Channel Title: Arxiv Insights Channel Description: My name is Xander Steenbrugge, and I read a ton of papers on Machine Learning and AI. But papers can be a bit dry & take a while to read. And we are lazy right? In this channel I try to summarize my core take-aways from a technical point of view while making them accessible for a bigger audience. If you love technical breakdowns on ML & AI but you are often lazy like me, then this channel is for you! <video> OpenAI Five: When AI beats professional gamers Description: In this episode I discuss OpenAI Five, a Machine Learning system that was able to defeat professional gamers in the popular video game Dota 2: - How was the system built? - What does this mean for AI progress? - What real world applications can be built on this succes? You can find all the OpenAI blogposts here: https://blog.openai.com/ If you enjoy my videos, all support is super welcome! https://www.patreon.com/ArxivInsights If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge </video> <video> Reinforcement Learning with sparse rewards Description: In this video I dive into three advanced papers that addres the problem of the sparse reward setting in Deep Reinforcement Learning and pose interesting research directions for mastering unsupervised learning in autonomous agents. Papers discussed: Reinforcement Learning with Unsupervised Auxiliary Tasks - DeepMind: https://arxiv.org/abs/1611.05397 Curiosity Driven Exploration - UC Berkeley: https://arxiv.org/abs/1705.05363 Hindsight Experience Replay - OpenAI: https://arxiv.org/abs/1707.01495 If you want to support this channel, here is my patreon link: https://patreon.com/ArxivInsights --- You are amazing!! ;) If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge </video> <video> 'How neural networks learn' - Part II: Adversarial Examples Description: In this episode we dive into the world of adversarial examples: images specifically engineered to fool neural networks into making completely wrong decisions! Link to the first part of this series: https://youtu.be/McgxRxi2Jqo If you want to support this channel, here is my patreon link: https://patreon.com/ArxivInsights --- You are amazing!! ;) If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge </video> <video> An introduction to Policy Gradient methods - Deep Reinforcement Learning Description: In this episode I introduce Policy Gradient methods for Deep Reinforcement Learning. After a general overview, I dive into Proximal Policy Optimization: an algorithm designed at OpenAI that tries to find a balance between sample efficiency and code complexity. PPO is the algorithm used to train the OpenAI Five system and is also used in a wide range of other challenges like Atari and robotic control tasks. If you want to support this channel, here is my patreon link: https://patreon.com/ArxivInsights --- You are amazing!! ;) If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge Links mentioned in the video: ⦁ PPO paper: https://arxiv.org/abs/1707.06347 ⦁ TRPO paper: https://arxiv.org/abs/1502.05477 ⦁ OpenAI PPO blogpost: https://blog.openai.com/openai-baselines-ppo/ ⦁ Aurelien Geron: KL divergence and entropy in ML: https://youtu.be/ErfnhcEV1O8 ⦁ Deep RL Bootcamp - Lecture 5: https://youtu.be/xvRrgxcpaHY ⦁ RL-adventure PyTorch implementation: https://github.com/higgsfield/RL-Adventure-2 ⦁ OpenAI Baselines TensorFlow implementation: https://github.com/openai/baselines </video> <video> Variational Autoencoders Description: In this episode, we dive into Variational Autoencoders, a class of neural networks that can learn to compress data completely unsupervised! VAE's are a very hot topic right now in unsupervised modelling of latent variables and provide a unique solution to the curse of dimensionality. This video starts with a quick intro into normal autoencoders and then goes into VAE's and disentangled beta-VAE's. I aslo touch upon related topics like learning causal, latent representations, image segmentation and the reparameterization trick! Get ready for a pretty technical episode! Paper references: - Disentangled VAE's (DeepMind 2016): https://arxiv.org/abs/1606.05579 - Applying disentangled VAE's to RL: DARLA (DeepMind 2017): https://arxiv.org/abs/1707.08475 - Original VAE paper (2013): https://arxiv.org/abs/1312.6114 If you want to support this channel, here is my patreon link: https://patreon.com/ArxivInsights --- You