Jeff Dean personally reviews Google AI 2019: 2 papers per day, covering 16 major directions, and a collection of important open source algorithms in one article

Latest update time：2020-01-10

Reads：

Qianming Biance Thirteen Guo Yipu sent from Aofei Temple
Quantum Bit Report | Public Account QbitAI

Another year, Jeff Dean, on behalf of Google AI, summarized the major trends in AI over the past year.

This is my brother-in-law's routine annual report as the general manager of Google AI , and it is also a show of strength as the world's largest AI company, and even the leading edge technology company.

He said that the past 2019 was a very exciting year.

There is still a flourishing of both academic and application, and open source and new technologies are promoted simultaneously.

Starting from basic research, to the application of technology in emerging fields, and then to looking forward to 2020.

Although the reporting format has not changed, artificial intelligence technology has taken a big step forward.

Jeff Dean summarized the AI achievements in 16 major areas and revealed that the number of AI papers published throughout the year reached 754 , with an average of 2 papers published every day.

Covering AutoML, machine learning algorithms, quantum computing, perception technology, robotics, medical AI, AI for good...

All of these things not only promote the influence of AI in all aspects of society at present, but also are a small display of future trends.

It is no exaggeration to say that if you want to know the progress of AI technology in 2019, it is most appropriate to read Jeff's summary; if you want to know where AI will go in 2020, reading Jeff's article can also be very beneficial.

For your convenience, we have compiled a small directory for you:

Machine Learning Algorithms : Understanding the Dynamic Training Properties of Neural Networks

AutoML : Keep an eye on automating machine learning

Natural language understanding : combining multiple methods and tasks to improve the technical level

Machine perception : Deeper understanding and perception of images, videos, and environments

Robotics : Self-supervised training, release of robot testing benchmarks

Quantum computing : Achieving quantum supremacy for the first time

Application of AI in other disciplines : from fly brains to mathematics, as well as chemical molecular research and artistic creation

Mobile AI applications : Locally deployed voice and image recognition models, as well as more powerful translation, navigation, and photography

Health and medical care : Has been used in clinical diagnosis of breast cancer and skin diseases

AI assists people with disabilities : Using image recognition and speech transcription technology to benefit vulnerable groups

AI promotes social welfare : predicting floods, protecting plants and animals, teaching children to read and learn mathematics, and investing more than 100 million yuan in 20 public welfare projects

TensorFlow is fully upgraded to build developer tools and benefit the researcher community

11 open datasets : from reinforcement learning to natural language processing to image segmentation

Global expansion of top conference research and Google research : publish a large number of papers and invest a lot of resources to support teachers, students and researchers in various fields to conduct research

Artificial intelligence ethics : Promoting research progress in fairness, privacy protection, and explainability of artificial intelligence

Looking ahead to 2020 and beyond : The deep learning revolution will continue to reshape how we think about computing and computers.

Machine Learning Algorithms

In 2019, Google conducted research in many different areas of machine learning algorithms and methods.

A major focus is on understanding the nature of training dynamics in neural networks .

In the following study, the researchers' experimental results show that scaling the amount of data parallelism can make the model converge faster and more effectively.

Paper address:

https://arxiv.org/pdf/1811.03600.pdf

Compared to data parallelism, model parallelism can be an effective way to scale models.

GPipe is a library that makes model parallelization more efficient:

While one part of the model is processing some data, other parts can be doing other work and calculating different data.

This pipeline approach can be combined together to simulate more efficient batch sizes.

GPipe library address:

https://ai.googleblog.com/2019/03/introducing-gpipe-open-source-library.html

Machine learning models are most effective when they can take raw input data and learn “disentangled” high-level representations.

These representations distinguish different kinds of examples by the properties that the user wants the model to be able to distinguish.

Advances in machine learning algorithms have been primarily aimed at encouraging the learning of better representations that generalize to new examples, problems, and domains.

In 2019, Google studied this issue in different contexts:

For example, they examined which properties influence representations learned from unsupervised data in order to better understand what factors contribute to good representations and effective learning.

