April 19 reflection questions
Please respond to these questions by Wednesday noon at the latest. We expect short answers. There is no correct answer and we are not expecting you to spend a long time on this; the goal is to check that you've read the articles and to raise discussion points that we will go over in class.
Please enter your MIT ID
Your answer
[Caruana et al] What are the different approaches that the authors suggest for interpreting the results from learning with few versus very many features, and why are the different approaches needed?
Your answer
[Caruana et al] (Optional) What utility does introducing a bias term serve in terms of improving the interpretability of the learned model?
Your answer
[Lei, Barzilay and Jaakkola ] What are the two objectives that a generator should be able to perform? How do the authors encode these two tasks in their cost function?measurement motifs? How are the two similar/distinct?
Your answer
[Lei, Barzilay and Jaakkola ] In your opinion, what are the limitations of the proposed approach and/or the authors’ evaluation?
Your answer
[Lei, Barzilay and Jaakkola ] (Optional) Why do the authors introduce the hidden s_t state?
Your answer
[Lipton] In Caruana et al, the authors imply that neural nets are not as interpretable as simpler models such as rule-based models. What argument does Lipton present that might go against this conclusion?
Your answer
Submit
Never submit passwords through Google Forms.
This content is neither created nor endorsed by Google. Report Abuse - Terms of Service - Additional Terms