04-11-2020 | | By Robin Mitchell
With the rising importance of AI and the increasingly more competitive service market, companies are looking to not only integrate AI into their products but also developing methods for improving AI systems. Why is AI becoming so popular, what applications can AI help with, who are SigOpt, and what does this acquisition mean for Intel?
Artificial Intelligence, or AI, is the ability for a computer system to be able to apply intelligence to a problem without needing to perform a large number of comparisons. For example, a computer system can be programmed to identify how to recognise pictures of cats by comparing the image to every single image of cats ever taken until an identical image is found, but this is incredibly slow and inefficient. Instead, an AI system is shown a series of pictures and its attempt to determine if there is a cat in the picture. If the answer form the AI system is correct (i.e. training the system), then its internal structure is slightly modified to reflect this answer. As more and more images are shown to the AI system, it becomes better at identifying cats. Eventually, it can identify any cat in any image without needing a like-for-like image database of all cats to compare to.
Thus, unlike traditional computer programs, AI can be trained from data and then begin to operate by themselves while often providing high success rates. The constant use of AI also trains it over time, improving its performance, and this, in turn, makes it more powerful. It is this ability to be able to learn from past experiences and continually improve itself that has made AI so popular.
AI can be applied in a wide range of different applications, but those that benefit the most are ones which have data that can be in different forms but have the same meaning. Such examples of varied applications include facial recognition, speech recognition, and object recognition.
Facial recognition is an area in particular that has greatly benefited from AI. The first step in facial recognition is to perform face detection, whereby a face can be identified in an image. While it may be obvious to humans, to a computer, two faces can look completely different. Two different faces convey the same information, a face, but both have their own unique characteristics, including eye shape, nose position, and mouth size.
Speech recognition is another example of how different pieces of data can convey the same meaning but take different forms. The many accents around the world mean that there are many different ways of saying the same word, and this can be hard for a computer system to decode. Thus, AI has proven to be incredibly valuable as it allows for a system to determine words and ignore accents.
SigOpt is a company that specialises in AI parameter optimisation, and their platform creates a standardized, scalable, enterprise-grade optimisation platform as well as an API for modelling pipelines. The platform helps to optimise hyperparameters that are used in AI systems to provide better analyses as well as tuning of systems. The black-box nature of SigOpt allows for designers to focus on improving AI training and performance instead of working on parameter determination, thus speeding up AI development. SigOpt can be integrated into most workflows irrespective of the machine learning platform used, model management, infrastructure, or library, and SigOpt even provide integration services, so designers do not need to change their system to suit SigOpt.
Recently, Intel announced that it has agreed with SigOpt to acquire them for an undisclosed amount. This move is one but of many demonstrating how many large semiconductor companies are purchasing and merging with cutting-edge technology providers as the world enters the next phase in technological development; the mass integration of AI and IoT. As previously stated, AI will become an incredibly important component in next-generation products, and a company that can not only develop AI systems but fine-tune them for fully optimised performance will be able to produce effective AI designs that can be targeted on mobile platforms which require reduced energy consumption.
According to Intel, the acquisition of SigOpt allows for Intel to use SigOpt products and services across its whole AI hardware product range to both accelerate and grow its AI offerings to developers and consumers. Intel already produces a range of hardware aimed at the AI market including the Intel Xeon scalable processors, Intel Movidius Vision Processing Units, FPGAs, and GPUs which would all benefit with SigOpt integration. With customers in the Fortune 500, SigOpt is a fast-growing company with a varied customer base, including universities, researchers, and even governments. The deal, which is expected to be completed by the last quarter of 2020, will still allow existing customers to use SigOpt. The development team at SigOpt will join and work with the machine learning performance team under Raja Koduri, Senior VP, chief architect, and general manager of Architecture, Graphics, and Software at Intel.