Lab Open Source No-Reward End-to-End Deep Learning Research

Recently, the Berkeley Artificial Intelligence Lab launched a no-reward, end-to-end deep reinforcement learning method that helps robots learn from a number of images depicting successful tasks without rewards. The robot will start learning with about 80 images and occasionally query the user for other tags. During the query, the robot will display an image to the user and ask the user to determine if the image represents the successful completion of the task. Researchers have set up about 25 to 75 such queries, and by using these queries, robots can learn directly in the real world in 1-4 hours of interaction time. It is understood that this method is one of the most efficient image-based robotic machine learning methods in the real world.

United States

Google pushes new AI tools to understand whale behavior with sound;

Amazon announced the launch of a new electric delivery drone;

Facebook and Berkeley Institute of Artificial Intelligence build an open research sharing platform;

The US House of Representatives plans to conduct anti-monopoly investigations on technology companies such as Google and Facebook;

The White House Office of Management and Budget (OMB) issued the Federal Data Strategy;

The US Department of Energy announced the launch of the Quantum Information Science Center program;

US RSIP released a visual AI module to address knee replacement problems;

American researchers develop self-driven micro-robots that are expected to achieve precise targeted therapy;

The Massachusetts Institute of Technology has developed a haptic glove with 548 sensors that can be used to identify objects by tapping;

Microsoft and Oracle reached an agreement to integrate cloud computing services;

Ford developed a two-foot delivery robot to solve the express delivery service.

United Kingdom

DeepMind researchers use the Vector Quantization Variational Automated Encoder (VQ-VAE) model for large-scale image generation;

British operator Three will launch 5G service in London in August;

The University of Oxford will hold a robotic art exhibition.


The University of Bonn, Germany, publishes the report “5G and Geopolitics and Global Competition”;

Italian scientists develop quadruped robots that are powerful enough to pull the plane;

Germany’s Hessen opened its first trackless dual-source electric truck test section.


Russia will use the robot to test before sending a manned landing command;

The Russian technology group said that the Kamazi unmanned truck will soon or will be in the streets of Russia.


Samsung Electronics and AMD reached a strategic cooperation to jointly promote the application of graphic display technology in mobile terminal equipment;

Samsung and MasterCard are working together to launch digital authentication for mobile phones.


Japan Softbank signed a 5G deployment contract with Ericsson and Nokia;

The University of Tokyo research team in Japan has discovered electrode materials with self-healing capabilities that are expected to extend battery life.


The Internet Society publishes “Improving IoT Security: Final Results and Recommendation Reports”;

China and Russia will invest 1 billion US dollars to establish a joint science and technology fund;

Canada introduced the Digital Charter to strengthen regulation of social media;

Apple announced the establishment of its first design development accelerator in China.

Britain uses AI to analyze urban social inequality

Imperial College scientist Azshara Sur and colleagues used deep learning methods to train a computer program that detects four major British cities (London, Birmingham, through publicly available street images and government statistics). Social and economic, environmental and health inequalities in Manchester and Leeds. After 525,860 London city images (corresponding to 156,581 postal codes), fine-tuning with 1% of West Midlands, Greater Manchester and West Yorkshire images, the program predicts results for the other three cities. Accuracy is comparable to that of London.

The research team assumes that some characteristics of the city and urban life, such as housing quality and living environment, have computer-readable direct visual signals, including building materials and disrepair, cars or local shops. Combined with government statistics on the city’s housing conditions, average income or mortality and morbidity, these images were used to train computer programs to detect inequalities in cities without statistics.

German think tanks suggest that the government strengthen technical development such as blockchain

Recently, the German Expert Committee on Research and Innovation submitted the “2019 German Research, Innovation and Technology Capability Assessment Report” to the German government, and proposed to strengthen investment in science and technology innovation and promote the development of emerging technologies such as blockchain and artificial intelligence.

The report proposes that the German government should further promote the development of blockchain and artificial intelligence technology based on the release of “Artificial Intelligence Strategy” and “High-Tech Strategy 2025” in 2018. Such as the introduction of blockchain national strategy, to improve public understanding and confidence in blockchain technology and its application; to clarify the specific artificial intelligence technology development goals, to prioritize the expansion of existing artificial intelligence research centers, and accelerate the research in the field of machine learning Wait.

The report suggests that in 2019 the German government should support the development of start-ups in the field of high-tech, solve the business and legal problems of financing difficulties and equity policies, and provide tax incentives for R&D activities of small and medium-sized enterprises. Accelerate the digital construction of German universities, upgrade digital hardware equipment and teaching content.

NTT announces VR experience walking new technology

The NTT Communication Science Foundation has announced a new technology that enables a seated person to experience a walking experience by combining a virtual reality (VR) head-mounted display device with a device that applies vibration to the sole of the foot. . Discussion is being introduced to cinemas and entertainment facilities.

The user will sit in a special chair and wear a headphone and headphones to hear the footsteps while walking, and get a similar feeling to the actual walking.

The technology of using VR for virtual experiences such as travel has been put into practical use. However, in most cases, there is a need to actually move back and forth when moving in VR. The new technology is characterized by the feeling of going back and forth even without a large enough space.

The person in charge of the institute said: “Not only is walking, but also wants to advance technology to running.”