In five years, the computing power required by AI will exceed 1 million times! Twelve organizations jointly published an 88-page article: 'Intelligent computing' is the antidote

Artificial intelligence is an industry that competes for financial resources. Without high-performance computing equipment, it is impossible to even fine-tune the model, let alone develop a basic model.

But if we only rely on hardware and the current development speed of computing performance, sooner or later we will not be able to meet the growing demand , so we still need Supporting software is used to coordinate and coordinate computing capabilities. At this time, "intelligent computing" technology is needed.

Recently, as many as twelve domestic and foreign research institutions including Zhijiang Laboratory, Chinese Academy of Engineering, National University of Defense Technology, and Zhejiang University jointly published a paper, for the first time A comprehensive survey was conducted on the field of intelligent computing, covering theoretical foundations, technological integration of intelligence and computing, important applications, challenges and future prospects.

Paper link: https://spj.science.org/doi/10.34133/icomputing.0006​

This is also the first review article that formally proposes the definition of intelligent computing and its unified theoretical framework. The full text is structured as follows.

AI opens the era of large-scale computing power

Human society is moving from the information society to the intelligent society, and computing has become the norm and promotion key factors in social development. In the new era of digital civilization where everything is interconnected, traditional data computing is far from meeting humanity's growing needs for higher levels of intelligence.

People’s growing interest in intelligent computing, coupled with the development of computing science, intelligent perception of the physical world, and understanding of the cognitive mechanism of human consciousness, have jointly improved the capabilities of computing. The level of intelligence accelerates the discovery and creation of knowledge.

In recent years, computing and information technology have developed rapidly. Due to the unprecedented popularity and success of deep learning, artificial intelligence (AI) has been established as the forefront of human exploration of machine intelligence, and has On this basis, a series of breakthrough research results have been achieved, including:

Convolutional Neural Network (CNN) proposed by Yann LeCun; Yoshua Bengio’s contribution in the field of causal reasoning of deep learning ;Geoffrey Hinton, one of the pioneers of artificial intelligence, proposed the Deep Belief Network (Deep Brief Network) model and backpropagation optimization algorithm in 2006.

J̈urgen Schmidhuber proposed the widely used Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM), and successfully used them to process sequence data , such as voice, video and time series data.

In March 2016, AlphaGo, the artificial intelligence Go program launched by DeepMind, competed against the world's top human Go master Lee Sedol, attracting unprecedented attention from around the world. This epoch-making human-machine war It ended with a landslide victory for artificial intelligence and became the catalyst that pushed the wave of artificial intelligence to a whole new level.

Another important enabler of artificial intelligence is the emergence of large-scale pre-trained models, which have begun to be widely used in natural language and image processing, Handle a wide variety of applications with the help of transfer learning.

For example, GPT-3 has proven that a large model with high structural complexity and a large number of parameters can improve the performance of deep learning. Inspired by GPT-3, a large number of large-scale deep learning models.

Intelligence and Computing

Computing power is one of the important factors supporting intelligent computing.

Given the astronomical number of data sources, heterogeneous hardware configurations and changing computing needs in the information society, intelligent computing mainly meets the computing power requirements of intelligent tasks through vertical and horizontal architectures.

Vertical architectures are characterized by homogeneous computing infrastructure, mainly improved by applying intelligent methods Resource utilization efficiency to improve computing power.

In contrast, horizontal architectureCoordinates and schedules heterogeneous and wide-area computing resources to maximize collaboration Computational efficiency.

For example, in April 2020, in response to the computing needs of global COVID-19 research, Folding@home combined 400,000 computing volunteers to achieve 2.5 Exaflops of computing in three weeks The computing power is greater than any supercomputer in the world.

Despite the great successes in intelligence and computing, these two fields still face some difficulties.

Challenges of Intelligence

Artificial intelligence using deep learning is currently in the process of interpretability, versatility, and evolvability. Aspects of sexuality and autonomy remain unresolved.

Most current artificial intelligence technologies can only play a weak role compared with human intelligence, and only play a role in specific fields or tasks, achieving powerful and general artificial intelligence. Intelligence still has a long way to go.

