In recent years, artificial intelligence has been developed in great leaps. Many applications in life have imported artificial intelligence technology to bring convenience and fun for life. This text will introduce the development trend of artificial intelligence and relevant solutions launched by NVIDIA which has a leading position in artificial intelligence field.
Differences among Artificial Intelligence, Machine Learning and Deep Learning
The concept of Artificial Intelligence (AI) has been discussed and presented in novels and movies, some of which talked about AI’s help for human beings or even mentioned whether AI has consciousness and emotion, while others described AI’s threat for human beings with a doomsday scene in the future. No matter if AI has advantages or disadvantages for human beings, the development has been an irresistible trend and it has been advanced toward the direction of improving human life. What extent AI will be evolved finally in the future is still inconclusive now.
AI is a kind of broad concept, which refers to the intelligent state presented by human-made machines and generally refers to the technology for presenting artificial intelligence via computer programs. Except for AI, Machine Learning and Deep Learning are also been mentioned frequently. We can use a concentric circle to define those nouns above in a range in which artificial intelligence is the peripheral circle, then the in-between machine learning and the middlemost is deep learning.
The most basic practice of machine learning is using algorithm to analyze data, learn more from the process and then conduct confirmation or prediction of the object in the world. Therefore, rather than using a set of specific commands to manually encode software programs for finishing a particular task, it is better to use mass data and algorithm to “train” machines to make them learn how to execute a task. This is the basic concept of machine learning.
Deep learning is a kind of technology for achieving machine learning, which originated from the artificial neural network achieved by the early machine-learning-human thinking mode. In most cases, the artificial neural network can change internal structure based on external information, which is a kind of self-adaptation systems or normally treated as being equipped with learning function. The artificial neural network has been used for solving various problems like machine vision and automatic speech recognition, which are difficult for traditional rule-based programs to solve.
After the invention of computers in 1940s, the concept of AI has been discussed in educational circles, but lot of ideas remained theoretical because they were limited by the arithmetic speed of computers. Since 2015, along with the rapid development and wide availability of Graphics Processing Unit (GPU), parallel processing became faster, cheaper and stronger, which made it possible for achieving deep learning of artificial neural network and also expanded the whole field of AI like autonomous vehicles and intelligent voice assistant etc. With the help of deep learning, artificial intelligence can even reach our long-standing science dreams such as humanoid robots that can do everything for human beings.
NVIDIA speeds up artificial intelligence applications
NVIDIA’s invention of the GPU, In 1999 sparked the growth of the PC gaming market, redefined modern computer graphics, and revolutionized parallel computing. More recently, GPU deep learning ignited modern AI — the next era of computing — with the GPU acting as the brain of computers, robots, and self-driving cars that can perceive and understand the world.
Based on over 15 years of CUDA development experience, NVIDIA lays the foundation for compute-intensive application ecosystem with the advantage of underlying CUDA primitive function strengthened by adopting GPU accelerated libraries such as linear algebra, advanced mathematics and parallel computing. NVIDIA software library provides the simplest GPU introduction methods and the adoption of NVIDIA CUDA-X AI library can achieve various kinds of algorithm of machine learning and analysis.
The development of AI applications should first train deep neural networks with large datasets. GPU accelerated deep learning framework launched by NVIDIA provides flexibility for designing and training user-defined deep neural networks and provides the interface to common programming languages like Python and C/C++ etc. Each main deep learning framework (such as TensorFlow and PyTorch etc.) has been accelerated by GPU. Therefore, data scientists and researchers can improve work efficiency in several minutes without any GPU programming.
As for AI researchers and application developers, the adoption of Tensor-core NVIDIA Volta and Turing GPU can accelerate training speed and improve deep learning performance. FP32 and FP16 mixed precision matrix multiply can greatly improve data throughput and reduce AI training time. In addition, NVIDIA deep learning software development kit (SDK) includes high-performance libraries which have achieved building block APIs and can be used for training and inference in the applications directly.
NVIDIA TensorRT™ is an SDK used for high-performance deep learning inference which includes deep learning inference optimizer and runtime that can provide low latency and high throughput for deep learning inference applications. During the inference process, the operating speed of applications based on TensorRT is 40 times faster than that of the platform only based on CPU. Depending on TensorRT, it can optimize all the neural network models being trained in the main framework and conduct calibration for low precision with high accuracy. Then it will be finally deployed to hyperscale data center, embedded or automotive product platforms.
NVIDIA has also founded the Deep Learning Institute (DLI) which provides hands-on training related to AI, accelerated computing and accelerated data science. Through DLI self-paced on-line training for individual, face-to-face workshops for teams and downloadable course materials suitable for college educators, researchers, data scientists and students can acquire practical experience via cloud GPUs and get a certificate of competency to support professional growth.
NGC of NVIDIA is the GPU software optimization center for deep learning, machine learning and high-performance computing (HPC), in which data scientists and researchers can concentrate on building solutions, collecting opinions and delivering commercial value.