emilwallner

Hey, I'm Emil, a resident at the Google Arts & Culture Lab.

This project started out as an internal list, but once I reached a few hundred projects, I thought it could be useful for others, so I turned it into a website.

If something is missing, just comment with a link to the project or portfolio below, and I'll review it. Also happy to receive feedback on the showcase.

Hey, I'm Emil, a resident at the Google Arts & Culture Lab.

This project started as an endeavour to get an overview of what creators were doing at the intersection of creativity and machine learning.

I thought I was up to date with the latest, but my reference list kept growing and I realised I was unaware of many projects by the community.

Once I reached a few hundred, I thought it could be useful for others, so I turned my spreadsheet into a visual database with Spreadsimple.

Something missing? Just comment with a link to the project or portfolio below, and I'll review it.

More projects coming every week!!

Here are a few good starting points:

- https://pmarchive.com/guide_to_career_planning_part0.html

- https://blog.samaltman.com/how-to-be-successful

- http://www.paulgraham.com/hs.html

- https://twitter.com/naval/status/1002103360646823936

- https://www.mrmoneymustache.com/2012/12/18/your-money-or-you...

- What should I do with my life? Po Bronson [Book]

On average, most forget about 50% of what they've read after 1 hour, and only remember 2% after a week.

It sounds like your problem has more to do with retrieval. You _learn_ something when you retrieve something from your working memory or long-term memory (you build myelin, a fatty substance that isolates your synapses, which enables you to later retrieve the information).

The optimal way to learn is to use spaced repetition. It naturally happens if you apply the knowledge frequently, or you can use a flashcard app to track it.

If you struggle to _understand_ something, it often has to do with overwhelming your working memory. You are connecting several concepts that you don't understand. On average, most can keep seven items in their working memory, but after that, you'll start feeling lost.

When you internalize a few concepts by retrieving them, followed by a long break or sleep, you free your working memory to connect more concepts.

The ones I've found most effective are schools for autodidacts (40-50 around the world), most of them are part of the 42 school network (I'm at the one in Paris), but there's also Holberton School and Epitech. Then, I'd say hackerspaces or playful/collaborative co-working spaces.

Solitude. Then, I'd say acceptance from friends/family, followed by a good routine.

I've tried learning alone, but after 6 - 12 months it becomes really difficult. Being around a self-learning community daily makes a world of difference, I'm about to start my fourth year and I'm loving it.

Being part of a program also reduces societal pressure. Any program really, just to be able to give a simple answer when people ask you what you are doing. Otherwise, it's easy to be met with doubt and cynical questions.

Lastly routine, I see self-learning as a day job. You want to clock N hours per day and don't spend any time procrastinating. I've been building habits for 5-10 years, and I'm reaching a point where I can be productive, but it takes time and patience.

François Chollet’s core point: We can't measure an AI system's adaptability and flexibility by measuring a specific skill.

With unlimited data, models memorize decisions. To advance AGI we need to quantify and measure skill-acquisition efficiency.

I made a summary about it here: https://twitter.com/EmilWallner/status/1193968380967043073 Announcement tweet: https://twitter.com/fchollet/status/1192121587467784192

It's a must-read for anyone who's interested in AGI.

Is there a math equivalent to Charles Petzold's Code? I.e. a brief history of math where you feel like you invent the key steps yourself.

Awesome, looking forward to it!

Solid work - hats off!

Ha, that's an excellent idea. Someone's already started by creating a program for matrix multiplication: https://scratch.mit.edu/projects/18640517/

While the results are really cool, they're not perfect, and seemingly even the best companies working with deep learning are still not getting it right every time. It makes me wonder if this technology will ever be useful for these tasks... I mean would you be willing to use BCIs even if they made small errors from time to time?