Blog address:
https://ai.googleblog.com/2019/04/evaluating-unsupervised-learning-of.html

Google showed that it is possible to use statistics of the margin distribution to predict the generalization gap, which helps understand which model generalizes most effectively.

In addition, off-policy classification has been studied in the context of reinforcement learning to better understand which models are likely to generalize best.

Blog address:
http://ai.googleblog.com/2019/07/predicting-generalization-gap-in-deep.html

We study methods for specifying reward functions for reinforcement learning so that the learning system can learn more directly from the true objective.

blog address:

http://ai.googleblog.com/2019/02/learning-to-generalize-from-sparse-and.html

AutoML

Google continued to focus on AutoML in 2019.

This approach can automate many aspects of machine learning and often achieve better results on certain types of machine learning meta-decisions, such as:

Google demonstrated how to use neural architecture search techniques to achieve better results on computer vision problems, with an accuracy of 84.4% on ImageNet while using 8 times fewer parameters than the previous best model.

Blog address:
http://ai.googleblog.com/2019/05/efficientnet-improving-accuracy-and.html

Google demonstrated a neural architecture search method that shows how to find efficient models suitable for specific hardware accelerators, thereby providing high-precision, low-computation running models for mobile devices.

Blog address:
http://ai.googleblog.com/2019/08/efficientnet-edgetpu-creating.html

Google showed how to extend the AutoML work to the domain of video models, finding architectures that achieve state-of-the-art results as well as lightweight architectures that can match the performance of hand-crafted models.

The result is a 50-fold reduction in computational effort.

Blog address:
http://ai.googleblog.com/2019/10/video-architecture-search.html

Google developed AutoML technology for tabular data and collaborated to release this technology as a new product called Google Cloud AutoML Tables.

Blog address:
http://ai.googleblog.com/2019/05/an-end-to-end-automl-solution-for.html

We show how to find interesting neural network architectures without using any training step to update the weights of the evaluated model, making architecture search more computationally efficient.

Blog address:
http://ai.googleblog.com/2019/08/exploring-weight-agnostic-neural.html

We explore architectures for discovering NLP tasks that significantly outperform vanilla Transformer models at significantly lower computational cost.

blog address:

http://ai.googleblog.com/2019/06/applying-automl-to-transformer.html

The study demonstrated that automatically learned data augmentation methods can be extended to speech recognition models.

Compared to existing human ML-expert driven data augmentation methods, significantly higher accuracy can be achieved with less data.

Blog address:
http://ai.googleblog.com/2019/04/specaugment-new-data-augmentation.html

Launched the first voice application using AutoML for keyword spotting and spoken language recognition.

In experiments, we found models that are better than those designed by humans: more efficient and with better performance.

Blog address:
https://www.isca-speech.org/archive/Interspeech_2019/abstracts/1916.html

Natural Language Understanding

The past few years have seen remarkable progress in models for natural language understanding, translation, natural conversation, speech recognition, and related tasks.

One theme that Google is working on in 2019 is:

Improve the state of the art by combining various methods or tasks to train more powerful models.

For example, only one model is used to train translation between 100 languages (instead of using 100 different models), which significantly improves the translation quality.

Blog address:
http://ai.googleblog.com/2019/10/exploring-massively-multilingual.html

We show how combining speech recognition and language models and training the system on multiple languages can significantly improve the accuracy of speech recognition.

Blog address:
http://ai.googleblog.com/2019/09/large-scale-multilingual-speech.html

The research demonstrates that it is possible to train a joint model to perform speech recognition, translation, and text-to-speech generation tasks.

It also offers certain advantages, such as preserving the speaker’s voice in the generated translated audio, and a simpler overall learning system.

Blog address:
http://ai.googleblog.com/2019/05/introducing-translatotron-end-to-end.html

We show how to combine many different objectives to produce a model that is significantly better at semantic retrieval.

For example, in GoogleTalk to Books, ask, "What scents bring back memories?"

