Nervous about this next interview with Konrad Kording

#1

I have an interview scheduled with Konrad Kording next week and I am super nervous about it. If any of you are familiar with his work and have specific topics I should focus on, please let me know.

0 Likes

#2

I’ve enjoyed the series so far, if you stick to the same approach it should be fine!

On their research page, it says:

Our conceptual work addresses information processing in the nervous system from two angles:
(1) By analyzing and explaining electrophysiological data, we study what neurons do.
(2) By analyzing and explaining human behavior, we study what all these neurons do together.

How do they even begin to model such a vast gap? (individual neurons and human behaviour). I’d love to hear him talk about that, with some examples of their research projects.

2 Likes

#3

Oh, and would he describe himself as a Bayesian? (in the context of AI)

1 Like

#4

What about a general question? One you might ask every future interviewee:

If they know something about Numenta’s work, how would they relate their own work to Numenta’s? What common characteristics do they find relevant or interesting? And what could Numenta do to benefit from the interviewee’s experience to improve on its model?

3 Likes

#5

He definitely talked about this.

and this.

Should be public by Friday.

2 Likes

#6

Not to pressure you, Matt (I know you have to juggle many tasks), but is there a part 2 of this interview?

Here’s the part 1 for those who haven’t seen it yet.

0 Likes

#7

Yes, will be out this month.

2 Likes

#8

Part 2 is here. Thanks Matt.

I don’t think I understood the time warp issue. Isn’t that part of the more general synchronicity problem? How does HTM deal with this? For instance when different cortical columns have to vote on determination of an object with different features, what guarantees that all the columns fire close enough of each other (in time) to influence the neighbors correctly? Is there a timing window for the votes to be compared? Or is there a level that behaves slower to allow build-up of the votes?

And this point Dr Kording talked about at the end, concerning the possibly problematic confusion between causality and correlation, isn’t that a bit like the study of thermodynamics, where you stop analysing the reactions of single molecules, and start calculating the statistical behavior of the entire room (or vessel, or body)? This only works because the statistical models are very robust. If that wasn’t the case, thermodynamics wouldn’t be a thing. Who says this is the same for a complex and diverse system as a brain?

0 Likes

#9

Synchronicity. Funny you should mention that. I have been thinking about the same thing for a long time.

1 Like

#10

My lawn hasnt needed a mow for a few weeks so I only just listened to part 2 today.

I really appreciated the amount of time spent discussing my question, it was a great conversation and overall a great episode.

1 Like