It's not rocket science, I think you can figure it out. To get started:

git clone https://github.com/emilwallner/Screenshot-to-code-in-Keras

pip install -U floyd-cli

(login to floydhub.com, a super simple platform for cloud GPUs)

cd Screenshot-to-code-in-Keras/floydhub

floyd init picturetocode

floyd run --gpu --env tensorflow-1.4 --data emilwallner/datasets/imagetocode/2:data --mode jupyter

Then, navigate to the folder floydhub/Bootstrap/test_model_accuracy.ipynb in your Jupyter Notebook and click on Cell > Run all

Now you can see the prediction and the correct markup for all the evaluation images. Ping me on twitter: @emilwallner, if you get stuck.

This is rad, thanks for sharing it.

Output source: https://emilwallner.github.io/bootstrap/pred_1/

This approach rewrites everything from scratch.

However, it's an interesting technical problem to adjust the code. Does anyone have a rough idea how to implement it?

The bootstrap version generalizes with 97% accuracy on a new image. Because the vocabulary is limited, you can train the model overnight. To make the model generalize with all the HTML/CSS markup you need significantly more compute.

The model uses 48 tokens at a time to make the next prediction. As long as it can keep track of where it is on the design image, it’s not a problem. In the final version, it has roughly the same accuracy (~97%) on short and long websites.

The tougher part is integrating JS or adding hover effects in CSS. In theory, this can be done with an attention layer, but I haven’t seen any papers on it.

I agree, researchers tend to focus on more academic problems and industry goes where the money is, leaving some areas without innovation. I’ve come across few papers that cover frond-end development.

I'm also starting to follow people and communities that work with deep learning in new ways. Here are some of my favorites:

[1] http://colah.github.io/

[2] https://iamtrask.github.io/

[3] https://distill.pub

[4] https://experiments.withgoogle.com/ai

It could try to classify the apple tree or the context, but it would require a lot of training data. If it's out of context, it should select a color based on probability. But it's hard to solve this with just input and output data. The simple solution is to use noncontradictory training data, i.e. only having green apples.

I have an urge to teach it simple logic. Instead of making it brown, it selects the color with the highest probability from a range of colors. However, I haven't come across a deep learning implementation like this to mimic.

Yes, you can find it here: https://youtu.be/i94OvYb6noo?t=59m6s

Good point. They are all example data. The "[[3,6],[6,9],[12,18]]" can be thought of as the coordinates of a comet, and 2 is your predicted correlation, the slope, followed by 1 your predicted constant, the y-intercept. In this case, you want to change 2 and 1 to find the combination that results in the lowest error. It the same with "Error = 4^2 + (-1)^2 + 6^2", it's an example of an error function. Does that make sense?

This is something I've also struggled with. I find it hard to read deep learning papers because I need to translate each math notation, thus struggling to get the bigger picture. I'm fond of the bottom-up approach, e.g. I started by mastering C and wrote my own libraries. But for deep learning I lean towards the opposite, starting with high-level libraries. When I want to understand the theory I search for simple python code that I can implement from scratch. This way I can understand the logic, without having to understand all the math behind it. I've mostly focused on doing Kaggle type of problems and used MOOCs when I get stuck. I've had little interest from larger companies, but I've managed to get a few offers from startups. Startups often have a couple of people with PhD-level knowledge but are also looking programmers that can code the models.

Thank you Paul!!

Hey, we just finished the core war game in C. Here is some more background to the history of the game: https://blog.codinghorror.com/core-war-two-programs-enter-on... Let us know what you think and if you have any feedback. Thanks!

Yo! I'm starting to learn machine learning, and I'd love feedback on my learning roadmap. Also, do you know any fun beginner challenges like Kaggle and OpenAI request for research? Thanks

I republished it: here is the new URL: https://medium.com/life-learning/stop-learning-4aaea50d82ca#...

Thanks for your input! It's true. 'Learning' is good for reaching a specific solution, but not for achieving your long-term plans.

You should do it in Stockholm, we have produced more unicorns per capita than any other city in the world: http://techcrunch.com/2016/01/26/sweden-is-a-tech-superstar-...