Welcome, Guest: Join Nairaland / LOGIN! / Trending / Recent / New
Stats: 2,640,106 members, 6,178,642 topics. Date: Wednesday, 03 March 2021 at 09:08 PM

A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. - Programming (2) - Nairaland

Nairaland Forum / Science/Technology / Programming / A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. (2968 Views)

Are You Interested In Robotics And Artificial Intelligence? / A Thread For Tutorial On Python Programming / Artificial Intelligence And Machine Learning Group (2) (3) (4)

(1) (2) (3) (4) (Reply) (Go Down)

Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Semtu(m): 5:19pm On Oct 30, 2020
Interesting... I'm good in Machine learning and currently rounding off a course on deep learning with tensorflow and Keras api. I guess Reinforcement learning is my next junction.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Drybones: 7:21pm On Oct 30, 2020
Semtu:
Interesting... I'm good in Machine learning and currently rounding off a course on deep learning with tensorflow and Keras api. I guess Reinforcement learning is my next junction.

Won't you at least practice deep learning?
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by ibromodzi: 10:11pm On Oct 30, 2020
Enceladus:
Conclusion

The experiment was a bit of success although the agent didn’t maximize its potential mainly because it can only sense the board only through the region around it’s head and a better option is the snake seeing the whole board as an image through a convolutional neural network. The agent was able to eat about 50 apples at max.

For the next experiment, I will be training an agent to play the old shooting game CONTRA and I am really excited about the training. I will also build upon the lesson learnt by this experiment by giving the agent more awareness of the environment by providing the agent the direct image render of the screen. IF the next experiment is successful, other prospective games for experimenting include Asphalt, Mortal Kombat etc. and also maybe soccer (but definitely the old versions like winning eleven 2008 and will probably need the help of a C++ dev to tune the game to a usable environment)
Adios guys.

This is amazing!

I also use Pytorch for deep learning. I recently trained a neural net model and wrote a utility function that can help eye care specialists (ophthalmologists and optometrist) differentially diagnose certain eye conditions (conjunctivitis and uveitis) in real time. I achieved an accuracy of close to 90% but I still need something higher than that. My ultimate goal is to build a stand alone system where the model can be embedded and used by clinical experts.

I also find NLP very interesting and I have developed, debugged, tested and deployed a keyword generating API that extracts important terms from a text. Currently, I'm working on a Chabot using Rasa. I want it to be able to hold reasonably conversations regarding eye health and make tentative diagnoses based on users' intents.

Data science as a whole is very interesting but I don't think anyone can go far by just watching YouTube videos and taking some courses online. Personally, I combine PDFs with videos and do a lot of hands-on.

1 Like

Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Enceladus(m): 11:38pm On Oct 30, 2020
ibromodzi:



I also use Pytorch for deep learning. I recently trained a neural net model and wrote a utility function that can help eye care specialists (ophthalmologists and optometrist) differentially diagnose certain eye conditions (conjunctivitis and uveitis) in real time. I achieved an accuracy of close to 90% but I still need something higher than that. My ultimate goal is to build a stand alone system where the model can be embedded and used by clinical experts.

Absolutely fantastic... The CNN (?) might even find clues and patterns oblivious to us that might help it identify these issues perfectly.


I also find NLP very interesting and I have developed, debugged, tested and deployed a keyword generating API that extracts important terms from a text. Currently, I'm working on a Chabot using Rasa. I want it to be able to hold reasonably conversations regarding eye health and make tentative diagnoses based on users' intents.

I focus on NLP mainly also, and I think they might be the last nut to crack before we get artificial general intelligence. Like for example GPT-3 is showing some signs it can reason and decode some quiet complex natural language and it understands a statement in multiple languages. I once started a project like that, but the sheer volume of the effort required made me abandon it sharply grin grin



Data science as a whole is very interesting but I don't think anyone can go far by just watching YouTube videos and taking some courses online. Personally, I combine PDFs with videos and do a lot of hands-on.

Absolutely. No hands-on coding while learning == Binge-watching Netflix.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by olamidedivotee(m): 10:06am On Oct 31, 2020
Enceladus:


I started with Tensorflow actually, but it doesn't feel pythonic to me so I switched to pytorch but with the Keras api now merged with the new Tensorflow, I think I might start to use it again.

