The technology industry continues to be a hotbed of innovation, with activity driven by the increasing demand for artificial intelligence (AI) applications in various industries, the availability of vast amounts of data, and the need for scalable and cost-effective computing resources, as well as growing importance of technologies such as distributed computing, parallel processing, and specialised hardware accelerators like graphics processing units (GPUs) and tensor processing units (TPUs). In the last three years alone, there have been over 3.6 million patents filed and granted in the technology industry, according to GlobalData’s report on Innovation in Cloud: Neural net architecture. Buy the report here.
However, not all innovations are equal and nor do they follow a constant upward trend. Instead, their evolution takes the form of an S-shaped curve that reflects their typical lifecycle from early emergence to accelerating adoption, before finally stabilising and reaching maturity.
Identifying where a particular innovation is on this journey, especially those that are in the emerging and accelerating stages, is essential for understanding their current level of adoption and the likely future trajectory and impact they will have.
300+ innovations will shape the technology industry
According to GlobalData’s Technology Foresights, which plots the S-curve for the technology industry using innovation intensity models built on over 2.5 million patents, there are 300+ innovation areas that will shape the future of the industry.
Within the emerging innovation stage, finite element simulation, deductive databases, and neural networks for data storage are disruptive technologies that are in the early stages of application and should be tracked closely. Multi-programming operating systems, AI assisted network management, and grid computing are some of the accelerating innovation areas, where adoption has been steadily increasing. Among maturing innovation areas are software-defined wide area network (WAN) and fog computing, which are now well established in the industry.
Innovation S-curve for cloud in the technology industry
Neural net architecture is a key innovation area in cloud
Neural network architecture encompasses a collection of algorithms and methodologies employed in the creation of artificial neural networks. These networks consist of interconnected nodes, referred to as neurons, organised into layers. Each neuron is connected to multiple other neurons and employs mathematical functions to transform inputs into outputs. The output of one neuron serves as input for subsequent neurons in the following layer, allowing for further data processing. This iterative process persists until the desired outcome is attained.
GlobalData’s analysis also uncovers the companies at the forefront of each innovation area and assesses the potential reach and impact of their patenting activity across different applications and geographies. According to GlobalData, there are 40+ companies, spanning technology vendors, established technology companies, and up-and-coming start-ups engaged in the development and application of neural net architecture.
Key players in neural net architecture – a disruptive innovation in the technology industry
‘Application diversity’ measures the number of different applications identified for each relevant patent and broadly splits companies into either ‘niche’ or ‘diversified’ innovators.
‘Geographic reach’ refers to the number of different countries each relevant patent is registered in and reflects the breadth of geographic application intended, ranging from ‘global’ to ‘local’.
Patent volumes related to neural net architecture
Source: GlobalData Patent Analytics
Samsung is a leading patent filer in neural net architecture. The company’s patents are aimed at invention describing a convolutional layer in a convolutional neural network that uses a predetermined horizontal input stride and a predetermined vertical input stride that are greater than 1 while the hardware forming the convolutional layer operates using an input stride of 1.
Each original weight kernel of a plurality of sets of original weight kernels is subdivided based on the predetermined horizontal and vertical input strides to form a set of a plurality of sub-kernels for each set of original weight kernels.
Each of a plurality of input feature maps (IFMs) is subdivided based on the predetermined horizontal and vertical input strides to form a plurality of sub-maps. Each sub-map is convolved by the corresponding sub-kernel for a set of original weight kernels using an input stride of 1. A convolved result of each sub-map and the corresponding sub-kernel is summed to form an output feature map (OFM).
Neural network architecture has revolutionised the field of artificial intelligence. By harnessing the power of cloud computing, neural networks can be deployed and trained on massive amounts of data with unprecedented speed and efficiency. It enables scalable and cost-effective solutions, as it provides access to high-performance computing resources and allows for distributed training across multiple machines.
To further understand how cloud is disrupting the technology industry, access GlobalData’s latest thematic research report on Cloud Computing – Thematic Intelligence.