This AI newsletter is all you need #6

Author(s): Towards AI Editorial Team

Originally published on Towards AI.

We are excited to announce that we will partner with the MineRL competition we shared a few weeks ago and host a Q&A on the server with one of the organizers/deep mind researchers/Open AI VPG paper team involved in the MineRL competition.

We created a Discord channel for you to ask any questions to someone working at DeepMind/OpenAI and we will select the most interesting questions to ask in our next podcast and/or interview. So if you are interested in OpenAI or DeepMind, but do not know much about what it’s like to work there or even be hired there, join the conversation and ask your question!

What happened this week in AI

Tons of exciting news this week — but the one that stands out is DeepMind’s new blog post where they announced that their AlphaFold 2 model predicted structures for nearly all proteins known to science. Yes, you read that right. DeepMind expanded its protein database by over 200x to over 200 million predicted structures that other scientists will use to better understand specific proteins, what they do and how they work, accelerating scientific research and discovery globally — allowing to create further breakthroughs.

This is a big deal for the scientific community as “AlphaFold has been accessed by more than half a million researchers and used to accelerate progress on important real-world problems ranging from plastic pollution to antibiotic resistance,” now expanding this knowledge base by 200x with the model’s second version.

Hottest News

1. DeepMind predicted structures for nearly all cataloged proteins known to science
   DeepMind predicted structures for nearly all cataloged proteins known to science. It will expand the AlphaFold database by over 200x — from nearly 1 million structures to over 200 million structures.
2. DeepSpeed Compression: A composable library for extreme compression and zero-cost quantization
   Microsoft Research open-sourced DeepSpeed Compression, a framework for compression and system optimization in deep learning models, learn more here.
3. Building AI models on mobile? This may be for you!
   PyTorch open-sourced the PlayTorch app to streamline the development of mobile AI experiences.

Most interesting papers of the week

1. Translating a Visual LEGO Manual to a Machine-Executable Plan
   A novel learning-based framework, the Manual-to-Executable Plan Network (MEPNet), which reconstructs the assembly steps from a sequence of manual images, taking a LEGO manual and creating a machine executable plan that can be executed to build the target shape (see image above).
2. Audio-driven Neural Gesture Reenactment with Video Motion Graphs
   A method that reenacts a high-quality video with gestures matching a target speech audio from a video & audio source, splitting and re-assembling clips from a reference video through a novel video motion graph encoding valid transitions between clips.
3. Panoptic Scene Graph Generation
   They “introduce panoptic scene graph generation (PSG), a new problem task that requires the model to generate a more comprehensive scene graph representation (see image above) based on panoptic segmentations rather than rigid bounding boxes,” which they say causes several problems that impede the progress of the field.
   They created a high-quality PSG dataset, containing 49k well annotated overlapping images from COCO and Visual Genome, for the community to keep track of its progress. Check out the code

Enjoy these papers and news summaries? Get a daily recap in your inbox!

The Learn AI Together Community section!

Meme of the week!

Featured Community post from the Discord

Another event was organized by a member of the community! Shared by @Zakrz#2739, Cohere AI Hackathon has workshops, keynotes, and mentoring sessions aiming to build with one of the world’s most powerful artificial intelligence language models.

Join the event happening from August 19 through August 21.

AI poll of the week!

What do you think? Join the discussion on Discord.

TAI Curated section

Article of the week

Data Science Essentials — Multicollinearity: This article explains multicollinearity. Multicollinearity may not seem like the most critical topic to grasp, but it is an important widespread concept for machine learning practitioners. The ability of an ML model to find independent variables that are statistically significant for prediction is diminished by high correlations between two or more independent variables. The author provides the most intuitive explanations of types of multicollinearity, causes of multicollinearity, and multicollinearity detection and management.

This week we published 24 new articles and welcomed six new writers to our platform. If you are interested in writing for us at Towards AI, please sign up here and we will publish your blog to our network if it meets our editorial policies and standards. https://contribute.towardsai.net/

Lauren’s Ethical Take on DeepMind’s AlphaFold 2 Expansion

What an incredible advancement! AlphaFold 2 has cataloged the structure of nearly every known protein. This freshly released dataset is open to anyone, and will soon be completely bulk downloadable through Google Cloud Public Datasets.

This is incredibly exciting, and the ethical implications are vast and varied. There is, of course, the proven effects and massive potential of reducing some of the greatest causes of suffering we face today and in the future, such as understanding and treating unique genetic diseases, addressing ecosystem health and biodiversity loss, and improving food supply. While this is a cause for both celebration and optimism, there is also potential for abuse that should be considered and mitigated, such as the creation of targeted biological weaponry using the database. DeepMind will have to decide what that mitigation looks like, but so far, their partnerships and respective advancements demonstrate a trend towards positive outcomes.

I want to highlight a passage from the conclusion of DeepMind’s blog post on the release:

“Just as maths is the perfect description language for physics, we believe AI might turn out to be just the right technique to cope with the dynamic complexity of biology. AlphaFold is an important first proof point for this, and a sign of much more to come.”

This analogy emphasizes that these predictions are not a perfect translation of the secrets of life, but rather an incredibly useful tool. It not only provides clarity, but immense opportunity for both appreciation and improvement of the lives that the information is based on. This technique is certainly cause for hope!

Job offers

Senior Data Scientist @ Storytel (Remote in Europe)

Machine Learning Operations Engineer @ Storytel (Remote in Europe)

Senior Data Engineer @ Storytel (Remote in Europe)

Senior Computer Vision Engineer @ Neurolabs (London & Remote)

Principal Machine Learning Engineer U+007C AI Product @ Jasper.ai (Remote)

Machine Learning Engineer @ Runway (Remote)

Interested in sharing a job opportunity here? Contact [email protected] or post the opportunity in our #hiring channel on discord!

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI