i want to know the explanation to previous exam last question

Hi Farzan!

Thanks for reaching out.

Can you please specify which exam you are referring to? Where is it that we say that "value function" is not an ML term at all? This will help us assist you better. Thank you.

Looking forward to your answer.
Best,
Martin

I have the same question as Farzan Reza above. 

Its question 4 on Practice exam: 2 as per the screenshot below:

Based on the below website, value function is one of the 4 main ingredients in Reinforcement learning:

https://towardsdatascience.com/introduction-to-reinforcement-learning-rl-part-1-introduction-c0d55c1240a3

Thanks in advance for your inputs!

Navin

Hi Navin!

Thanks for reaching out.

First of all, thank you for searching on the web for additional information. However, we cannot guarantee truthfulness of external websites, nor can we ensure they provide enough context when discussing a certain topic.

Value function is a term that can be applied for reinforcement learning only, because there we can talk about finding a policy that should maximize the reward. In contrast to that, out of this context, Rachel and Ross seem to be discussing general terms, related to mathematics more so than to reinforcement learning. That's why value function does not necessarily relate to a better performance per sé. Otherwise, applying the 'mathematical' terms to the context of reinforcement learning, it works. You can think of it as : we can say 'objective function' in the context of reinforcement learning, but cannot generalize when saying 'value function'. 

In other words, it is important to not confuse 'value function' and 'objective function'. As explained in the beginning of the following video: https://learn.365datascience.com/courses/intro-to-data-and-data-science/machine-learning-ml-techniques/

we present the machine learning in general from a single standpoint and explore how to use the same theory in the different types of machine learning later.
Hope this helps.
Kind regards,
Martin

Thanks Martin!

Based on the lectures, I understood that "the objective function" measures the inaccuracy of the model and needs to be minimized however the answer to the above question states that maximizing the reward equals maximizing the objective function?

Hi Navin!

Thanks for reaching out.

We either minimize the objective function to minimize the error (supervised, unsupervised learning) and thus improve the ML model, or we maximize it to optimize the reward (reinforcement learning). But its the same type of equation.

Here's a further explanation, if you'd like to read the same explanation in a different way:

https://learn.365datascience.com/question/minimise-or-maximise-the-objective-function/

Hope this helps.
Kind regards,
Martin