

Best,

The MoSi-DRL Team

Dear students,

This is a short reminder about tomorrows deadline regarding your talks. Until August 5th 23.59 CEST, you are required to send the following to your supervisor via email:

• A preliminary version of your slides. For now, this can either be one PDF document, or 3 separate documents (one for each of your sub topics).
• A category assignment regarding which group member is going to present which subtopic/category (a,b, or c).
• Four multiple-choice questions (including solutions) which should serve the purpose of testing whether your audience has understood the content of your talks. These can be appended to your slides within the same document, or you can submit them as a separate PDF document.

Note that this submission is necessary for passing the seminar.

 

Best,

The MoSi-DRL Team

 

 

Dear students, 

this is the schedule for the talks:

August 12:
08.30 Value-based Methods
~ 09.00 Policy-gradient Methods
Coffee Break
10.30 DRL for Continuous Action Spaces
~ 11.45 End

August 13:
08.30 Improving Exploration
~ 09.45 Special Topics
Coffee Break
11.30 Project Presentation
~12.30 End

 

Best, 

The MoSi-DRL Team

Dear students, 

 

we have more information to share with you:

1. The talk of your group should last for one hour. Thus, the talk of each of your members should last for 20 minutes. You will need to adhere to the given time +- 5 %.

2. As 6 hours of talks are too long for you to stay focused and all of you said to also be available on August 13, we would like to have the talks on August 12 and August 13 in the morning on both days. If you have any conflicts, please contact us. 

 

Further, your advisor will stretch out to make an appointment with you to discuss your progress in understanding the papers. 

Best, 

 

 

Dear Students, 

this is just a short reminder that you have to send your topic preferences by mail until Sunday at the latest. In case you haven't done so already, please use the following scheme:

Favour: 1,4 
Block: 3

(and please keep in mind that: 1 = Value-based Methods, 2 = Policy-gradient Methods, 3 =  DRL for Continuous Action Spaces, 4 = Improving Exploration, 5 = Special Topics / Model-Based DRL) 

Further, keep in mind that you have to register in the LSF. You can do that until July 26. Please don't forget to do that, as we have no options of letting you pass the seminar without an LSF registration. 

 

Best, 

The MoSi-DRL Team

Dear students, 

we just uploaded the organization slides from todays kick-off meeting under Materials. 

There are a few changes to the version you saw this morning, all as discussed:

- we added information about the multiple choice questions you have to provide about your talk.
- we changed the information about the topic selection (such that now everyone has to write an e-mail). 
- the typo with the topic selection deadline was fixed (it is July 11). 

We will publish the group assignment and the associated advisors next Monday (July 12). 

All appointments can now also one found under Timetable.

If you have questions, don't hesitate to contact us. 

 

Best, 

The MoSi-DRL Team

Dear students, 

we plan to have all talks as a (digital) block on a single day. 

To check your availability, you can now find four registrations (August 5,6, 12, and 13) on your Personal Status page. Please make use of that function in the following procedure:

- If you are fully available on that day, register for it. 
- If you are unavailable on that day due to an exam, do not register for it. 
- If you are available in principle but have a conflicting appointment, register for that day and write an email to Timo containing the information when you are unavailable. 

Please register for all dates that you are available on, not just the ones you like. If you don't take part (i.e., don't register for any day) we assume that you are always available. If you really can't attend any of these days, please also contact Timo. 

Please do so until Friday (June 25)!

 

Best, 

The MoSi-DRL Team