The shortcomings of the "big model" and the strengths of the "knowledge map"

In the past two years , Large models in the field of artificial intelligence are hot . Take the field of naturallanguageprocessing as an example , since BERT Born in the sky , Scores soared in various evaluations , After surpassing the human level in the Stanford reading comprehension evaluation collection , Various larger and larger naturallanguageprocessing models are emerging , And constantly refresh new records in various evaluations .

chart 1 This is the case of different pretreatment models in recent years , It can be seen that the scale of the model increases exponentially . thus , Many people have made the bigger the model, the better （Larger model, better performance） believe firmly , And gradually formed AI In the field of “ The arms race ”.

chart 1 “ Big model ” Exponential growth of parameters

exactly ！ If you only consider various evaluations , If the goal is to brush the list , This trend is indisputable . Whether Microsoft or OpenAI Of Megatron-Turing NLG, It's still Google PaLM, As well as the Enlightenment of domestic research institutes such as Zhiyuan 、 Baidu Wenxin 3.0 etc. , It is obvious to all that the evaluation results are outstanding . But all this is not without criticism .

One side , The power of this oversized model is controversial , What exactly does it have “ Super power ”, Or is it because the training corpus is rich enough that the model is just “ memory ” These content ？ Further , This big model based on deep learning , And its natural and unexplained features .

On the other hand , This large model built on supercomputing resources and data resources , In the face of real application, it seems that the heart is more than the strength is insufficient , In most cases , It only applies to “ Brush a list ”, Even if you have a lot of money , And you can't really use those that look very nice “ Big model ”.

Besides , In many fields of professional application , For example, failure analysis of cutting-edge medical device manufacturing , Because its corpus is so small in a wide range of data , There is no advantage in using these large models .

therefore , People of insight in this “ Super computing power ”+“ Huge amounts of data ”+“ Big model ” Beyond the paradigm of , Put forward to “ Algorithm ”+“ Calculate the force ”+“ data ”+“ knowledge ” The new paradigm of .

such as , Academician Zhang cymbal said “ Human intelligence cannot be learned through simple big data learning , What to do with that ？ It's simple , Add knowledge , Let it have the power of reasoning , The ability to make decisions .”

Knowledge map is the latest method to store and express knowledge in the field of artificial intelligence , At present, it is driving the further development of artificial intelligence , It is also considered as one of the core technologies to realize cognitive intelligence .

in fact , For all living beings , These big models “ Out of reach ”.

In recent years, I have been doing technical research on naturallanguageprocessing and knowledge mapping 、 Work on product development and industrial implementation . During this period, I visited leading enterprises in all walks of life , Understand that whether it is a financial giant , The leader in each segment of the manufacturing industry is still a monopoly large state-owned enterprise , These big models are not used in their business , To improve efficiency 、 cost reduction 、 Enhance competitiveness, etc . Even the AI giants have not made good use of these big models .

And the cognitive intelligence technology with knowledge map as the core , Because I joined “ knowledge ” This is a summary of human development experience , Make it possible to get rid of “ Big model ” The defects of , It is more suitable for practical application in all walks of life .

also , Cognitive intelligence technology with knowledge map as the core , Because it has less computing resources , The inference can explain , Thus, it has great advantages in product implementation and industrial application .

In order to introduce this kind of cognitive intelligence technology with great prospects and very practical industrial applications , I put a lot of effort into it , Sort out the research results of cutting-edge technologies of knowledge map , Summarize the experience in artificial intelligence product development and industrial application for more than ten years , It has written down the recommendation of more than ten well-known experts in the academic and business circles 《 Knowledge map ： Cognitive intelligence theory and practice 》 A Book .

Content abstract

This book systematically and comprehensively introduces the core technology of knowledge map , Existing macro integrated technical system , There are also key technologies and algorithm details , The content includes ：

Methodology of knowledge map pattern design —— Six Tao method ;

Entity extraction and relation extraction in knowledge map construction ;

Attribute graph model and graph database in knowledge storage , highlighted JanusGraph Distributed graph database ;

The basis of graph theory in Knowledge Computing , And centrality 、 Community detection and other classical graph calculation algorithms ;

Logical reasoning in knowledge reasoning 、 Geometric transformation reasoning and deep learning reasoning , And its programming examples .

Last , This book focuses on finance 、 Take the application scenarios of the three major industries of medical treatment and intelligent manufacturing for example , It combs the application value and application form of knowledge map .

chart 2 《 Knowledge map ： Cognitive intelligence theory and practice 》 Content framework

The book features

People who are suitable for reading

This book is suitable for practitioners and researchers in the artificial intelligence industry to systematically learn the knowledge map , It is also suitable for front-line engineers and technicians , And can serve as enterprise management personnel 、 government agent 、 Policy makers 、 Reference materials for public policy scholars , And University computers 、 Reference materials for teachers and students of Finance and artificial intelligence and teaching materials of training schools .

Knowledge atlas as knowledge acquisition in artificial intelligence 、 Storage and technology used , Is to make up for deep learning and “ Big model ” A good remedy for weakness , It is an effective method to stitch up the cutting-edge technology theory and industrial application of cognitive intelligence .“ The sun and the moon are beautiful , A hundred grains of grass and trees are beautiful ”, Knowledge is an essential nutrient for the continuous progress of artificial intelligence , Therefore, the knowledge map is also the ladder of artificial intelligence progress .

hope 《 Knowledge map ： Cognitive intelligence theory and practice 》 This book can provide some help for readers in exploring this artificial intelligence , It can make a modest contribution to the industrial application of cognitive intelligence .

Scan code for details of this book