I’m currently working a traffic light(Computer Vision) model to be deployed to PyBoard and Arduino.


Sounds interesting, could you explain more about it. And why PyBoard and not raspberry pi.
I'm very interested in AI i need someone to put me through.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Enceladus(m): 10:23am On Oct 31, 2020
olamidedivotee:

I'm very interested in AI i need someone to put me through.

Do you code?
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by ibromodzi: 12:37pm On Oct 31, 2020
Enceladus:




Absolutely fantastic... [b]The CNN (?) [/b]might even find clues and patterns oblivious to us that might help it identify these issues perfectly.



I focus on NLP mainly also, and I think they might be the last nut to crack before we get artificial general intelligence. Like for example GPT-3 is showing some signs it can reason and decode some quiet complex natural language and it understands a statement in multiple languages. I once started a project like that, but the sheer volume of the effort required made me abandon it sharply grin grin




Absolutely. No hands-on coding while learning == Binge-watching Netflix.

Yeah, CNN. Come to think of it, despite the tremendous success deep learning has achieved in medicine, clinical experts are of the opinion that they can not rely on a technology whose decision making process is not evidence based (and this is what standard medical practice is all about).

Given a vast amount of training examples, neural nets are very good at extracting features and making precise predictions that can even surpass human performances in certain cases. These remarkable achievements are quite fascinating but there's a problem; we do not know how these features are being extracted given the black box nature of neural networks. This is a major concern for health care experts and it is quite debilitating to the adoptability of AI in medicine.

As for NLP, frankly speaking, I enjoy doing it more than CV and I hope to take some advanced courses in it to gain more knowledge about the field.

GTP-3 is a monster but we are yet to see a truly intelligent system. Anyone to come up with that will control the world for the next 50 years because so many industries will be disrupted and countless number of businesses will go into extinction.

What NLP library/libraries do you enjoy using most? Spacy and Textblob top the list of my own favorite libraries. Also do you process any other language apart from English?
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by olamidedivotee(m): 12:53pm On Oct 31, 2020
Enceladus:

Do you code?
yes, very well. With Python but I'm still using my phone cos no laptop.

1 Like

Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Abcruz(m): 1:33pm On Oct 31, 2020
ibromodzi:


Yeah, CNN. Come to think of it, despite the tremendous success deep learning has achieved in medicine, clinical experts are of the opinion that they can not rely on a technology whose decision making process is not evidence based (and this is what standard medical practice is all about).

Given a vast amount of training examples, neural nets are very good at extracting features and making precise predictions that can even surpass human performances in certain cases. These remarkable achievements are quite fascinating but there's a problem; we do not know how these features are being extracted given the black box nature of neural networks. This is a major concern for health care experts and it is quite debilitating to the adoptability of AI in medicine.

As for NLP, frankly speaking, I enjoy doing it more than CV and I hope to take some advanced courses in it to gain more knowledge about the field.

GTP-3 is a monster but we are yet to see a truly intelligent system. Anyone to come up with that will control the world for the next 50 years because so many industries will be disrupted and countless number of businesses will go into extinction.

What NLP library/libraries do you enjoy using most? Spacy and Textblob top the list of my own favorite libraries. Also do you process any other language apart from English?


You're totally correct about the skepticism regarding the adoption of AI in medicine. Some doctors are strongly against it because of the threat of losing their professional relevance through automation. On the other hand, people are scared of entrusting their health challenges in the hands of a machine who has no empathy or emotion and probably designed and deployed by someone who isn't medically grounded to recognize misdiagnosis of the system.

Nonetheless, medicine is a very complicated field of endeavor that cannot be practiced without the human touch. It will be in the best interest of both doctors and patients if AI is designed to automate routine clinical practices while the doctors focus more on care based medicine.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Abcruz(m): 1:41pm On Oct 31, 2020
Concerning General Intelligent AI, a prototype called "Baby X" has already been developed by soulmachine. So far, it can learn from it's environment with the aid of the camera and can respond to simple questions just like a normal human baby. The baby moves his body autonomously. This is only the beginning.