The result is, "For me, the scent of jasmine and the aroma of baking pans remind me of my carefree childhood."

Blog address:
http://ai.googleblog.com/2019/07/multilingual-universal-sentence-encoder.html

We show how to use an adversarial training procedure to significantly improve the quality and robustness of language translation.

Blog address:
http://ai.googleblog.com/2019/07/robust-neural-machine-translation.html

With the development of models such as seq2seq, Transformer, BERT, Transformer-XL, and ALBERT, Google's language understanding technology capabilities continue to improve and have been applied to many core products and features.

In 2019, the application of BERT to core search and ranking algorithms brought the biggest improvement in search quality in the past five years (and one of the biggest improvements ever).

Machine Perception

Models for better understanding of static images have made significant progress over the past decade.

What follows is Google’s main research in this area over the past year.

This includes a deeper understanding of images and videos, as well as perception of life and the environment, including:

We investigate finer-grained visual understanding in shots, enabling more powerful visual search.

Blog address:
https://www.blog.google/products/search/helpful-new-visual-features-search-lens-io/

The Nest Hub Max's smart camera features are shown off, such as Quick Gestures, Face Match, and Smart Video Call Framing.

Blog address:
https://blog.google/products/google-nest/hub-max-io/

We investigated better video depth prediction models.

Blog address:
https://ai.googleblog.com/2019/05/moving-camera-moving-people-deep.html

We use temporal periodic consistency to learn better representations for fine-grained temporal understanding of videos.

Blog address:
https://ai.googleblog.com/2019/08/video-understanding-using-temporal.html

Learning temporally consistent representations in text, speech, and video with unlabeled video.

Blog address:
https://ai.googleblog.com/2019/09/learning-cross-modal-temporal.html

It is also possible to predict future visual input by observing the past.

Blog address:
https://ai.googleblog.com/2019/03/simulated-policy-learning-in-video.html

And proved that the model can better understand the action sequences in the video.

Blog address:
https://ai.googleblog.com/2019/04/capturing-special-video-moments-with.html

robot technology

The application of machine learning to robot control is an important area of research at Google, which believes that it is an important tool to enable robots to operate effectively in complex real-world environments, such as everyday homes and businesses.

Google's work in robotics in 2019 includes:

1. In Remote Robot Navigation via Automated Reinforcement Learning, Google demonstrated how to combine reinforcement learning with remote projects to enable robots to navigate more effectively in complex environments, such as the Google office building.

Related Links:

http://ai.googleblog.com/2019/02/long-range-robotic-navigation-via.html

2. In PlaNet , Google demonstrated how to effectively learn a world model from images alone, and how to use this model to complete tasks with fewer learning attempts.

Related link:
http://ai.googleblog.com/2019/02/introducing-planet-deep-planning.html

3. In TossingBot , Google unified the laws of physics and deep learning, allowing the robot to learn intuitive physical principles through experiments, and then throw objects into the box according to the learned rules.

Related Links:

http://ai.googleblog.com/2019/03/unifying-physics-and-deep-learning-with.html

4. In the study of Soft Actor-Critic, Google proved that the way to train reinforcement learning algorithms can be achieved by maximizing the expected reward or by maximizing the entropy of the strategy.

This could help the robot learn faster and be more robust to changes in its environment.

Related link:
http://ai.googleblog.com/2019/01/soft-actor-critic-deep-reinforcement.html

5. Google has also developed a self-supervised learning algorithm for robots, which allows robots to learn to assemble objects by disassembling them in a self-supervised way. This shows that robots can learn knowledge from disassembly just like children.

Related link:
http://ai.googleblog.com/2019/10/learning-to-assemble-and-to-generalize.html

6. Finally, Google also launched ROBEL , a benchmark for low-cost robots, which is an open source platform for low-cost robots to help other developers develop robot hardware faster and more conveniently.

Related link:
http://ai.googleblog.com/2019/10/robel-robotics-benchmarks-for-learning.html

Quantum computing

In 2019, Google made a major breakthrough in quantum computing, demonstrating quantum superiority to the world for the first time: in a computing task, quantum computers are far faster than classical computers.

