Facebook has open sourced their PyTorch based Natural Language Processing modeling framework. According to them it: 👇

blurs the boundaries between experimentation and large-scale deployment.

Looking forward to trying this out. 🤓

Src: Facebook

Do The Digital Worm 🐛

Step 1: recreate the brain of the C. elegans worm as a neural network 🧠

Step 2: ask it to park a car 🚗

Researchers digitized the worm brain, the only fully mapped brain we have, with a 12 neuron network. The goal of this exercise was to create a neural network that humans can understand and parse since the organic version it is based on is well understood. 🗺

An interesting realization that came out of this exercise: 👇

Curiously, both the AI model and the real C. elegans neural circuit contained two neurons that seemed to be acting antagonistically, he said—when one was highly active, the other wasn’t.

I wonder when this switching neuron feature will be rolled into an AI/ML/DL architecture. 🤔

Src: Motherboard

DeepFakes Get More Realistic 😖

Remember back when I said I was terrified about deepfakes? Well, it’s not getting any better. 😟

Apparently researchers at Carnegie Mellon and Facebook’s Reality Lab decided there is nothing to worry about and the method for making them needed to be better. So they give us Recycle-GAN. ♻️

We introduce a data-driven approach for unsupervised video retargeting that translates content from one domain to another while preserving the style native to a domain, i.e., if contents of John Oliver’s speech were to be transferred to Stephen Colbert, then the generated content/speech should be in Stephen Colbert’s style.

Fantastic. Just what we need. A system that transfers content while maintaining stylistic integrity all while not needing a great deal of tweaking/input to make it happen. 😵

Also, why is Facebook helping to make fake content easier to create? Don’t they have enough problems on this front already? 🤔

Src: Carnegie Mellon

ELMo Really Does Know His Words 👹

I’m super interested in the world of NLP (natural language processing), so the news that performance increased dramatically with ELMo piqued my interest. 💡

The biggest benefit in my eyes is that this method doesn’t require labeled data, which means the world of written word is our oyster. 🐚

Yeah, yeah, word embeddings don’t require labeled data either. ELMo can also learn word meanings at a higher level, which I think means it will have far more impact and a wider range of applications. 📶

ELMo, for example, improves on word embeddings by incorporating more context, looking at language on a scale of sentences rather than words.

Our cuddly, red muppet still picks up the biases we embed in our writings though. So plenty more work to be done. 🛠

Src: Wired

(Compute) Size Doesn’t Matter 📏

Fast.ai was recently part of a team that set some new speed benchmarks on the ImageNet image recognition data set. Why is this noteworthy? Because they did it on an AWS instance that cost $40 total. 🏅

We entered this competition because we wanted to show that you don’t have to have huge resources to be at the cutting edge of AI research, and we were quite successful in doing so. We particularly liked the headline from The Verge: “An AI speed test shows clever coders can still beat tech giants like Google and Intel.”

Jeremy Howard

AI has a mystique about it that makes it seem like it’s only for uber-tech nerds that love math and have access to the biggest computers, but that’s not true. Yes it’s technical, but it’s not impossible. And there are plenty of resources to help those curious get started. It is not nearly as difficult as it seems. We just have a lot of language and storytelling baggage attached to it. ⚗️

Very few of the interesting ideas we use today were created thanks to people with the biggest computers. And today, anyone can access massive compute infrastructure on demand, and pay for just what they need. Making deep learning more accessible has a far higher impact than focusing on enabling the largest organizations..

Jeremy Howard

Src: fast.ai

When Gradients Explode (or Vanish) 💥

This is a nice quick read on how to combat exploding or vanishing gradients, a problem that wreak havoc on your deep learning model. 👹


  • Exploding gradient? Use gradient clipping. It sets a ceiling on your gradient but keeps the direction. ✂️
  • Vanishing gradient? If you’re using an RNN, use an LSTM. ✔️

Src: Learn.Love.AI.

Fakes Get More Real 🎥

This is why I’m terrified of DeepFakes! But also, think of the potential for visual artistic mediums like movies and TV. But also, think of the implications for politics. I find it fitting that they used a lot of political figures as examples since this could majorly disrupt the field. 🗳️

My first concern was in regards to detection, especially since this method seems to solve the blinking problem that SUNY’s detection method was able to capitalize on. I wondered if this technique would generate noise and irregularities that would aid in detection, which the error section of the video suggests is the case. Here is what the SIGGRAPH team notes in relation to my concerns:

Misuses: Unfortunately, besides the many positive and creative use cases, such technology could also be misused. For example, videos could be modified with malicious intent, for instance in a way which is disrespectful to the person in a video. Currently, the modified videos still exhibit many artifacts, which makes most forgeries easy to spot. It is hard to predict at what point in time such modified videos will be indistinguishable from real content to our human eyes. However, as we discuss below, even then modifications can still be detected by algorithms.

Implications: As researchers, it is our duty to show and discuss both the great application potential, but also the potential misuse of a new technology. We believe that all aspects of the capabilities of modern video modification approaches have to be openly discussed. We hope that the numerous demonstrations of our approach will also inspire people to think more critically about the video content they consume every day, especially if there is no proof of origin. We believe that the field of digital forensics should and will receive a lot more attention in the future to develop approaches that can automatically prove the authenticity of a video clip. This will lead to ever better approaches that can spot such modifications even if we humans might not be able to spot them with our own eyes (see comments below).

Detection: The recently presented systems demonstrate the need for ever improving fraud detection and watermarking algorithms. We believe that the field of digital forensics will receive a lot of attention in the future. Consequently, it is important to note that the detailed research and understanding of the algorithms and principles behind state-of-the-art video editing tools, as we conduct it, is also the key to develop technologies which enable the detection of their use. This question is also of great interest to us. The methods to detect video manipulations and the methods to perform video editing rest on very similar principles. In fact, in some sense the algorithm to detect the Deep Video Portraits modification is developed as part of the Deep Video Portraits algorithm. Our approach is based on a conditional generative adversarial network (cGAN) that consists of two subnetworks: a generator and a discriminator. These two networks are jointly trained based on opposing objectives. The goal of the generator is to produce videos that are indistinguishable from real images. On the other hand, the goal of the discriminator is to spot the synthetically generated video. During training, the aim is to maintain an equilibrium between both networks, i.e., the discriminator should only be able to win in half of the cases. Based on the natural competition between the two networks and their tight interplay, both networks become more sophisticated at their task.

Src: Stanford

Demand a Recount: DeepFake Edition 🗳

I’m not the only person worried about the potential impact of Deepfakes on politics (not that I claimed to be or thought I was). Apparently there is a Twitter wager about when a DeepFake political video will hit 2 million views before getting debunked. 🐦

I had mostly been thinking about the potential of these tools to be used like social media ads were in the last election. That is, until I read this quote. 👇

I think people who are just kind of having fun are, in aggregate, more dangerous than individual bad actors.

The pool of people that would use these just for fun to cause confusion and mayhem is a lot larger than the pool of people looking to use them specifically for political instability or influence. And the true danger is in the viewers seeing and believing it because very few will likely see or care about it being unveiled as fake. 🚫

Deepfake videos could exploit modern societies split by partisanship into echo chambers where information—authentic or not—tends to reinforce preexisting beliefs.