Bjrn

• Bjrn 8 Jan 2020 6:22 UTC

  2 points

  on: What were the biggest dis­cov­er­ies /​ in­no­va­tions in AI and ML?

  From A ‘Brief’ History of Neural Nets and Deep Learning, Part 4:

  So, why indeed, did purely supervised learning with backpropagation not work well in the past? Geoffrey Hinton summarized the findings up to today in these four points:

  1. Our labeled datasets were thousands of times too small.

  2. Our computers were millions of times too slow.

  3. We initialized the weights in a stupid way.

  4. We used the wrong type of non-linearity.

  I think this blog series might help provide a partial answer to your question.

• Bjrn 27 Nov 2019 8:01 UTC

  2 points

  on: My Anki patterns

  Would it be possible to share some CSS/​images showing off your card æsthethethics?

• Bjrn 8 Nov 2019 6:15 UTC

  1 point

  on: Per­sonal qual­ity experimentation

  It seems largely true to me that it is not hard to create a temporary change in one of the axes, but I am curious if these can lead to permanently changed settings in the long run. I would be very curious to hear from anyone who’s created lasting change through an experiment like this and what axis in particular they changed.

  Personally, I’ve experimented with acting much more extroverted than I typically am in certain social contexts (interviews are a good example) and the new setting feels comfortable or even better at times. Gaining more experiences like this certainly makes the new setting seem more appealing but it does not seem to become easier, and I will always revert back to my default setting once the context changes. Maybe the shyness axis is particularly difficult to change.