A task that would take a classical computer 10,000 years to calculate can be completed in just 200 seconds by a quantum computer. This research was featured on the cover of the October 24 issue of Nature magazine.

△ Google's Sycamore processor for quantum computing

Google CEO Pichai said: "It means the same thing as the first rocket successfully escaping the earth's gravity and flying to the edge of space." Quantum computers will play an important role in fields such as materials science, quantum chemistry and large-scale optimization.

Google is also working to make quantum algorithms easier to express and the hardware easier to control, and it has found ways to use classical machine learning techniques in quantum computing.

Application of AI in other disciplines

Google has published many papers on the application of artificial intelligence and machine learning in other scientific fields, mainly in the field of multi-organizational collaboration.

Proceedings:
https://research.google/pubs/?area=general-science

This year's highlights include:

Interactive automatic 3D reconstruction of the fly brain, using a machine learning model to carefully draw each neuron in the fly brain, which Jeff Dean called a milestone in mapping the structure of the fly brain.

Related blog:
https://ai.googleblog.com/2019/08/an-interactive-automated-3d.html

In learning better simulation methods for partial differential equations, Google uses machine learning to accelerate partial differential equation calculations, which are at the heart of studying fundamental computational problems such as climate change, fluid dynamics, electromagnetism, heat conduction, and general relativity.

△ Simulation of two solutions to the Burgers equation

Google also uses machine learning models to judge smells and GNN to judge molecular structures to predict what it smells like.

Also in chemistry, Google has developed a reinforcement learning framework to optimize molecules.

Related paper:
https://www.nature.com/articles/s41598-019-47148-x

In terms of artistic creation, Google AI has made even greater efforts, such as the artistic expression of AI+AR

https://www.blog.google/outreach-initiatives/arts-culture/how-artists-use-ai-and-ar-collaborations-google-arts-culture/

Re-choreographing dances with machines:

https://www.blog.google/technology/ai/bill-t-jones-dance-art/

New exploration of AI composition:

https://www.blog.google/technology/ai/behind-magenta-tech-rocked-io/

It also extends to a fun AI composition Doodle:

https://www.blog.google/technology/ai/honoring-js-bach-our-first-ai-powered-doodle/

Mobile AI Applications

A lot of what Google does is to use machine learning to give mobile phones new capabilities. These models can run on the phone, and even if airplane mode is turned on, these features can still be used.

Now, the mobile phone speech recognition model, vision model, and handwriting recognition model have all been implemented.

Related Blogs:

Speech recognition
https://ai.googleblog.com/2019/03/an-all-neural-on-device-speech.html
Visual model
https://ai.googleblog.com/2019/11/introducing-next-generation-on-device.html
Handwriting recognition model
https://ai.googleblog.com/2019/03/rnn-based-handwriting-recognition-in.html

Jeff Dean said this paves the way for more powerful new features.

In addition, this year's highlights on Google's mobile phones include:

The Live Caption function can automatically add subtitles to videos played by any application on the phone.

Related blog:
https://ai.googleblog.com/2019/10/on-device-captioning-with-live-caption.html

The Recorder app lets you search the audio you've recorded on your phone.

Related blog:
https://ai.googleblog.com/2019/12/the-on-device-machine-learning-behind.html

Google Translate's camera translation function has also been upgraded to support multiple languages including Arabic, Hindi, Malay, Thai and Vietnamese. It not only translates between English and other languages, but also between languages other than English. It can also automatically find where the text is in the camera frame.

Related blog:
https://www.blog.google/products/translate/google-translates-instant-camera-translation-gets-upgrade/

We also released a facial enhancement API in ARCore to help you achieve real-time AR gameplay.

Face Augmentation API:
https://developers.google.com/ar/develop/java/augmented-faces/

There is also mobile gesture recognition, and once this is done, gesture interaction can be performed.

RNNs were also used to improve handwriting input recognition on mobile phone screens.

