Study group Polska

What about decreasing size 512 ?

Hmm… I want to train my model with size 512 (or even bigger later), just looking for the most efficient and affordable solution. I know that lower size will work - I started from size 128 :slight_smile:

Did you try preemptive vms, a lot cheaper. https://cloud.google.com/preemptible-vms/

I tried, but still… I prefer coffee to increase blood pressure. :slight_smile:
I was very unlucky with this feature: “Compute Engine gives you 30 seconds to shut down when you’re preempted”.

OK I understand it:)

Resnet originally was trained on 224 and 299 sizes

Also, what is better bigger bs or bigger image size ???

I was playing with some models and results were interesting many times counter-intuitive

M

Coffee not that good for you either :slight_smile:
So far I was lucky didn’t loose my data…

From fastai GCP setup guide:

According to our benchmarks, a K80 is 84% slower than a P4

part 2 dates were announced here

Dates March 18 - April 30 (7 weeks)
Schedule 6:30-9 p.m.

on
Mon Mar 18
Mon Mar 25
Wed Apr 3
Wed Apr 10
Wed Apr 17
Tue Apr 23
Tue Apr 30|

anyone thinking to continue with part 2? if yes, what are we going to do in between? :slight_smile:

3 Likes

Online meetup tonight?

yeah, ill be there :slight_smile:

are you going to join @sayko @Blanche @Michal_w @tillia @piotr.czapla @radek @Gaurav85 @Emsi ?

20.00, join us using this link

1 Like

hej @wojtekcz dolaczysz do nas? czekamy :slight_smile:

1 Like

it was nice chatting with you today…

although next week there is no lecture, we will meet online anyway, same day, same time same link :slight_smile:

links to some interesting articles we discussed today:

CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning

Deep Learning , anyone can use

An annotation tool powered by active learning

Possible part 2 notebooks

Attention Is All You Need

Capsule neural network

3 Likes

I won’t probably attend because I have recruitment task for a new job (which will include python and possibly some ML), so wish me luck :smiley: Hopefully I will have some time for fast.ai in the second half of December.

4 Likes

There are two interesting things to follow up. Math course and the ML course I’m thinking about Math by Rachel.

Can you post it again, I’ve tried once to join and it seems it was old link as I haven’t notice anyone.

please use the link that is in the top post

did you see or planning to watch any interesting talks at NeurIPS this year? live stream.

yeah both courses look great and if can add kaggle to the mix would be perfect :slight_smile:

So we have 2 free spots on team for STX next hackaton next friday in Poznań, anyone wanna join?

1 Like

thx for the offer but i would have to pass. but please share results :).

@piotr.czapla @Michal_w @Blanche @wojtekcz @tillia @sayko

link for today’s virtual meetup at 20.00

apologies but i will not be able to make it today, but the link above will be ready for you!

are you interested to continue the virtual meetups in the future? to practice the part 1 material and get ready for part 2? we could meet 1/month, 1/3-weeks, 1/2-weeks or so, let me know!