Finally, there are also major challenges in upgrading from data-based intelligence to more diverse forms of intelligence, including perceptual intelligence, cognitive intelligence, autonomous intelligence, and human-machine fusion intelligence. Theoretical and technical challenges.

Computing Challenges

The digital wave has brought unprecedented growth in applications, connections, terminals and users, As well as the amount of data generated, huge computing power is required.

For example, the computing power required for artificial intelligence doubles every 100 days, that is, it is expected to increase more than 1 million times in the next five years.

As Moore's Law wears off, it becomes challenging to keep up with such rapidly increasing computing power requirements.

Moore's Law: The number of transistors that can be accommodated on an integrated circuit will double approximately every two years

The processing of large-scale tasks in an intelligent society relies on the effective combination of various specific computing resources. The traditional hardware model cannot adapt well to intelligent algorithms, which greatly limits the development of software. .

What is intelligent computing?

To this day, there is still no widely accepted definition of intelligent computing.

Some researchers regard intelligent computing as a combination of artificial intelligence and computing technology, but this view limits the definition of intelligent computing to the field of artificial intelligence, while ignoring the inherent limitations and the important role of triadic interactions between humans, machines and things.

Another school of thought regards intelligent computing as computational intelligence, which imitates human or biological intelligence to implement optimal algorithms to solve specific problems, and refers to intelligent computing as computational intelligence. Mainly regarded as an algorithmic innovation.

In this paper, the researchers propose a new definition of intelligent computing from the perspective of solving complex scientific and social problems, taking into account the three basic spaces of the world, namely human The increasingly close integration of social space, physical space and information space.

Definition of Intelligent Computing

Intelligent computing refers to the era of digital civilization that supports world interconnection, including new computing theory methods, architectural systems and technologies areas of competence. Intelligent computing completes computing tasks at the minimum cost based on specific actual needs, matches sufficient computing power, calls the best algorithm, and obtains the best results.

The new definition of intelligent computing is proposed in response to the rapidly growing computing needs in the integration of human society, the physical world, and the information space. Intelligent computing is people-oriented and pursues high computing power, high energy efficiency, high intelligence, and high security.

Its goal is to provide ubiquitous, efficient, secure, autonomous, reliable and transparent computing services to support large-scale and complex computing tasks. Figure 1 shows the overall theoretical framework of intelligent computing, which embodies various computing paradigms that support human-physical-information integration.

First of all, intelligent computing is neither existing computers, cloud computing, edge computing, nor other computing technologies such as neuromorphic computing, optoelectronic computing, and quantum computing. Computing) is not a simple integration. Rather, it is a form of computing that solves real problems by systematically and comprehensively optimizing existing computing methods and resources based on task requirements.

In contrast, the existing major computing disciplines, such as supercomputing, cloud computing and edge computing, belong to different fields: supercomputing aims to achieve high computing power, cloud computing Emphasizing cross-platform/device convenience, edge computing pursues service quality and transmission efficiency.

Intelligent computing dynamically coordinates data storage, communication and computing between edge computing, cloud computing and super computing fields, building various cross-domain intelligent computing systems to support end-to-end End-to-end cloud collaboration, inter-cloud collaboration and supercomputing interconnection.

Intelligent computing should make full use of existing computing technology, and more importantly, promote the formation of new intelligent computing theories, architectures, algorithms and systems.

Secondly, the concept of intelligent computing is proposed to solve the problems in the future integration and development of human-physical-information space.

With the development of information technology applications in the big data era, the boundaries between physical space, digital space and human society have become increasingly blurred.

The human world has evolved into a new space, which is characterized by the close integration of humans, machines and things. Social systems, information systems and the physical environment constitute a dynamically coupled large system. , in this system, people, machines and things are integrated and interacted in a highly complex way, which promotes the development and innovation of new computing technologies and application scenarios in the future.

The above is the detailed content of In five years, the computing power required by AI will exceed 1 million times! Twelve organizations jointly published an 88-page article: 'Intelligent computing' is the antidote. For more information, please follow other related articles on the PHP Chinese website!