[ContinualAI Meetup] Real-World Applications of Continual Learning

It’s time for the next ContinualAI Monthly Online Meetup!

This wednesday at 5 p.m. CEST, we’ll be talking about Real-World Applications of Continual Learning, you’re all invited to join (no registration required!) :smile::rocket:

As always we will start with a series of spotlight 15 minutes presentations from 4 Continual Learning experts and then have a fun, 30 minutes open debate on the topic.

Some of the questions we’d like to tackle:

:round_pushpin: Is Continual Learning ready for real-world production environments and business scenarios?
:round_pushpin: What are the best practical use-cases in which CL may excel today or in the near future?
:round_pushpin: How can we scale current research algorithms to bigger and more interesting problems?
:round_pushpin: Are we ready as businesses, research and society for the next wave of adaptive AI agents?!

:round_pushpin: Recording of the Meetup available here: https://www.youtube.com/watch?v=GteGII2e7ro

Hello everyone, I want to thank Tom Diethe, Keval Doshi and Dani Kiyasseh for their great presentations!

I just uploaded the slides for my presentation on SlideShare: https://www.slideshare.net/LorenzoPellegrini5/the-core-app-continual-object-recognition-on-the-edge


Here are the papers from my research group that will allow you to grasp the context and the main ideas behind the Continual Learning algorithm the app uses:

Rehearsal-Free Continual Learning over Small Non-I.I.D. Batches

Which describes the problem and possible solutions of dealing with a long stream of non-I.I.D. batches.

Latent Replay for Real-Time Continual Learning

Which describes the Latent Replay mechanism the app uses, along with its trade-offs, pros and cons.

I also recommend the Medium blog post from @vincenzo.lomonaco:

Trailer and Download

The app trailer is available on YouTube: https://www.youtube.com/watch?v=Bs3tSjwbHa4
(the latest version of the app is actually way faster than the one we used when creating this trailer)

You can download and install the alpha version of the application here:

The app asks for camera access (of couse) while the external memory card permission is used to store the resulting caffemodel so it can be exported and tested on a server environment. The only technical requirement is having and ARM-v8 CPU, but I’m planning to extend the support to the older ARM-v7 CPUs as well.

There is no source code yet but we’re planning to release it soon :smiley: !

Have a good day, stay safe!

1 Like

Thanks for the invite! Very interesting discussions.

Slides from my talk here:

1 Like

hey! if some interesting Applied CL papers came up in the discussion and someone wants to point them out to me, that would be nice :slightly_smiling_face:
I’m trying to add some Applied paper to my page https://github.com/optimass/continual_learning_papers/blob/master/README.md#applications

Hi @optimass,

That’s a great list - looks really useful. Here are two of our continuous learning papers that might be relevant (both available as Open Access):

They both demonstrate results on real world data. The anomaly detection one includes a benchmark with bunch of data gathered from customers (the algorithm there has been applied commercially).


1 Like

just added them!

1 Like