Studying In South Bay

Part 2 (2019)
41

Still on my way, not used to public transit

42

Hey Peter
Iā€™m outside the study group area. Can you let me in.

ā€¦
Thanks

43

http://cs231n.stanford.edu/slides/2018/cs231n_2018_lecture06.pdf

44

Here the video covering the Stanford slides

1 Like
45

Batch norm

2 Likes
46

I also highly recommend the first few videos of the Convolutional Neural Networks of Andrew Ng on Coursera

1 Like
47

Will check that out. thanks for sharing.

48

I found this post while looking for something else, but it sounds like your question. Hope it helps or at least gives you some direction:

49

Haha. I was precisely going through this topic now. The subsequent videos of andrew ngā€™s video explains how batch norm works and in this he explains covariate shift - https://www.youtube.com/watch?v=nUUqwaxLnWs

These videos provide a very good explanation of the underlying concept. Here he also talks about where bias gets eliminated in batch norm and it gets replaced with hyper params for batch norm gamma * (normalized_data) + beta. .

50

Yes I went through all of those a few months ago. Itā€™s time to revisit. I really like how he explains everything.

51

The slides are excellent. Went through it again.

1 Like
52

lightgbm seems to be overwhelmingly doing well in a lot of kaggle competitions. Do you all want to discuss about this sometime next week after the lec3 discussion?

1 Like
53

Hi,
I am located in South Bay. Is there a face-to-face meeting group in the area or only this online forum thread?

54

Last two weeks we have been meeting at Peter Kaiserā€™s place at Santa Clara.

55

I would like to join the group meetings in the future, if possible

56

Sounds good. Peter has been hosting it in his apartment complex. We have been meeting on Thursday from 6.30pm to 9.30pm for last 2 weeks.

57

Hi Serge!

Yes, the study group we have been doing is open and welcome to everyone and anyone. If the group growers to larger then 7 people I donā€™t think Iā€™ll be able to host it anymore but for the time being below is the info. I will also post again a day or two before Thursday. Hope to see you there if you can make it.

Thursday, April 4th from 6-9pm at 610 E Weddell Dr, Sunnyvale, CA 94089.

We meet in the co-working room by the front of the building. We normally sit in the meeting room so if you donā€™t see us just give me a call. The co-working space also has free coffee and cocoa. You are also welcome to bring food if you need to but just be sure to be tidy as it is a community space for everyone in the apartment complex.

There is guest parking in the garage and if that is full there is some space for street parking. Call me or text me on my cell phone once you have arrived and I can let you in. 952-250-4121. Or use this as a virtual key for the intercom system - http://tinyurl.com/y57xkwju

Looking forward to seeing and meeting everyone who can attend.

1 Like
58

Thank you @peterk !

59

I was going through my this video and Iā€™m unable to understand why is not being zero centered a problem for activation functions such as sigmoid and relu.

60

Jeremy presented this paper in part 1 - https://arxiv.org/abs/1805.11604 where he says that batch norm doesnā€™t reduce covariate shift