Related blog:
https://ai.googleblog.com/2019/03/rnn-based-handwriting-recognition-in.html

When it comes to navigation and positioning, GPS is often only an approximate location, but AI can play a key role.

Combined with the data from Google Street View, if you hold up your phone and turn it around, the phone will become like a friend who knows the way, and will point out to you based on the street view and map: which building is this, which street is this, this is south, this is north, and you should go this way. ‍

Related blog:
https://ai.googleblog.com/2019/02/using-global-localization-to-improve.html

In addition, in order to ensure user privacy, Google has also been researching federated learning. The following paper is an article transcribed by the Google team in 2019 on the progress of federated learning:

https://arxiv.org/abs/1912.04977

As for the old topic of mobile phone photography, Google improved the selfie capabilities of mobile phones in 2019. ‍

Related blog:
https://ai.googleblog.com/2019/04/take-your-best-selfie-automatically.html

Background blur and portrait mode also got a boost in 2019.

Related blog:
https://ai.googleblog.com/2019/12/improvements-to-portrait-mode-on-google.html

There was also a huge improvement in the night scene challenge of photographing stars, and a paper was published at SIGGRAPH Asia.

Related blog:
https://ai.googleblog.com/2019/11/astrophotography-with-night-sight-on.html

Related papers:
https://arxiv.org/abs/1905.03277
https://arxiv.org/abs/1910.11336

Health and Medical

2019 marks the first full year for the Google Health team.

At the end of 2018, Google reorganized the Google Research health team, Deepmind Health, and the health-related hardware department to create the new Google Health team.

1. Google has made many achievements in the diagnosis and early detection of diseases:

The use of deep learning models to detect breast cancer has a higher accuracy than human experts, reducing false positive and false negative cases in diagnosis. This research was recently published in Nature magazine.

In addition, Google has also made some new achievements in diagnosing skin diseases, predicting acute kidney injury, and detecting early lung cancer.

2. Google combines machine learning with other technologies in other medical technologies, such as adding enhanced display technology to microscopes to help doctors quickly locate lesions.

Google has also built a human-centric similar image search tool for pathologists, allowing examination of similar cases to help doctors make more effective diagnoses.

AI assists people with disabilities

AI is becoming more and more closely integrated into our lives. In the past year, Google has used AI to help us in our daily lives.

It’s easy for us to see beautiful images, hear our favorite songs, or talk with our loved ones, yet more than a billion people around the world don’t have access to these same ways of seeing the world.

Machine learning technology can help people with disabilities by converting these audiovisual signals into other signals. The AI assistant technologies provided by Google include:

Lookout helps people who are blind or have low vision identify information about their surroundings.

Live Transcribe, a real-time transcription technology, helps deaf or hard-of-hearing people quickly convert speech into text.

Project Euphonia enables personalized speech-to-text conversion, and this research improves the accuracy of automatic speech recognition for people with diseases such as ALS that cause slurred speech.

There is also a Parrotron project that also uses end-to-end neural networks to help improve communication, but the research focuses on speech-to-speech conversion.

For the blind and low-vision community, Google is using AI to generate descriptions for images. When a screen reader encounters an image or graphic without a description, Chrome can now automatically create a description.

Lens for Google Go, a tool that reads text in audio form, has greatly helped illiterate users obtain information in the world of words.

AI for social good

Jeff Dean said that machine learning has great significance in solving many major social problems. Google has been making efforts in some areas of social problems and is committed to enabling others to use their creativity and skills to solve these problems.

Take flooding for example. Hundreds of millions of people are affected by flooding every year. Google uses machine learning, computing, and better databases to make flood predictions and send alerts to millions of people in affected areas.

They even held a workshop and invited many researchers to specifically solve this problem.

Related blogs:
https://www.blog.google/technology/ai/tracking-our-progress-on-flood-forecasting/
https://ai.googleblog.com/2019/09/an-inside-look-at-flood-forecasting.html
https://ai.googleblog.com/2019/03/a-summary-of-google-flood-forecasting.html

In addition, Google has also done some work related to machine learning and animal and plant research.

