Servus! Reinfrocement learning and neural nets are my domain

  • 7 Replies
  • 265 Views
*

Marco

  • Roomba
  • *
  • 12
Servus! Reinfrocement learning and neural nets are my domain
« on: August 08, 2017, 09:51:23 am »
Hello everybody,

First some info about myself.
I'm a master student from Germany. For almost two years, I've worked with neural nets and reinforcement learning. I specifically focus on game simulations. I applied neural nets to the behavior of combat units in StarCraft and I implemented a game as testbed for bench-marking AI algorithms. This game is called and particularly features a framework for easily starting to implement AI. Also, it comes with a tool for automating matches and customizing sequences of matches. I spent the last couple of months trying to get DeepMind's Deep Reinforcement Learning approach working. As the game is implemented with Unity (+ C#), I I used this C# port of ConvNetJS. Well, that port clearly lacks in performance so I'm currently working on a DQN implementation myself, which shall be added to ConvNetSharp. My main goal is to natively use neural networks and reinforcement learning inside of Unity plus GPU support. Pretty soon, I'll approach to write a publication about deep reinforcement learning and continues values (maybe training an AI to play the game Moorhuhn). This is going to happen before my master thesis, That potential publication shall support my desire to do a PhD after my master degree.

So as I'm going for a PhD and as I'm probably the only one at my school to work in that field, I'm seeking for people to discuss with and maybe to join forces. Most of the great papers are simply not composed by single individuals. Also, I already have one or two things to discuss already.

Here is some stuff which I enjoyed recently:


Best
Marco

*

Zero

  • Trusty Member
  • *******
  • Starship Trooper
  • *
  • 332
  • Fictional character
    • SYN CON DEV LOG
Re: Servus! Reinfrocement learning and neural nets are my domain
« Reply #1 on: August 08, 2017, 10:19:34 am »
Kamehameha!! ;D

Serious stuff...

Welcome Marco O0

EDIT: changed name
« Last Edit: August 08, 2017, 01:50:50 pm by Zero »

*

keghn

  • Trusty Member
  • *******
  • Starship Trooper
  • *
  • 495
Re: Servus! Reinfrocement learning and neural nets are my domain
« Reply #2 on: August 08, 2017, 03:29:10 pm »
  Welcome Marcos. Good luck with you PhD.
 Love NN theory. I see you are the master of the details.
 I am doing a lot of work on unsupervised NN logic. Bit different then what you doing. Has more of temporal component like
as in lstm.

 I have my own highly detailed theories on machine life, pre agi, and agi brains, That use NN.
 So i am not the master of what tool and software you like to use.
 I wish you good luck with your PhD. 

 

*

Art

  • At the end of the game, the King and Pawn go into the same box.
  • Global Moderator
  • ******************
  • Hal 4000
  • *
  • 4406
Re: Servus! Reinfrocement learning and neural nets are my domain
« Reply #3 on: August 08, 2017, 03:42:19 pm »
Welcome aboard Marcos!

Best of luck with your thesis, Masters and PhD although we know there's very little luck actually involved.  O0
In the world of AI, it's the thought that counts!

*

WriterOfMinds

  • Trusty Member
  • **
  • Bumblebee
  • *
  • 26
    • WriterOfMinds Blog
Re: Servus! Reinfrocement learning and neural nets are my domain
« Reply #4 on: August 09, 2017, 07:48:59 am »
Hi Marco! It's good to see another active member around here. Have fun.

*

Freddy

  • Administrator
  • **********************
  • Colossus
  • *
  • 6088
  • Mostly Harmless
Re: Servus! Reinfrocement learning and neural nets are my domain
« Reply #5 on: August 09, 2017, 01:39:09 pm »
Hello and welcome to the site Marco :)

*

korrelan

  • Trusty Member
  • ********
  • Replicant
  • *
  • 678
  • Look into my eyes! WOAH!
    • Google +
Re: Servus! Reinfrocement learning and neural nets are my domain
« Reply #6 on: August 09, 2017, 08:04:49 pm »
Welcome Marco.

 :)
It thunk... therefore it is!


 


Users Online

12 Guests, 0 Users

Most Online Today: 34. Most Online Ever: 208 (August 27, 2008, 09:36:30 am)

Articles