AI concept again in the past two years, is now being gradually fall application stage, however, the general feedback from vendors and users point of view, the application of artificial intelligence landing mass does not seem to imagine the well, began to have more and more such scene fragmentation, high application costs, laboratory scene to a big gap between the practical effect of the scene is exposed and other issues, which has become the current stage of the application process AI landing a new pain point. The reason for producing these pain points, and AI developers lack of strong development platform and development tools in the development process are not unrelated.
Bitter AI developers, who understand
For example, if an AI developer, want to train a model to develop a face recognition application, it may be what problems it?
First of all, the first question might encounter is data labeling . Because the first step of this work is to find the open source face data collection and tagging data for each face, open face data set is not difficult to find, such as Labelled Faces in the Wild, CASIA WebFace , LFW , etc. data collection on the internet everywhere, easy to find. But the data label is a big problem, and requires a lot of man-hours.
The second problem encountered is probably the parameter adjustment of. In the training data, the most important job is the training of face data to adjust the parameters, there will be a lot of parameter adjustment in the training data, due to the different tone participants convergence speed and accuracy of the final model, the parameter adjustment is Face recognition program to improve the recognition rate of the key steps, and this part of the work due to the need to repeatedly test optimization, usually making a lot of effort, but how in the shortest time to find the optimal parameter configuration, which is in the development of face recognition the second major problem facing applications.
The third problem is that training time is too long. After the parameter adjustment, on the need for large-scale distributed training. And this large-scale distributed training, it takes a long time, this lengthy time consuming for developers, is the main reason that AI can not improve the efficiency of application development.
Finally, there is a training model deployment problems. Under normal circumstances, the deployment model is very complex, you need to write application code into the system, and need to be maintained and updated at any time. Without a simple and convenient platform to completely rely on manual operation, then the workload is very large.
So, think about it, a face recognition application is not complicated, you need to face so many problems, not to mention more sophisticated AI applications it? And this is not just an isolated AI developers and prospective developers will encounter. AI has a lot of dreams, young blood boiling, is likely to be blocked these problems outside the gate AI application development.
Recent drops, millet and other big companies are downsizing, how programmers spend this wave of layoffs?
Pain AI developers, Huawei cloud really understand
Pain AI developers, Huawei's cloud really understand. Like on the above recognition applications, as long as a Huawei cloud ModelArts solution will get every minute.
In fact, Huawei cloud AI developers to help it, already layout.
Last year, Huawei released a Pratt & Whitney AI strategy, let AI high but not expensive, so businesses can afford to put to good use, and must be assured. To this end, Huawei released the full stack whole scene AI solution. The so-called full-stack, from a technical perspective, it refers to the function of AI solutions include Huawei force up to 256T FLOPS count, high-performance AI chip, chip enable, training and reasoning and application enablement framework, including full-stack solutions; so-called full scene, AI refers to the Huawei solution is the industry's only cover public cloud, private cloud, a variety of edge computing, networking and consumer industry terminal terminal high, medium and low whole scene, count the force across the whole scene a few million times solution.
Calculated from the Atlas of artificial intelligence platform, ModelArts stop AI development platform, to stop AI communications development platform, Huawei's cloud services platform and ServiceStage micro intelligent edge platform, Huawei AI solution enables high, medium and low whole scene all unified architecture, a development applies to all deployment scenarios, migration, collaboration, reduce AI developers in the development process of each layer development threshold, will leave the Huawei cloud complex, simply leave the majority of developers, greatly increasing development efficiency AI developers. Huawei's cloud through this "extraordinary" the whole scene full stack AI development solutions, AI developers will "extraordinary achievement."
To help AI developer friends more in-depth understanding of Huawei's cloud full stack AI development solutions whole scene, Huawei cloud organized the "DevRun • Select the extraordinary, Huawei cloud developers Salon 2019", at present, the activity has been in Beijing, Suzhou, Nanjing Hangzhou four cities held by the majority of developer friends a warm welcome. In the July 5, the event will come to Shenzhen. In the salon, the cloud will be invited Huawei Huawei cloud technology large coffee, combined with practical cases, developers friends how based Huawei Atlas AI chip / module, ModelArts development platform for one-stop AI, AI intelligent communication platform, Huawei's intelligent edge services and cloud-native application service quickly implement AI application development, allowing developers relying on Huawei's cloud this "extraordinary" platform developed to combat the "extraordinary" application.
Activities agenda:
The so-called "extraordinary selection, will be enormously successful," developer friends, do not miss this rare opportunity!
Click to read the original text, sign up now!