Q1: Could you please give me an example to consider that which model should be applied when we want to predict downtime of the machine, when we have only the data collected from sensors which installed on that machine?

Ans: It's really depend on the problem and quality of the result. If you hope to know which method have better result.  You can use some exist toolbox to put part of your data to get quick result for your first evaluation. Google AutoML maybe is a good tool to help you.

Q2: Every time scientist found a  new technique of AI  what is the trigger point,  where these knowledge come from,  (eg : Newtown found law of gravity from sitting under apple tree)

Ans: This is a good question.  For AI point of view, there are included many field of technology.  like computer knowledge, psychology and philosophy..etc.  From Algorithm point of view, the methods of AI was established on the statistics. Because all the tech was be created to solve the problem with large data.  

Q3: How to know how much data is enough?

Ans: It's really depend on the case and requirement. You can use few samples with good quality data to train a model for specific requirement. (The cover range and data tolerance was limited).  If you have lot of data but without the good quality, the performance of the model result is still bad.  So don't focus on the data qt'y,  but the data quality and coverage. Or you can share your application in detail, we can have expert to provide the right answer for data requirement. 

Q4: When u show confusion matrix and explain about the term recall and precision....It sounds like some of terms sensitivity and specificity in diagnostic testing?

Ans: Following is the rule of the precision, recall (sensitivity) and specificity sensitivity, recall, hit rate, or true positive rate (TPR).

Q5: How do we apply tag/label on the features?

Ans: You tag the data, the data with features.  (ex: the dog picture with the feature of dog, you just tag the picture as dog. Machine will learn that. )

Q6: What is the next generation after Deep Learning?

Ans: Some people said maybe the Quantum computer is next important technology. 

Q7: What will be the most plausible AI model to recognize animals in the world? Why is this so difficult?

Ans: AI model need data. If you have enough labelling data, training a model to recognize animals is not a difficult things. (You can search or collect as much as labelling animal pictures. Than you can use supervised learning to train a model for your recognize requirement. )

Q8: Please provide some more URLs for us to play around with RL, and GAN  network  ( similar to the TensorFlow playground URL for NN)?

Ans: Please try to type notexist to searching in google. Yo can fine a lot of website for GAN.

Q9: In developing of autonomous cars, what learning method they use?

Ans: Supervised learning is most useful method for image related prediction.

Q10: In manufacturing, what kind of application and what kind of the ML model is used in production?

Ans: Almost all kind of methods used in manufacturing. Because different problem need to use suitable method.

Q11: Apart from robotics and systems control requiring interaction, what are the other fields where we can use reinforcement learning?

Ans: A GAME agent is good target. Like AlphaGo.  RL is not a good solution for real world cases. Because the award(Score) is hard to define well.

Q12: Please explain how are ML methods for classification, regression, clustering etc. different from their ‘classical statistical methods’ counterparts like linear/logistic regression, principal component analysis. thank you.

Ans: You can read the "Introduction to machine learning with Python" Andreas C. Müller and Sarah Guido to know more in detail.

REF: TTAIC & AIGO & III