are amazing!! ;) If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge </video> <video> The Molecular Basis of Life Description: If you want to support this channel, here is my patreon link: https://patreon.com/ArxivInsights --- You are amazing!! ;) If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge -------------------------------- Life is a molecular marvel of astounding complexity. In this video we take a dive into the world of molecular engines, proteins and the physical processes that power life on earth. This video serves as an introduction to molecular biology and a primer for the future videos that will cover Machine Learning applications in computational biology and protein design. ::Chapters:: 00:00 Intro 02:37 What are Proteins, and why should I care? 04:10 Getting a sense of scale 07:39 From DNA to Proteins 12:17 From Structure to Function 17:23 The Coronavirus 19:39 Application Potential of AI assisted computational Biology 22:11 Pensight and Patreon Links </video> <video> Editing Faces using Artificial Intelligence Description: Link to Notebooks: https://drive.google.com/open?id=1LBWcmnUPoHDeaYlRiHokGyjywIdyhAQb Link to the StyleGAN paper: https://arxiv.org/abs/1812.04948 Link to GAN blogpost: http://hunterheidenreich.com/blog/gan-objective-functions/ If you want to support this channel, here is my patreon link: https://patreon.com/ArxivInsights --- You are amazing!! ;) If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge -------------------------------- This episode covers one of the greatest ideas in Deep Learning of the past couple of years: Generative Adversarial Networks. I first explain how a generative adversarial network (GAN) really works. After this general overview, we go into the specific objective function that is optimized during training. We then dive into Nvidia's StyleGAN model and learn how we can manipulate it's latent space to morph arbitrary images of faces. This video comes with a complete Google Colab notebook to reproduce & play with all the examples shown in the video! ::Chapters:: 00:00 Intro 02:55 Video overview 03:35 Introduction to GANs 05:40 5 min Deepdive on the Training Objective for GANs 10:07 State-of-the-art GAN techniques: StyleGAN 14:40 Manipulating the latent space of GANs </video> <video> An introduction to Reinforcement Learning Description: This episode gives a general introduction into the field of Reinforcement Learning: - High level description of the field - Policy gradients - Biggest challenges (sparse rewards, reward shaping, ...) This video forms the basis for a series on RL where I will dive much deeper into technical details of state-of-the-art methods for RL. Links: - "Pong from Pixels - Karpathy": http://karpathy.github.io/2016/05/31/rl/ - Concept networks for grasp & stack (Paper with heavy reward shaping): https://arxiv.org/abs/1709.06977 If you enjoy my videos, all support is super welcome! https://www.patreon.com/ArxivInsights If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge ::Chapters:: 00:00 Intro 01:03 So what is Reinforcement Learning? 03:39 Learning without explicit examples 07:25 Main challenges when doing RL 15:04 Are the robots taking over now? </video> <video> 'How neural networks learn' - Part I: Feature Visualization Description: Interpreting what neural networks are doing is a tricky problem. In this video I dive into the approach of feature visualisation. From simple neuron excitation to the Deep Visualisation Toolbox and the Google DeepDream project, let's open up the black box! Links: Distill.pub post on Feature Visualisation: https://distill.pub/2017/feature-visualization/ Sander Dieleman post on music recommendation: http://benanne.github.io/2014/08/05/spotify-cnns.html Blogpost on Deep Feature visualisation: http://yosinski.com/deepvis Github link to DeepVis Toolbox: https://github.com/yosinski/deep-visualization-toolbox Paper by Zeiler & Fergus: https://arxiv.org/abs/1311.2901 If you want to support this channel, here is my patreon link: https://patreon.com/ArxivInsights --- You are amazing!! ;) If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge </video> <video> AlphaGo - How AI mastered the hardest boardgame in history Description: In this episode I dive into the technical details of the AlphaGo Zero paper by Google DeepMind. This AI system uses Reinforcement Learning to beat the world's Go champion using only self-play, a