Open-source projects would become one of the AI trends in 2019. That is an opinion of multiple AI specialists, including the Executive Director at LF Deep Learning Foundation Ibrahim Haddad. So, how does open source influence AI advance even now?
Open source in the course of AI advance
The speed of the development of computing capacities and innovations has tremendously influenced the AI sphere. New machines allow to solve tasks of any complexity in a more effective way, new network topologies emerge, more and more data for analyzing is accumulated. As provided by consulting firm Frost & Sullivan, the global AI market would reach $52.5 bn with an annual growth rate of 31% by 2022.
Open-source developments would become one of the major directions due to the severe shortage of AI experts. These are projects with open source, which enables remote work after neural network building and training. Currently, there exist open-source developments shared via the Internet by such companies as Microsoft and small startups.
For example, Google turned Kubernetes from a closed internal system into an open-source project. So, what is the point of maintaining the same principle? As reported by the Executive Director at LF Deep Learning Foundation Ibrahim Haddad, this would allow to fulfill important and large-scale ideas more effectively as well as develop the community of AI experts.
Open-source projects in the field of AI
Large companies that were among the early adopters of AI open-source projects have already benefited from this tendency. For example, cloud services and open source help Microsoft provide solutions to clients.
By the way, today’s most demanded and popular libraries such as TensorFlow, Keras, and Caffe are also open-source. Developers across the globe are getting more and more interested in the creation of such projects. Meanwhile, the number of unusual and interesting solutions grows annually. Open-source AI-based algorithms can paint, act as a voice assistant, produce video games graphics, and more.
Algorithm Obvious produces paintings
The France-based art collective called Obvious created a generative adversarial network capable of producing paintings. ‘Portrait of Edmond Belamy’ was sold at auction for $432 500. The algorithm works on a database with 15 000 man-made portraits of various periods. Having analyzed them, the network produced a series of its 11 unique works of art.
The portrait is printed on canvas by means of inject printing, put into a frame and has a signature in the form of a mathematic formula demonstrating the connection between two parts of network algorithms – generator and discriminator. The art collective took Robbie Barrat’s development as a base. He later was surprised commenting: “I was writing this source in my free time while being a student”.
NVIDIA uses AI to produce a real universe in video games
NVIDIA came out with an AI-imbedded model capable of producing universes without modeling and a traditional graphics engine. This product is undergoing the development stage: it has an open source, which allows all enthusiasts to take part in the debugging process. The project utilizes deep learning for a neural network that is trained to analyze a video and builds a 3D universe.
The algorithm can produce vehicles, trees, buildings but there remains lot of details to be improved. The company also uses motion modeling technology (sports, walking, dancing) and can apply it for different characters in a real-time mode. The project was initially created to improve the quality of old video games.
Oracle GraphPipe for AI models serving
Oracle GraphPipe is a special tool that simplifies machine learning models serving. The developer provided access to its source code and placed on GitHub for free use. The tool works with projects based on such libraries as TensorFlow, MXNet, Caffe2, and PyTorch. They are applied for IoT devices and AI-powered platforms.
The tool helps deploy models without creating user API and does away with copying data while deserialization. This solution allows to seamlessly deploy models out of the existing frameworks. Oracle GraphPipe offers open-source tools necessary for work with AI, for example TensorFlow framework.
Mycroft – open-source AI voice assistant
Mycroft is an alternative to voice assistants of Apple, Amazon, Google, and Microsoft. Their technologies of natural language processing Siri, Alexa, and Cortana are patented and confidential. By contrast, Mycroft uses open source. It was projected and created by Joshua Montgomery and his team in the US.
Mycroft can be downloaded and used for free. Developers have an opportunity to change its source in order to extend and improve the functions of NLP. More than 700 independent experts are currently contributing to the software. Within the project, there were created smart speakers Mark 1 and Mark 2.
Open-source AI Baidu for cancer diagnostics
The Chinese search engine heavyweight launched a healthcare project fueled by open-source artificial intelligence aimed at effective cancer diagnostics. The AI algorithm can analyze biopsy images and recognize cancer cells. As shown by the results of the company’s testing, the neural network with deep learning is better at recognizing cancer compared to pathologists.
This AI has open source available to any healthcare employee. The project was created to combat cancer and simplify doctors’ work (not replacing them).
Open-source AI projects expose big potential and can be observed in completely different directions. These are healthcare developments, assistants, and graphics production. Such projects and large-scale ideas not only bring together developers but also create a competitive environment for the best application of AI technologies. What is more, open-source communities may solve the problem of staff shortage in the AI realm.