پروژه مونت کارلو (80670))= 2864

---

شرح پروژه

#پروژه_جدید
#کدپروژه : 80670
موضوع : پروژه مونت کارلو =
سلام وقت بخیر 2864
برای انجام پروژه مزاحم شدم
: سلام مجدد حدود یکماه وقت دارم
: (Your paper should at least include c)

2.     Explain sample complexity for RL Wikipedia link

3.      Explain the Double Q-learning link

4.      Explain the Atari paper(not much math, but you will have to present it) . Link

5.     Explain the Automatic Domain Randomization  link(Only section 5)
: Basic rules:

1.      Report is expected to be 2-3 pages. This will include background, model/solution method studied, and the results. (The code must be appended separately)

2.     All original work should be cited, including papers, publicly available code used (website or GitHub)

3.     There will be a small interview where you will explain your work.

4.     You have the option to also do the final exam and take the better grade between the exam and the project.

5.     The grade will be assigned based on the clarity of your understanding and/or implementation. In particular, the difficulty of the problem does not contribute to the grade.

6.     In the case of a paper, a clearly formulated result and outline of its proof is expected (rather than a copy-paste from the original source!)


Suggestions:

Simple project ideas:


In what following example numbers refer to the book (SB):

Reinforcement Learning: An Introduction, Sutton and Barto, 2nd Edition. This is available for free here and references will refer to the final pdf version available here.


1.      Blackjack Example 5.3 page 99

2.     Windy gird world Example 6.4 page 130

3.     Shortest path in other simple situations

4.     Gradient Bandit Algorithm (SB 2.9) for non-stationary bandits. (Slightly harder)

5.     Use Gymnasium (a fork of openAI gym) to design an agent learn an optimal policy for one of their environments.

  Somewhat more involved:

1.     Checkers see  SB 16.2. The difficulty here is to construct suitable reward function (done historically by propagating a minimax tree search).  The implementation of TD is fine. (Also you need some effective dynamic programming in the sense of programming)

2.     Ambitious would be to implement Backgammon



Some paper ideas:

1.     Stochastic Approximation Theory results with proofs:

a)     Robbins-Monro Algorithm

b)    Stochastic gradient method

c)     Q-learning converges to optimal.

(Your paper should at least inc
: باید یکی ازاین تمرینهایی که استاد انتخاب کردن رو ترجیحا ساده ترینش  روبه زبان مونته کارلو بنویسیم
: فکر میکنم منظورش اینه

این پروژه شامل 1 فایل مهم است، لطفا قبل از ارسال پیشنهاد حتما نسبت به بررسی این فایل اقدام فرمایید.

مهارت ها و تخصص های مورد نیاز

---

---