Rijstwafel

8 points

joined

Rijstwafel hasn't added a bio yet

History

History

Thanks everyone!

At some point Bishop mentions “risk”. I’ve seen this term come up elsewhere too but never really understood what it means. Does anyone here know?

Interesting stuff. I’d really like to become more familiar with LA. Can anyone recommend me a book that has some useful exercises on this stuff?

Thanks guys! I understand it better now. When would you encounter this problem in real-life applications, though? I just don't get why Bishop gives us this information, haha :)

@hdidwania I understand. Thanks. His point would then be that we can't simply optimize a change-of-variables function if the transformation is non-linear? Still find it hard to see the practical value of this information :) But it makes more sense, thanks!

@ jdry1729 I think the 1-star exercises are a good start? Or would those be too many? Personally, I've already done the first few chapters' exercises so I don't really mind, but I think it is crucial to discuss them to gain a thorough understanding.

Hi!

I have a few questions:

- Is my intuition right that the choice of using a polynomial for fitting comes from Taylor expansions? I.e. polynomials can resemble any many (any?) functions to some extent?
- What point is Bishop trying to make with the whole change-of-variables paragraph on pages 18-19? Could someone explain this intuitively?
- I'd really like to discuss some exercises but I think the plan was to do them at the end of each chapter?

Thanks!

Hi bdhammel!

About your second point: an approach to neurons with greater natural realism can be found in Spiking Neural Networks. Natural neurons fire after they have received a stimulus that built up until their 'potential' reached a certain threshold. The problem with replicating these neurons artificially is that this function isn't differentiable. However, there have been some successes recently. For a paper: see https://arxiv.org/pdf/1804.08150.pdf .

Hi Yamit!

Definitely interested. Two other guys and I just started a discord channel for PMRL, too. Do you want to merge, i.e. the sharing link to the channel?

Some suggestions we have:

- Discuss the material on Discord every saturday.
- Half a chapter + the 1-star exercises each week.
- I have an Overleaf file with my own summary of the chapters I read so far + my solutions to some exercises. By suggesting edits and commenting questions in the summary we wanted to crystallise the material into an intuitively readable summary.

replied to
Study Group for Goodfellow's "Deep Learning"

I’m in!