remarkable display of clever engineering on the path to stronger AI systems. DeepMind Blogpost: https://deepmind.com/blog/alphago-zero-learning-scratch/ AlphaGo Zero paper: https://storage.googleapis.com/deepmind-media/alphago/AlphaGoNaturePaper.pdf If you want to support this channel, here is my patreon link: https://patreon.com/ArxivInsights --- You are amazing!! ;) If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge </video> <video> AlphaFold and the Grand Challenge to solve protein folding Description: If you want to support this channel, here is my patreon link: https://patreon.com/ArxivInsights --- You are amazing!! ;) If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge -------------------------------- AlphaFold is DeepMinds latest breakthrough addressing the protein folding problem. Using an advanced Deep Learning architecture that achieves end-to-end learning of protein structures, this work is arguably one of the most influential papers of this decade and is likely to spark enormous advanced in computational biology and protein design. This video covers the entire architecture of the model as well as training principles that led to the incredible results of AlphaFold2! AlphaFold Nature paper: https://www.nature.com/articles/s41586-021-03828-1 AlphaFold Codebase: https://github.com/deepmind/alphafold Work from the Baker lab: https://www.bakerlab.org/ Fabian Fuchs' amazing blog on equivariance: https://fabianfuchsml.github.io/alphafold2/ Ongoing Open Source effort to reproduce AlphaFold: https://github.com/lucidrains/alphafold2 ::Chapters:: 00:00 Intro 02:28 The Protein Folding Problem 05:29 AlphaFold1 revisited 06:10 Multiple Sequence Alignments (MSA) 08:10 Distograms 12:29 AlphaFold2 14:52 The Evoformer 19:07 The Structure Module 28:13 Zooming out: looking at the future </video> <video> Why humans learn so much faster than AI Description: - Link to edited game versions: https://rach0012.github.io/humanRL_website/ - Link to the Paper: https://openreview.net/pdf?id=Hk91SGWR- "Why are humans such incredibly fast learners?" This is the core question of this paper. By leveraging powerful prior knowledge about how the world works, humans are able to quickly figure out efficient strategies in new and unseen environments. Current state-of-the-art Reinforcement Learning algorithms however, usually don't have strong priors and this is one of the fundamental challenges in current research on Transfer Learning. If you want to support this channel, here is my patreon link: https://patreon.com/ArxivInsights --- You are amazing!! ;) If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge </video> <video> 'How neural networks learn' - Part III: Generalization and Overfitting Description: In this third episode on "How neural nets learn" I dive into a bunch of academical research that tries to explain why neural networks generalize as wel as they do. We first look at the remarkable capability of DNNs to simply memorize huge amounts of (random) data. We then see how this picture is more subtle when training on real data and finally dive into some beautiful analysis from the viewpoint on information theory. Main papers discussed in this video: First paper on Memorization in DNNs: https://arxiv.org/abs/1611.03530 A closer look at memorization in Deep Networks: https://arxiv.org/abs/1706.05394 Opening the Black Box of Deep Neural Networks via Information: https://arxiv.org/abs/1703.00810 Other links: Quanta Magazine blogpost on Tishby's work: https://www.quantamagazine.org/new-theory-cracks-open-the-black-box-of-deep-learning-20170921/ Tishby's lecture at Stanford: https://youtu.be/XL07WEc2TRI Amazing lecture by Ilya Sutkever at MIT: https://youtu.be/9EN_HoEk3KY If you want to support this channel, here is my patreon link: https://patreon.com/ArxivInsights --- You are amazing!! ;) If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: https://pensight.com/x/xander-steenbrugge </video>

Contacts and Links

Description

My name is Xander Steenbrugge, and I read a ton of papers on Machine Learning and AI.
But papers can be a bit dry & take a while to read. And we are lazy right?

In this channel I try to summarize my core take-aways from a technical point of view while making them accessible for a bigger audience.

If you love technical breakdowns on ML & AI but you are often lazy like me, then this channel is for you!

Videos from channel