Portfolio

🎙️ Interview with David Stutz from Google DeepMind at #HLF23

When @GoogleAI scaled machine translation to 1000 languages with monolingual data, they wrote a 57 long paper about it https://arxiv.org/abs/2205.03983. Ms. Coffee Bean summarized it 😊: 📺 https://y...

Ms. Coffee Bean’s take on DALL-E 2’s secret “language”. 📺 https://youtu.be/MNwURQ9621k Short summary of current opinions and theories on why DALL-E 2 can create sensible pictures out of gibberish t...

Imagen from Google Brain 🧠 is competing with DALLE-2 when it comes to generating amazing images from just text! Here is an overview of Imagen and other diffusion-based text-to-image generators such...

Special AI ☕ break edition about exciting developments in mathematical deep learning – inspired by physics. 🔥 📺 https://youtu.be/m2bXL5Z5CBM Make sure to check it out if you want a quick summary ov...

Let’s VALSE 💃! We present our own work on the VALSE benchmark testing large pretrained vision and language models by targeting linguistic phenomena. Work accepted at ACL 2022. https://youtu.be/huml...

It’s time for a coffee break! ☕ We’re explaining PaLM, Google AI’s Pathways Language model. 👇 Yes, it’s that agglomeration of 540 billion parameters that can explain jokes. 😏 📺 https://youtu.be/yi-...

The 10 billion parameter SEER model from @MetaAI is *fairer*, even though it is trained on *uncurated* data. How so? 😕 Check out our take on this. 👇 https://youtu.be/XHAoV_nKr1o

Regularization in deep learning, Lipschitz continuity, gradient regularization, adversarial defense, gradient penalty. These were topics of our daily Quiz questions! Tim Elsner (@daidailoh on Twitt...

Diffusion models beat GANs in image synthesis, GLIDE generates images from text descriptions, surpassing even DALL-E in terms of photorealism! Check out this video to learn how diffusion models wor...

We were invited to the Machine Learning Street Talk podcast! We talked about linguistics, symbolic AI and our YouTube journey. 🙌 Check out the episode: 👉 https://youtu.be/p2D2duT-R2E

Multi-head self-attention (MSA) and convolutions are complementary, or so it turns out. So what are the differences between MSA and Convs? In this video, we present the lessons of the recent "How d...

📺 https://youtu.be/gzNlfvu1hqo Announcement: ☕⚔️🍵 AMA with AI Coffee Break & Chai Time Data Science (Sanyam Bhutani) over at Weights & Biases . Happening Saturday, the 26th of Feb. at 5 PM CET / 8 ...

Can a ConvNet outperform a Vision Transformer? What kind of modifications do we have to apply to a ConvNet to make it as powerful as a Transformer? Spoiler: it’s not attention. This is our animated...

What a Christmas surprise: 🎄 Tonight, we surpassed 10,000 subscribers on YouTube! 🤯 https://www.youtube.com/aicoffeebreak Ms. Coffee Bean is still drunk (with joy). Thanks to our wonderful viewers ...

Why would one build a transformer to solve linear algebra problems when there is numpy.linalg? 😱 📺 Check out the video to find out why F. Charton trained transformers to solve #LinearAlgebra proble...

Want to know about our conversation about #AI (mis)conceptions at @_eavi? Then check out📺 the video recording: 👉 https://youtu.be/XI5pvWzg_N4 Or read the textual summary: https://eavi.eu/ai-for-goo...

“Masked Autoencoders Are Scalable Vision Learners” paper explained and animated! 📺 Say goodbye to contrastive learning and say hello (again) to autoencoders in #ComputerVision! Love the simple, ele...

What could go wrong when comparing models based on the number of parameters? Size does not matter | The efficiency misnomer | What does the number of parameters mean? 👉 https://youtu.be/l3bTCP6x-FM

Do Transformers process sequences of FIXED or of VARIABLE length? | #AICoffeeBreakQuiz question explained. 👉 https://youtu.be/Xxts1ithupI

A confused Coffee Bean summarizes the generalization – interpolation – extrapolation debate in #MachineLearning of Yann LeCun, Gary Marcus, Francois Chollet and others. Her take on the discussion s...

SimVLM paper explained | What the paper doesn't tell you 👉 https://youtu.be/XU4ZHUnnbpg

Ms. Coffee Bean explains, visualizes, and comments on the sobering take about 📊 NLU Benchmarking by S. Bowman and G. Dahl. Check out the short, animated video: 👇 https://youtu.be/W57u1j16iC8

🔥 NEW video! 🔥 Swin Transformer paper explained, visualized, and animated by Ms. Coffee Bean. Let us know what you think. We’re a little rusty after our vacation. 📺 Check out the short video: 👉 htt...

How to know if a face was generated by a (Style)GAN(2)? 📜 We discuss “Eyes Tell All: Irregular Pupil Shapes Reveal GAN-generated Faces.” by Guo et al. 2021. 📺 If you're interested in cues for spott...

How do modern search engines work? Find out by watching this short animated video. 👇 📺 https://youtu.be/YkK5IKgxp-c Best enjoyed during your Coffee Break! ☕

Sound the opinionated video alarm! 🚨 Ms. Coffee Bean and her PhD student debate their different opinions triggered by the “Foundation Models” paper. 📺 On the opportunities and risks of calling pre-...

What are tokenizers good for? Why do transformers need them? Find out in this short animated video! 📺 👇 https://youtu.be/D8j1c4NJRfo

Our submission to the Veritasium Science Communication contest! The shortest explanation about the today’s capabilities of artificial intelligence 🧠, or rather “When is your iPhone smarter than Ms....

How can data leakage happen during data preparation? Find out by joining Ms. Coffee Bean’s visualization 📺 of the answer to this question. 👉 https://youtu.be/8_UBDTHAHqY You can find the original s...

We had a great discussion with Prof. Melanie Mitchell! For an honest, down-to-earth view on the limits of (artificial) intelligence, check out the latest Machine Learning Street Talk episode: 👉 htt...

In the previous video, we explained the basics of positional embeddings with sines and cosines as they were introduced in the "Attention is all you need" paper. In our latest video, we dive even fu...

Positional embeddings in transformers explained and visualized! Check out the video here 👇 https://youtu.be/1biZfFLPRSY

📺 Relative position representations explained by Ms. Coffee Bean. Learn about them as they were introduced in the "Self-Attention with Relative Position Representations." paper by S. Peter, J. Uszk...