Dan Ganea

Staff Engineer @Fivetran in the SQL Server team. Ex-Kotlin backend, Ex-C++ embedded. Former ML researcher. Published a paper, wrote path-tracers. Coffee-brewer.

[Dan Ganea @thedanganea]

@antirez Yes, the same process of discovering new mathematics can still happen.

There's two reasons why it's debatable whether or not knowing the truth/proof is detrimental to that:
- It provides a known 'path' already.
- Less of a carrot for mathematicians to prove smth.

[Dan Ganea @thedanganea]

@doodlestein Ah, yes. From the creator of “SIMD-accelerated guard that blocks rm -rf and git reset --hard”

[Jean-Baptiste Cordonnier @jb_cordonnier]

We added the visualisation of the attention patterns of Vision Transformer!

https://t.co/zKZGRLCTmp

Some heads learn translation equivarient attention to extract patches at fixed shifts. Other heads rely on color similarity or maybe more semantic features deeper in the network.

[Vivek Verma @vcubingx]

IT'S FINALLY OUT!

After months of work, I'm happy to announce that the first video in a series on visualizing deep learning is out. It's on visualizing the structure of a neural network.

RTs appreciated :)

https://t.co/oXuOPuN0AV

[Erik Bekkers @erikjbekkers]

I recently gave an intro to group equivariant CNNs at Mathematics of Machine Learning (https://t.co/bevBEx9yY6).

Thank you Philip Aston, @MJEhrhardt, Catherine Higham, & Clarice Poon for organizing!

The talk can be rewatched here:
https://t.co/fkKnnhmCOi
Hope you like it! :)

[Programming Wisdom @CodeWisdom]

✨ More Woz wisdom: "Your first projects aren’t the greatest things in the world, and they may have no money value, they may go nowhere, but that is how you learn – you put so much effort into making something right if it is for yourself."

[Justin Johnson @jcjohnss]

Videos for my Fall 2019 course "Deep Learning for Computer Vision" are now on YouTube!

This is an evolution of @cs231n that I used to teach at Stanford:
- All content refreshed
- New topics: Transformers, Video, 3D, etc
- HW in @PyTorch + @GoogleColab

https://t.co/6nqZKTpmxv

[Reza Zadeh @Reza_Zadeh]

A basic introduction to Attention, Transformers, & GPT, by @ilyasut, who considers Attention as the most important neural architectural advance since the LSTM. https://t.co/0IVL81QA5n

[Dan Ganea @thedanganea]

Every time I think there couldn't possibly be a medium where Doom isn't ported ...

[John Carmack @ID_AA_Carmack]

Someone noticed that when you have hundreds / thousands of cores in a supercomputer, the individual utilization boxes in Task Manager start to look like pixels. People started making pictures by doing different amounts of work on specific processors.

It escalated quickly.

[elvis @omarsar0]

📺 Transformers Explained 📺

A fantastic and nicely animated explanation of Transformers.

Love the explanations and examples. It makes everything so easy to understand.

by Letitia Parcalabescu

https://t.co/8oFsJntsJA

[Robert Lange @RobertTLange]

Great tutorial on Meta-Learning by @yeewhye covering optimisation-based, black-box & a probabilistic perspective on learning task invariances at #MLSS2020. Re-watch the videos here:

📺(Part I): https://t.co/bHQVIS8Y8j
📺(Part II): https://t.co/OrKMpxeYYW

[Anton Kaplanyan @Kaplanyan]

An extremely well-written post on radimetric quantities and lighting with great interactive illustrations https://t.co/5STBgwERtJ

[Yannic Kilcher 🇸🇨 @ykilcher]

New Video 🔥 Slot Attention is a module that can be built into any pipeline to create an N-to-1 assignment of a set of features to slots. Ideal for object discovery / classification.
https://t.co/PxZA4zShBU
@FrancescoLocat8 @thomaskipf @dirkweissenborn @TomUnterthiner @kyosu

[Robert Lange @RobertTLange]

The lottery ticket hypothesis 🎲 states that sparse nets can be trained given the right initialisation 🧬. Since the original paper (@jefrankle & @mcarbin) a lot has happened. Checkout my blog post for an overview of recent developments & open Qs.

✍️: https://t.co/R46SCEYf8p

[Nando de Freitas @NandoDF]

This brilliant ⁦@OpenAI⁩ work and the video of ⁦@karpathy⁩ I shared recently are very exciting AI frontiers. The story repeats itself: Big net, curated data, and common sense are the ingredients. Congrats ⁦@ilyasut⁩ et al. https://t.co/6gTC96VKfu

[François Chollet @fchollet]

Interested in learning how to build OCR models? Here's a great code example from @A_K_Nain showing a model that can break Captchas. https://t.co/dmXDCNEppI

[Karan Desai (KD) @kdexd]

Introducing "VirTex": a pretraining approach to learn visual features via language using fewer images.
Pretrain: CNN+Transformer from scratch on COCO Captions.
Transfer CNN: Results on 6 vision tasks match/exceed ImageNet pretraining (10x size wrt COCO)!
https://t.co/A3F00jmT9N

[Laura Leal-Taixe @lealtaixe]

Our series of lectures for the Advanced Deep Learning in Computer Vision course @TU_Muenchen with @MattNiessner is now public! Learn about graph NN, neural rendering, metric learning, GANs, and more! https://t.co/IKGWRFp4ty

[Jeremy Howard @jeremyphoward]

"If you want to learn deep learning by example, from a team that has seemingly written the book on it, Deep Learning for Coders with fastai and PyTorch is for you."

Thanks @mattmayo13 @kdnuggets! 😊
https://t.co/K3d64GStXc

[Zachary Lipton @zacharylipton]

Super excited to announce that “Dive Into Deep Learning” now officially supports PyTorch! https://t.co/V7ZyrjS0hY