Mai Ken stannum predicts: To 2025 global artificial intelligence applies market dimensions total value to will reach 127 billion dollar, artificial intelligence will be the breakthrough site that numerous intelligence industry develops.
Undertake combing through distributinging to artificial intelligence industry, offerred plan of modes of life and relation to their environment of artificial intelligence industry, main component is:
Graph graph of modes of life and relation to their environment of industry of 1 artificial intelligence
Be opposite the key below 4 respects spread out service of technology of information of the intelligent infrastructure construction that condition of core line of business includes, intelligence and data, intelligence, intelligent artifacts the introduction, sum up application of artificial intelligence industry and industrial development trend.
Intelligent infrastructure
Intelligent infrastructure offers computational ability to prop up for artificial intelligence industry, its limits includes frame of chip of intelligent sensor, intelligence, distributed computation to wait, it is the important safeguard that artificial intelligence industry grows.
1. intelligence chip
Intelligent chip can be divided from applied angle for training and inference two sort. From the point of deploy setting, can divide carry kinds two steps are big for high in the clouds and equipment.
Because training course involves huge training data and construction of complex deepness nerve network, need giant computational dimensions, group of chip of main use intelligence will finish. Measure photograph comparing with the computation of training, computational quantity of inference is less, but still involve many matrix operation. Current, training and inference come true in high in the clouds normally, only very high to requirement of real time sex equipment can be handed in by equipment end is undertaken handling.
Will look by technical framework, intelligent chip can be divided for general kind of chip (CPU, GPU, FPGA) , be based on FPGA half custom-built change completely chip, custom-built turn ASIC chip, kind cerebral computation chip (IBM TrueNorth) . Additional, main artificial intelligence processor still has the DPU, BPU, NPU, EPU artificial intelligence chip that applies to different setting and function.
As Internet user quantity and data dimensions expand quickly, artificial intelligence expands pressing to the requirement of computational function growth, the demand that to CPU computational function promotes exceeded rub the growth rate of Er law. In the meantime, get account of technology of be confined to, traditional processor function also cannot according to rub Er law continues to grow, chip of intelligence of the generation below development is imperative.
Intelligent chip of future basically is develop in two direction:
It is the chip that imitates human cerebra structure;
2 it is quanta chip.
Intelligent chip is the strategic commanding elevation of artificial intelligence times,
2. intelligence sensor
Intelligent sensor is the sensor that has information processing function, intelligent sensor contains microprocessor, have collect, the function such as news of processing, trade, it is sensor compositive turn the result that combines with microprocessor photograph.
Intelligent sensor belongs to the nerve ending of artificial intelligence, use at feeling outside ambient in the round. The large-scale deploy of of all kinds sensor and application created indispensable condition to realize artificial intelligence. Different application setting, be like: How is intelligence prevented, the medical treatment of intelligent household, intelligence requirement that put forward to differ to sensor application.
Future, extend ceaselessly as what artificial intelligence uses a field, the market will be perpetual to the demand of sensor grow in quantity, market dimensions hopeful broke through 460 billion dollar 2020. Future, gao Min is spent, high accuracy, high reliability, miniature is changed, compositive turn the important tide that will become intelligent sensor to develop.
Frame of 3. distributed computation
Face huge data processing, sophisticated knowledge reasoning, groovy stand-alone computation mode cannot have been propped up. So, it is OK that computational mode must assign huge computational job must assign huge computational job into small stand-alone susceptive computation task, namely the computational framework that technology of cloud computation, margin calculation, big data provided a basis.
Current and popular distributed computation frame, be like: OpenStack, Hadoop, Storm, Spark, Samza, Bigflow. All sorts of deepness opening a source learn frame to also emerge in endlessly, include TensorFlow, Caffe, Keras, CNTK, Torch7, MXNet, Leaf, Theano, DeepLearning4, Lasagne, Neon to wait a moment among them.