They partnered with seven wildlife conservation organizations to use machine learning to help analyze wildlife photo data and find where these wildlife populations are.

Related blog:
https://www.blog.google/products/earth/ai-finds-where-the-wild-things-are/

Google is also working with the National Oceanic and Atmospheric Administration to use underwater sound data to determine the location of whale populations.

Related blog:
https://www.blog.google/technology/ai/pattern-radio-whale-songs/

Google has released a set of tools to study biodiversity using machine learning.

Related blog:
A New Workflow for Collaborative Machine Learning Research in Biodiversity
https://ai.googleblog.com/2019/10/a-new-workflow-for-collaborative.html

They also held a Kaggle competition to use computer vision to classify various diseases on cassava leaves. Cassava is the second largest source of carbohydrates in Africa, and cassava diseases affect people's video safety issues.

https://www.kaggle.com/c/cassava-disease

Google Earth's Timelapse feature has also been updated, and you can even see data on population movement and migration.

Related blogs:
https://ai.googleblog.com/2019/06/an-inside-look-at-google-earth-timelapse.html
https://ai.googleblog.com/2019/11/new-insights-into-human-mobility-with.html

In terms of education, Google has developed the Bolo application with voice recognition technology to guide children to learn English. This application is deployed locally and can run offline. It has helped 800,000 Indian children learn to read, and the children have read a total of 1 billion words. In a pilot in 200 villages in India, 64% of the children's reading ability has improved.

It's like a Google version of Liulishuo.

Related blog:
https://www.blog.google/technology/ai/bolo-literacy/

In addition to literacy, there are more complex learning subjects such as mathematics and physics. Google has created the Socratic app to help high school students learn mathematics.

In addition, in order to enable AI to play a greater role in public welfare, Google held the AI Impact Challenge, collecting more than 2,600 proposals from 119 countries.

Eventually, 20 proposals that could solve major social and environmental problems stood out. Google invested $25 million (more than 170 million RMB) in these proposals and made some achievements, including:

Doctors Without Borders (MSF) has created a free mobile app that uses image recognition tools to help clinicians in underserved areas analyze antimicrobial images and provide recommendations on what drugs to use for patients. The project has been piloted in Jordan.

Doctors Without Borders project report:
https://www.doctorswithoutborders.org/what-we-do/news-stories/news/msf-receives-google-grant-develop-new-free-smartphone-app-help

A billion people in the world make their livings on small farms, but pests and diseases can wipe out their livelihoods.

Therefore, an NPO called Wadhwani AI uses image classification models to identify pests on farms and give recommendations on which pesticides to spray and when to spray, thereby increasing crop yields.

Illegal deforestation of tropical rainforests is a major factor influencing climate change. An organization called Rainforest Connection uses deep learning for bioacoustic detection. Using some old mobile phones, it can track the health of the rainforest and detect threats.

△ 20 public welfare projects funded by Google

Developer tools build and benefit the researcher community

As the world's leading AI company, Google is also a pioneer in open source, constantly contributing to the community, focusing on TensorFlow.

Jeff Dean said that the past year has been an exciting year for the open source community because of the release of TensorFlow 2.0.

This is the first major upgrade to TensorFlow since its release, making it easier than ever to build ML systems and applications.

The relevant reports of quantum bit are as follows:

GoogleTF2.0 released in the early morning! "Changes everything, surpasses PyTorch"

In TensorFlow Lite, they added support for fast mobile GPU inference; and released Teachable Machine 2.0, which allows you to train a machine learning model with just one button without writing any code.

The relevant reports of quantum bit are as follows:

TensorFlow Lite released a major update! Support for mobile GPUs, inference speed increased by 4-6 times

There’s also MLIR, an open source machine learning compiler infrastructure tool that addresses the growing complexity of software and hardware fragmentation, making it easier to build AI applications.