1 Like

Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Enceladus(m): 2:07pm On Oct 31, 2020
ibromodzi:


Yeah, CNN. Come to think of it, despite the tremendous success deep learning has achieved in medicine, clinical experts are of the opinion that they can not rely on a technology whose decision making process is not evidence based (and this is what standard medical practice is all about).

Given a vast amount of training examples, neural nets are very good at extracting features and making precise predictions that can even surpass human performances in certain cases. These remarkable achievements are quite fascinating but there's a problem; we do not know how these features are being extracted given the black box nature of neural networks. This is a major concern for health care experts and it is quite debilitating to the adoptability of AI in medicine.

As for NLP, frankly speaking, I enjoy doing it more than CV and I hope to take some advanced courses in it to gain more knowledge about the field.

GTP-3 is a monster but we are yet to see a truly intelligent system . Anyone to come up with that will control the world for the next 50 years because so many industries will be disrupted and countless number of businesses will go into extinction.

What NLP library/libraries do you enjoy using most? Spacy and Textblob top the list of my own favorite libraries. Also do you process any other language apart from English?

Yeah sure like Abcruz said, use of AI in medicine and in the military (robot soldiers, autonomous UAV's e.t.c) is of great ethical concerns. And like you rightly said the fact that NN's are black boxes doesn't help the case either.

Sure we don't have a true AGI yet but gpt-3 is giving us a few flashes of the future and in that it is raking in billions of dollars for open-ai even at its infancy.

Textacy built on spacy (mainly because of phrase detection like vbs), huggingface libraries (transformers, neuralcoref) with TextBlob are my go-to tools. I've also work with Italian and passively working on pidgin.

1 Like

Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by ibromodzi: 2:08pm On Oct 31, 2020
Abcruz:



You're totally correct about the skepticism regarding the adoption of AI in medicine. Some doctors are strongly against it because of the threat of losing their professional relevance through automation. On the other hand, people are scared of entrusting their health challenges in the hands of a machine who has no empathy or emotion and probably designed and deployed by someone who isn't medically grounded to recognize misdiagnosis of the system.

Nonetheless, medicine is a very complicated field of endeavor that cannot be practiced without the human touch. It will be in the best interest of both doctors and patients if AI is designed to automate routine clinical practices while the doctors focus more on care based medicine.

Well said. Handling of routine clinical practices by intelligent machines will greatly reduce heavy reliance of many patients on a single doctor (talk of patient doctor ratio) most especially in countries like Nigeria.

1 Like

Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Enceladus(m): 2:16pm On Oct 31, 2020
olamidedivotee:

yes, very well.
With Python but I'm still using my phone cos no laptop.

Ok. Since you code, you can get a good introduction from the book i mentioned above, another good material is deeplizard's channel on youtube. But like ascertained above you have to code along with your tutors to cement the knowledge. Since you don't have a laptop yet, you can can start with the machine learning book first and use your (android ?) phone to code along (Pydroid on play store can install most of the non deep learning libraries). GL
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by ibromodzi: 2:33pm On Oct 31, 2020
Abcruz:
Concerning General Intelligent AI, a prototype called "Baby X" has already been developed by soulmachine. So far, it can learn from it's environment with the aid of the camera and can respond to simple questions just like a normal human baby. The baby moves his body autonomously. This is only the beginning.

Interesting!
I'm reading about Baby X for the first time here. I'll surely read more about it. However, as a quibble, the fact that it learns from its environment suggests it uses some sort of reinforcement learning (RL)principles and RL is not new in the realm of AI.

Nevertheless, it is good to see that the AI community is still enthusiastic about the development of AGI, given the huge amount of resources being dedicated to the endeavor.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by ibromodzi: 2:39pm On Oct 31, 2020
Enceladus:


Yeah sure like Abcruz said, use of AI in medicine and in the military (robot soldiers, autonomous UAV's e.t.c) is of great ethical concerns. And like you rightly said the fact that NN's are black boxes doesn't help the case either.

Sure we don't have a true AGI yet but gpt-3 is giving us a few flashes of the future and in that it is raking in billions of dollars for open-ai even at its infancy.

Textacy built on spacy (mainly because of phrase detection like vbs), huggingface libraries (transformers, neuralcoref) with TextBlob are my go-to tools. I've also work with Italian and passively working on pidgin.

