Study Group in Italy

Andrea. Glad to meet yā€™all! :wink:

So if Iā€™m not making mistakes, weā€™ll have the first lesson at 3 oā€™clock, October 23. Am I right? :thinking:

Like I said previously, good idea! We could use the EU study group for everything but local meetups.

Hi everybody,
this is Riccardo from Italy too.

1 Like

Hello everyone!!!

This is Federico, I am Italian but I am now living in Finland!!!

Hi all! Here Iā€™amā€¦ Andrea from Milan, Italy. letā€™s roll :slight_smile:

Hello guys, did you attend the live class this ā€œmorningā€?

This year I canā€™t make it (job hours are not that compatible with waking up in the middle of the night), but if you can I strongly suggest following live.
Itā€™s a completely different experience!

1 Like

I did, just get up :wink:
@DavideBoschetto : totally agree!

What about trying to solve a classification problem as study group to practice with new library?
Could be a Kaggle dataset or competition:

About 70mb:

About 116MB:

Other ā€œsmallā€ (and usually fast to train and to make experiments) dataset for image classification to pick:
https://www.kaggle.com/datasets?sortBy=hottest&group=public&page=1&pageSize=20&size=medium&filetype=all&license=all&tagids=13302%2C14102

ā€¦ On small datasets like that you can even practice with CPU!

I attended too!
Now at workā€¦ :thinking:

I would try to focus on specific real-problems/issues that, if solved, could be of direct and immediate benefit of those in need.

For instance:
https://www.synapse.org/#!Synapse:syn4224222/wiki/401743

More over, if we can apply our learnings as a group to specific problems, we can also double-check what the state of the art currently states.

What do you think guys?

1 Like

Other available resources on the same topic:

Full data-set:

1 Like

Iā€™m in! Choose the one you like.

BTW: Iā€™ve used fast.ai v1 to this old competition yesterday and got very good results at the first try: 32 errors over 4462 in the validation set.

Hi guys, Iā€™m in too.

1 Like

Ciao @ste,
thanks for the reply!

I found the histological data-set for breast cancer interesting for several reasons:

  1. It seems that images normalization used in the paper uses a different criteria than the standard one proposed by @jeremy in the class

  2. It seems balanced

  3. There are state-of-the-art metrics

  4. Can provide an impact to health of women

It would be great to run this group-session ā€œliveā€ (via hangout?)

Thanks,
d.

2 Likes

Great! Letā€™s find a way to collaborate on the same dataā€¦
Anyone would join us?

In the past Iā€™ve tried to participate to ā€œSkin Lesion / Melanoma Detectionā€ previous editions, but didnā€™t have so much time and understanding of mixing CNN & Tabular dataā€¦

https://challenge2018.isic-archive.com/

Hello guys, I collected all my learning/questions in a class-lesson related thread here:

https://forums.fast.ai/t/lesson-1-chat/27332/948?u=danielegaliffa

What are your learnings so far?

Thanks

1 Like

@danielegaliffa regarding to your last question, Iā€™ll suggest to take a look at @devforfu and @gianferrarif work:

https://forums.fast.ai/t/how-to-get-an-empty-convlearner-for-single-image-prediction/28245/49

Nice article @gianferrarif !

Way better than Jian Yangā€™s hot dog detector :wink:

2 Likes

Hello, I have just launched a discussion thread on fastai study groups to gather feedback from organizers and participants to identify best practices and avoid some gaps.

You will find more information in this post. Thank you if you can take a few minutes to participate in the discussion :slight_smile:

1 Like