Are you sure Atari was not choosen first, and then its meaning :) ?
Talk1: how many times the first answer is the correct one?
Talk 4: Given all this data and the diagrams in the dashboard: Do you have concrete scenarios where all this is helpful for developers/managers?
Speaker 1: How do you synthesize values for missing arguments (e.g., the example correct method is given an array as input, but also takes a second parameter)?
Talk 1: how does the approach handle symbolic path constraints that cannot be solved by the SMT solver?
Talk 1: Is the approach able to scale up with a dataset such as StackOverflow?
Talk 1: is the approach compatible with obfuscated code (often the case with Android apps)?
Talk 2: to what extent do the features that you used rely on types in their encoding? what would it take to apply your approach to a dynamically-typed language?
Talk1: What would happen if you applied your approach to existing (buggy) code where the code provides an initial close but errant model?
Talk 5: Isn't it a step backwards to trade now applicability against quality, and thus, original techniques such as ROSE and SVD must be considered as well?