Great. I haven't used any hugging face library before. I'll be glad if you can recommend some resources on them. I also work with Arabic texts.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Abcruz(m): 4:25pm On Oct 31, 2020
ibromodzi:


Interesting!
I'm reading about Baby X for the first time here. I'll surely read more about it. However, as a quibble, the fact that it learns from its environment suggests it uses some sort of reinforcement learning (RL)principles and RL is not new in the realm of AI.

Nevertheless, it is good to see that the AI community is still enthusiastic about the development of AGI, given the huge amount of resources being dedicated to the endeavor.

Definitely, Baby X uses reinforcement learning to retain knowledge, It uses computer vision to see the environment, it uses natural language processing(NLP) to understand what is been spoken and how to respond.

A book containing different diagrams was shown to it and they asked him what do you see? and he replied everything correctly.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by shuddyy: 4:40pm On Oct 31, 2020
Massive interesting stuffs I have read here. It has further motivated to learn more on ML and AI.
I'm an enthusiast just getting to learn the basics. My interest encouraged me to take a research project on NLP. I hope to learn in the process and build my knowledge of AI and ML therefrom.

Please if you are available to handle this project for a fee i will be glad you contact me.
Its a project on NLP.

Development of a Machine Learning Model for Multi-labeled Toxic Comment Classification.

I'm expected to develop a Machine Learning model based on LSTM.

The model would be used to classify the kaggle Toxic Comment dataset and compare the result of the classification with the results in some literature that would have reviewed in chapter two.


Cmparison will be based on some selected performance metrics eg

Accuracy
Sensitivity or recall
Specificity
F-Measure

My email: aashuddyy@gmail.com
07032320992

1 Like

Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Drybones: 5:05pm On Oct 31, 2020
Abcruz:


Definitely, Baby X uses reinforcement learning to retain knowledge, It uses computer vision to see the environment, it uses natural language processing(NLP) to understand what is been spoken and how to respond.

A book containing different diagrams was shown to it and they asked him what do you see? and he replied everything correctly.
There is a popular robot Sophia.
I think that is more advanced than this, it has a Twitter account verified,and has spoken in conferences, even has a Saudi Arabian citizenship.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Semtu(m): 5:06pm On Oct 31, 2020
shuddyy:
Massive interesting stuffs I have read here. It has further motivated to learn more on ML and AI.
I'm an enthusiast just getting to learn the basics. My interest encouraged me to take a research project on NLP. I hope to learn in the process and build my knowledge of AI and ML therefrom.

Please if you are available to handle this project for a fee i will be glad you contact me.
Its a project on NLP.

Development of a Machine Learning Model for Multi-labeled Toxic Comment Classification.

I'm expected to develop a Machine Learning model based on LSTM.

The model would be used to classify the kaggle Toxic Comment dataset and compare the result of the classification with the results in some literature that would have reviewed in chapter two.


Cmparison will be based on some selected performance metrics eg

Accuracy
Sensitivity or recall
Specificity
F-Measure

My email: aashuddyy@gmail.com
07032320992

Let me take this up, I've sent you a message on WhatsApp
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Abcruz(m): 5:23pm On Oct 31, 2020
Drybones:

There is a popular robot Sophia.
I think that is more advanced than this, it has a Twitter account verified,and has spoken in conferences, even has a Saudi Arabian citizenship.

AI is a software not an hardware.
Sophia is a robot not an AI.
Most of the things Sophia speaks at conferences are directly writing into it or controlled remotely just as the developers at hanson robotics rightly stated.

An AI wholely make use of digital presence to understand and interact while leveraging on complex algorithms to see, hear, process, understand and react with a measurable precision.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Enceladus(m): 5:43pm On Oct 31, 2020
ibromodzi:


Great. I haven't used any hugging face library before. I'll be glad if you can recommend some resources on them. I also work with Arabic texts.

I mostly read through their readme file and documentation on github. Here is also a good article. Arabic text really? How do you deal with it's Semitic nature?
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Drybones: 6:38pm On Oct 31, 2020
Enceladus:
The training took about 26 hours on an hyper threaded Intel Xeon CPU (good reward aware policy was learned around 5 hours). Although I tried training with a free tesla K80 GPU from google [color=#990000][/color]
Can you shed light on the training?
Does it mean you left your system on for 26hours?
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Enceladus(m): 7:33pm On Oct 31, 2020
Drybones:

Can you shed light on the training?
Does it mean you left your system on for 26hours?

