Italian Virtual Study Group

Ciao,
It would be great to create a virtual study group and maybe even organize a meetup/event after the quarantine :wink:
Did you prefer Teams/Slack/Zoom/Telegram… other?

5 Likes

It would be great to keep everything here as main conversation, so also other can see IMHO. But whatever it will work is fine for me, in what projects are you going to work on?

2 Likes

I am exploring/working on different problems/projects. Some for my work and others for my curiosity.
For example: camera pose, action recognition, body tracking, NLP, recommendation systems…
My job is my passion :wink:

2 Likes

Ciao a tutti,
let’s try to keep stuff here. Then let’s see if other options could add value.

I’m working in R&D dept on different projects and PoCs… some are action recognition in videos, object detection and tracking, NLP on text and voice applications, as well as VR/MR and volumetric videos.

3 Likes

I’m interested to join and follow along.

3 Likes

Hi all, nice to meet you!

1 Like

Hello everybody!

1 Like

Hi buddies! :wink:

1 Like

Are we planning to create a Zoom Channel or group?! It would be great

2 Likes

Hello guys. I created a Zoom public Channel if interested.
The name is:

Fast.ai - Italian Virtual Study Group

3 Likes

Happy to see many of the old ones here! And welcome to the new ones!! Nice the idea of Zoom, we should try it …

1 Like

Hello guys!
It would be interesting to see if we can contribute to the following thing.

I don’t think they release as “open-source” any anonymous data, but it would make sense to try to ask.
What do you think?

1 Like

I am trying to find more details about this project.
It isn’t clear if they use an existing product or not. In any case, should be great to have access to a labeled image dataset.

1 Like

From a small research it seems that the software is a product by the chinese company infervision.

Partendo dall’analisi delle immagini della TC polmonare l’algoritmo di intelligenza artificiale, elaborato in Cina a partire da una base di 1000 casi clinici di pazienti affetti da Covid-19, è in grado rispondere in appena 20 secondi.

The so-called “Infer-read” system is based on technology developed by a Chinese company called InferVision and adapted for use in Italy. Rome’s Polyclinic Bio-Medical University purchased the license for the system and InferVision engineers based in Germany traveled to Rome to train staff on its use.

References:
http://www.china.org.cn/world/Off_the_Wire/2020-03/23/content_75850266.htm

2 Likes

Ciao a tutti! Come va questo study group, vi riunite ogni tanto? @danielegaliffa io ho un account free su Zoom…ho provato a cercare Fast.ai - Italian Virtual Study Group fra i channel ma non trovo nulla. Forse serve un account a pagamento? E se usassimo Slack invece?

1 Like

Ciao! I was wondering (since this group has many awesome Italian speakers) any examples / code links in how to train an NLP classifier for a text in italian? Grazie mille :smiling_face_with_three_hearts:

1 Like

Ciao scusami per il delay nel rispondere, ma ho visto solo ora la notifica del tuo messaggio. Perdonami.

Purtroppo non ci siamo ancora mai riuniti :frowning:

In merito al canale di Zoom, non lo trovo più nemmeno io tra i miei canali…boh :frowning:

Ad ogni modo, se hai altre modalità più smart/snelle (es: slack) va benissimo!
Appena hai pronto il canale, let us know così proviamo a riunirci!

Grazie mille!

2 Likes

Ciao Daniele!

Grazie per la risposta ma ormai il corso è praticamente finito…mi sa che abbiamo perso l’occasione. Peccato, erano 4 anni che volevo partecipare a questo corso, e quando finalmente ero riuscito a trovare un pò di tempo, il mondo è cambiato. Si vede che non era destino!

To train a classifier you need an Italian LM trained on a large corpus. Once you have it, the procedure (ULMFit workflow) is the same of https://github.com/fastai/fastai2/blob/master/nbs/examples/ulmfit.ipynb there are few things that are language specific (ie: the tokenizer if you use a language specific one like spacy that, at the time, was the fastai default).

I’ve trained an Italian LM from scratch on the whole Italian wikipedia (several GB of text) a couple of years ago: pretty sure that’s no more compatible with the last version of the library.
It was a fun experience: AFAIR it took about 5 days of training, and force me to write the algorithm as a script in order to properly check the health of such a long process.

1 Like