At NeurIPS 2019, they showed how to use JAX, an open source high-performance machine learning research system:

https://nips.cc/Conferences/2019

In addition, they also open-sourced MediaPipe, a framework for building ML pipelines for perception and multimodal applications:

https://github.com/google/mediapipe

And the efficient floating-point neural network inference operator library XNNPACK:

https://github.com/google/XNNPACK

Of course, Google has also given you some free deals.

Jeff Dean introduced that as of the end of 2019, they have provided free access to Cloud TPU to more than 1,500 researchers around the world through TensorFlow Research Cloud, and their introductory courses on Coursera have more than 100,000 students, etc.

At the same time, he also introduced some "heartwarming" cases, such as a college student who discovered two new planets with the help of TensorFlow and established a method to help others discover more planets.

College students are also using TensorFlow to identify potholes and dangerous road cracks in Los Angeles, among other things.

Another aspect is open data sets.

11 datasets are open

After launching the dataset search engine in 2018, Google is still working on this this year and doing its best to add more features to this search engine.

In the past year, Google has opened up 11 datasets in various fields. Here are some resources to share. Please keep them~

Open Images V5, adding segmentation masks to the annotation set, has a sample size of 2.8 million across 350 categories, QuantumBit reports:

2.8 million samples! Google opens the largest segmentation mask dataset in history and starts a new round of challenges

The “Natural Questions” dataset, the first dataset to use naturally occurring queries and find answers by reading the entire page, rather than extracting the answer from a small paragraph, has 300,000 question-answer pairs, and BERT cannot reach 70 points, Quantum Bit reported:

Google releases super-difficult question-answering dataset "Natural Questions": 300,000 question-answer pairs, BERT can't even reach 70 points

Datasets for detecting deepfakes:

https://ai.googleblog.com/2019/09/contributing-data-to-deepfake-detection.html

In the football simulation environment Google Research Football, the intelligent agent can play football freely in this FIFA-like world and learn more football skills. Quantum Bit reports:

Google built a virtual football field, allowing AI to do reinforcement learning training like playing FIFA | Open source with API

Landmark dataset Google-Landmarks-v2: includes 5 million images and 200,000 landmarks. QuantumBit reports:

5 million images, 200,000 landmarks, Google releases another large dataset

YouTube-8M Segments dataset, a large-scale classification and temporal localization dataset, including manually verified labels at the 5-second segment level of YouTube-8M videos:

https://ai.googleblog.com/2019/06/announcing-youtube-8m-segments-dataset.html

AVA Spoken Activity dataset, a multimodal audio + visual video perceptual dialogue dataset:

https://research.google.com/ava/

PAWS and PAWS-X: for machine translation. Both datasets consist of highly structured sentence pairs with high lexical overlap, and about half of the sentences have corresponding multilingual translations:

https://ai.googleblog.com/2019/10/releasing-paws-and-paws-x-two-new.html

Let two people have a conversation, and simulate the natural language conversation dataset of human conversation through digital assistants:

https://ai.googleblog.com/2019/09/announcing-two-new-natural-language.html

Visual Task Adaptation Benchmark: This is a visual task adaptation benchmark launched by Google to compete with GLUE and ImageNet.

It helps users better understand which visual representations can be generalized to more new tasks, thereby reducing the data requirements for all visual tasks:

http://ai.googleblog.com/2019/11/the-visual-task-adaptation-benchmark.html

The largest publicly available database of task-oriented conversations, the Mode Guided Conversation Dataset, has over 18,000 conversations across 17 domains:

http://ai.googleblog.com/2019/10/introducing-schema-guided-dialogue.html

Global expansion of top conference research and Google research

According to official statistics from Google, Googlers published 754 papers in the past year.

Jeff Dean also listed some top conference results:

CVPR has more than 40 papers, ICML has more than 100 papers, ICLR has more than 60 papers, ACL has more than 40 papers, ICCV has more than 40 papers, NeurIPS has more than 120 papers, and so on.