I made use of a virtual machine from google cloud actually. The vm was attached to my google drive and it saves the agent checkpoint every n hours.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Abcruz(m): 8:26pm On Oct 31, 2020
Enceladus:


I made use of a virtual machine from google cloud actually. The vm was attached to my google drive and it saves the agent checkpoint every n hours.

Is the virtual machine free or paid? And did you leave the PC on for 26hrs while running the virtual machine?
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Enceladus(m): 8:50pm On Oct 31, 2020
Abcruz:


Is the virtual machine free or paid? And did you leave the PC on for 26hrs while running the virtual machine?

The vm is paid, but I used a bit of an hack to minimize the price to something around N600 for the whole 26 hours. And no my PC was mostly off during the training but I occasionally login to the vm to check training progress.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by ibromodzi: 10:02pm On Oct 31, 2020
shuddyy:
Massive interesting stuffs I have read here. It has further motivated to learn more on ML and AI.
I'm an enthusiast just getting to learn the basics. My interest encouraged me to take a research project on NLP. I hope to learn in the process and build my knowledge of AI and ML therefrom.

Please if you are available to handle this project for a fee i will be glad you contact me.
Its a project on NLP.

Development of a Machine Learning Model for Multi-labeled Toxic Comment Classification.

I'm expected to develop a Machine Learning model based on LSTM.

The model would be used to classify the kaggle Toxic Comment dataset and compare the result of the classification with the results in some literature that would have reviewed in chapter two.


Cmparison will be based on some selected performance metrics eg

Accuracy
Sensitivity or recall
Specificity
F-Measure

My email: aashuddyy@gmail.com
07032320992

Hi! Let's work on this together. Just sent you a mail.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Predstan: 10:12pm On Oct 31, 2020
olamidedivotee:

yes, very well.
With Python but I'm still using my phone cos no laptop.

Lol. Code very well on phone You haven’t started coding.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by ibromodzi: 10:19pm On Oct 31, 2020
Enceladus:


I mostly read through their readme file and documentation on github. Here is also a good article. Arabic text really? How do you deal with it's Semitic nature?

Very easy to work with once you understand the language. The major work lies in the text preprocessing.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by ibromodzi: 10:24pm On Oct 31, 2020
Enceladus:


The vm is paid, but I used a bit of an hack to minimize the price to something around N600 for the whole 26 hours. And no my PC was mostly off during the training but I occasionally login to the vm to check training progress.

So what happens if my code throws errors?
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Abcruz(m): 10:49pm On Oct 31, 2020
Enceladus:


The vm is paid, but I used a bit of an hack to minimize the price to something around N600 for the whole 26 hours. And no my PC was mostly off during the training but I occasionally login to the vm to check training progress.

Wow that's cool. Do you mind sharing the hack?
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Abcruz(m): 10:52pm On Oct 31, 2020
Predstan:


Lol. Code very well on phone You haven’t started coding.

There's nothing wrong with it. I have been using my phone to code since my PC crashed and I find it flexible.
Re: A Thread On My Experiment With Reinforcement Learning & Artificial Intelligence. by Predstan: 6:01am On Nov 01, 2020
Abcruz:


There's nothing wrong with it. I have been using my phone to code since my PC crashed and I find it flexible.

Lol. What IDLE are you using on you phone? What libraries are you using Again you haven’t started coding. You are still joking and playing hide and seek

(1) (2) (3) (4) (Reply)

Learning C++ And Dreamweaver / Learn Programming For Free, Right From Your Couch! / Asp.net/mssql Versus Php/mysql

(Go Up)

Sections: politics (1) business autos (1) jobs (1) career education (1) romance computers phones travel sports fashion health
religion celebs tv-movies music-radio literature webmasters programming techmarket

Links: (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Nairaland - Copyright © 2005 - 2021 Oluwaseun Osewa. All rights reserved. See How To Advertise. 197
Disclaimer: Every Nairaland member is solely responsible for anything that he/she posts or uploads on Nairaland.