+++ AI is helping scholars restore ancient Greek texts on stone tablets +++ Open Sourcing Amundsen: A Data Discovery And Metadata Platform +++ AI May Not Kill Your Job—Just Change It +++ A Survey of Deep Learning Techniques for Autonomous Driving +++ The State of Modeling, Simulation, and Data Utilization within Industry. An Autonomous Vehicles Perspective+++ Microsoft AI School and Microsoft Learn +++ Kepler.GL & Jupyter Notebooks: Geospatial Data Visualization with Uber’s opensource Kepler.GL +++
Breakthrough — Or So They Say
Solving Rubik’s Cube with a Robot Hand. Artificial intelligence research organization OpenAI has achieved a new milestone in its quest to build general purpose, self-learning robots. The group’s robotics division says Dactyl, its humanoid robotic hand first developed last year, has learned to solve a Rubik’s cube one-handed. OpenAI sees the feat as a leap forward both for the dexterity of robotic appendages and its own AI software, which allows Dactyl to learn new tasks using virtual simulations before it is presented with a real, physical challenge to overcome. Quote: “We demonstrate that models trained only in simulation can be used to solve a manipulation problem of unprecedented complexity on a real robot. This is made possible by two key components: a novel algorithm, which we call automatic domain randomization (ADR) and a robot platform built for machine learning. ADR automatically generates a distribution over randomized environments of ever-increasing difficulty. Control policies and vision state estimators trained with ADR exhibit vastly improved sim2real transfer. For control policies, memory-augmented models trained on an ADR-generated distribution of environments show clear signs of emergent meta-learning at test time. The combination of ADR with our custom robot platform allows us to solve a Rubik’s cube with a humanoid robot hand, which involves both control and state estimation problems.” News coverage can be found here, here, here, here, here, and on arXiv:1910.07113.
AI is helping scholars restore ancient Greek texts on stone tablets. Researchers at Oxford University and DeepMind Technologies have created “Pythia,” an ancient text restoration system that outperforms experts in guessing the missing text from partially destroyed Ancient Greek inscriptions. The fully automated system, the first of its kind, fills the missing characters, or even whole words, by making alternative proposals. It can thus become a very useful tool for experts trying to read and restore ancient inscriptions, dozens of which are discovered each year to be added to the existing trove of several thousands. Pythia learned to recognize such patterns on some 35,000 inscriptions ranging from the 7th centuty BC to the 5th century AD (1,500 to 2,700 years old) that contain over 3 million words. When shown an incomplete inscription, Pythia makes up to 20 hypotheses, that is, proposes up to 20 different letters or words, leaving it to the experts to make the final choice. News coverage can be found here, here, here, and on arXiv:1910.06262. Some insight into the code can be found on github and Google Colab.
Tools and Frameworks
Open Sourcing Amundsen: A Data Discovery And Metadata Platform. In order to increase productivity of their data scientists and research scientists, Lyft has developed a data discovery application built on top of a metadata engine. Now, Amundsen has been open-sourced and can be found on github. Amundsen follows a micro-service architecture and is comprised of five major components: (1) Metadata Service handles metadata requests from the front-end service as well as other micro services. By default the persistent layer is Neo4j, but can be substituted. (2) Search Service is backed by Elasticsearch to handle search requests from the front-end service. By default the search engine is powered by ElasticSearch, but can be substituted. (3) Front-End Service hosts Amundsen’s web application. (4) Databuilder is a generic data ingestion framework which extracts metadata from various sources. (5) Common is a library repo which holds common codes among all micro services in Amundsen.
Business News and Applications
AI May Not Kill Your Job—Just Change It. The emergence of artificial intelligence (AI) and machine learning (ML) poses a new set of opportunities – and challenges – for work and workers. The tasks that can be done by machine learning are much broader in scope than previous generations of technology have made possible. A new paper from MIT and IBM’s Watson AI Lab, featured on Wired, shows that for most of us, the automation revolution probably won’t mean physical robots replacing human workers. The team of researchers analyzed 170 million online US job listings that were posted between 2010 and 2017. They found that, on average, tasks such as scheduling or credential validation, which could be performed by AI, appeared less frequently in the job listings in the more recent years. The recent listings also included more “soft skills” requirements like creativity, common sense, and judgment. AI is taking over more easily automated tasks and workers are being asked to do things that machines can’t do. Some low-wage occupations like home health care, hairstyling, or fitness training are insulated from the impact of AI because those skills are hard to automate. But middle-wage earners are starting to feel the squeeze and need to adjust their job focus. If you’re in sales, for example, you’ll spend less time figuring out the ideal price for your product, because an algorithm can determine the optimal price to maximize profits. Instead, you might spend more time managing customers or designing attractive marketing materials or websites. The findings of the study are in line with earlier publications by the McKinsey Global Institute and The Partnership on AI.
Publications
A Survey of Deep Learning Techniques for Autonomous Driving (arXiv:1910.07738). A rather long survey, with a total of 38 pages, on applications of deep learning in the context of autonomous driving, by four researchers of the Transilvania University of Brasov (Romania). While I can’t judge the quality in depth, the preprint may help to navigate the topic area serve as a springboard for selected deep dives.
The State of Modeling, Simulation, and Data Utilization within Industry. An Autonomous Vehicles Perspective (arXiv:1910.06075). Another interesting look at autonomous driving, here from the perspective aviation industry, more specifically Boeing, who are trying to learn and draw parallels from their colleagues in ground-based transport. Quote: “The purpose of this paper is to address and decompose the simulation capabilities within the key players of the autonomous vehicle and self-driving car industry (Toyota, Waymo, BMW, Microsoft, NVIDIA, Uber, etc.), as well as several notable startups within the high fidelity 3D mapping and simulation domain (Mapper, HERE, Cognata, etc.).”
Tutorials
Microsoft AI School and Microsoft Learn. Microsoft offers plenty of videos and background readings in the categories AI School, AI Lab projects, and Microsoft Learn.
Kepler.GL & Jupyter Notebooks: Geospatial Data Visualization with Uber’s opensource Kepler.GL. kepler.gl is a web-based visualisation tool for large geospatial datasets built on top of deck.gl. Uber open-sourced it last year, and its functionality is impressive. This tutorial shows how you can incorporate kepler.gl into a Jupyter notebook, combining the flexibility of Jupyter Notebooks with Kepler’s great visualisation capabilities.