They also held 15 separate workshops at Google on topics ranging from improving global flood warnings to how to use machine learning to build systems that better serve people with disabilities to accelerating the development of algorithms, applications, and tools for quantum processors (NISQ).

It has also funded more than 50 doctoral students around the world through its annual doctoral scholarship program, provided support to start-up companies, and so on.

Similarly, in 2019, Google's research locations continued to expand globally, opening a research office in Bangalore. At the same time, Jeff Dean also issued a recruitment demand: If you are interested, hurry up and come to the bowl~

Artificial Intelligence Ethics

As in previous years, at the beginning of this report, Jeff first talked about Google's work on artificial intelligence ethics.

This is also Google's clear declaration on AI practice, ethics, and technology for good.

In 2018, Google released the Seven Principles of AI and developed application practices around these principles. In June 2019, Google handed in a report card showing how these principles were put into practice in research and product development.

Report link:
https://www.blog.google/technology/ai/responsible-ai-principles/

Jeff Dean said that these principles basically cover the most active areas of artificial intelligence and machine learning research, such as bias, security, fairness, reliability, transparency and privacy in machine learning systems.

Therefore, Google's goal is to apply the technologies in these fields to work and continue to conduct research to continue to advance the development of related technologies.

On the one hand, Google has also published a number of papers at academic conferences such as KDD'19 and AIES 19 to explore the fairness and interpretability of machine learning models.

For example, we study how Activation Atlases can help explore the behavior of neural networks and how they can help the interpretability of machine learning models.

Related link:
Exploring Neural Networks with Activation Atlases
https://ai.googleblog.com/2019/03/exploring-neural-networks.html

On the other hand, Google's efforts have been put into practice and real products have been produced.

For example, TensorFlow Privacy was released to help train privacy-preserving machine learning models.

Related link:
Introducing TensorFlow Privacy: Learning with Differential Privacy for Training Data
https://blog.tensorflow.org/2019/03/introducing-tensorflow-privacy-learning.html

In addition, Google has released a new dataset to help research on identifying deepfakes.

Related link:
Contributing Data to Deepfake Detection Research
https://ai.googleblog.com/2019/09/contributing-data-to-deepfake-detection.html

Looking ahead to 2020 and beyond

Finally, Jeff also looked at the development process of the past 10 years and looked forward to the research trends in 2020 and beyond.

He said that over the past decade, there have been remarkable advances in the fields of machine learning and computer science, and we now have computers that are better able to see, hear and understand language than ever before.

With sophisticated computing devices in our pockets, we can harness these capabilities to better help us perform many of the tasks we do in our daily lives.

We redesigned our computing platform around these machine learning methods by developing specialized hardware that enables us to tackle larger problems.

These have changed the way we think about computing devices in data centers, and the deep learning revolution will continue to reshape the way we think about computing and computers.

At the same time, he also pointed out that there are still a lot of unresolved problems. This is also the research direction of Google in 2020 and beyond:

First, how do you build a machine learning system that can handle millions of tasks and can automatically and successfully complete new tasks?

Second, how can we achieve state-of-the-art progress in important areas of AI research, such as avoiding bias, improving explainability and comprehensibility, improving privacy, and ensuring safety?

Third, how can computing and machine learning be applied to make progress in important new areas of science, such as climate science, health care, bioinformatics, and many other fields?

Fourth, how can we ensure that more diverse researchers are proposing and exploring the ideas and directions that the machine learning and computer science research community pursues? How can we best support new researchers from diverse backgrounds entering the field?

Finally, what do you think of the breakthroughs and progress of Google AI in the past year?

Welcome to interact in the comment area~

Report portal:
https://ai.googleblog.com/2020/01/google-research-looking-back-at-2019.html

Google 2019 paper portal:
https://research.google/pubs/?year=2019

The author is a contracted author of NetEase News and NetEase "Each has its own attitude"

-over-

AI Insider | Seize new opportunities for AI development

Expand your network of high-quality contacts, obtain the latest AI information & paper tutorials, welcome to join the AI Insider Community to learn together~

Communicate with experts | Enter the AI community