While news outlets forecast the coming AI revolution, practitioners know the AI revolution has already started. Companies like Google, Amazon, Facebook, IBM, and Microsoft are pouring millions into funding for artificial intelligence and machine learning, processing billions of data points every day. So far, 2017 has seen over $6 billion in AI funding for independent ventures as well. Hedge funds use AI to trade billions of dollars of securities in fractions of a second, and new advances in transportation, shipping, health care, and retail promise to make AI-implementation commonplace.
The global market for artificial intelligence is projected to grow to over $3 trillion by 2024, an astronomical growth rate as AI makes our lives easier and more efficient. However, in order to achieve that lofty growth, the artificial intelligence market will have to overcome several bottlenecks that limit adoption. Neuromation offers a solution to the current fragmented AI landscape, increasing access and speed with which AI technologies can be deployed.
The Supply & Demand Gap in AI
While demand for AI is high and only growing, the supply of good AI models and data is slowing down the development process. At Neuromation, our solution addresses the need for data sets, computing power, and a marketplace for AI models.
Good Data to Train Neural Networks
We’re living in the age of big data. It’s fairly easy to obtain huge data sets in virtually any field where you’d like to apply AI. The problem is those data sets haven’t been described and annotated so that they’re usable for training neural networks. Traditionally, humans have to do the job of annotation by hand, making dataset creation challenging, costly, and prone to bias. For example an average retail set has 150,000+ items. A deep neural network needs thousands of labeled photo examples for each item. Doing this by hand would take years.
Neuromation’s solution utilizes synthetic data to create training datasets for AI applications. We automate the process of labelling and annotating data by creating datasets that are tailored to the task at hand. Using synthetic data drives down the cost and time required to create a training dataset.
Computing Power to Learn from Big Data Sets Quickly
Another limitation of current AI is the processing power available to developers and researchers. The computing power required to train a neural network often means that an algorithm make take days or weeks in training with a single computer. Often times, the model needs to be tweaked and run again, making the development process is slow and onerous.
We propose tapping the existing processing power of miners on blockchain networks. Researchers and companies working on AI are willing to offer compensation in return for processing power, and allowing miners to earn money from working on machine learning problems means greater rewards for the miners and faster processing times for AI developers. The Neuromation white paper estimates miners could earn 3 to 5 times more mining NeuroTokens versus other types of mining.
Sharing, Importing, and Requesting Models
The next generation of development in AI will require tools for uploading and sharing AI models on a marketplace for use in various applications. In addition to providing synthetic data and computing power on the Neuromation marketplace, we’ll also build, train, and maintain models for clients and allow clients to import their own models. In the year after Neuromation’s launch, we expect to sell over 1,000 models, and the model marketplace will be one of our primary revenue drivers.
Practical Applications of Neuromation
Synthetic data and distributed processing have exciting practical applications for everyday businesses. For example, Neuromation has partnered with leading retail brands to identify products on a shelf and give suggestions about shelf layout and efficiency based on images taken from a smartphone. Using our technology, we’ve achieved a 95%+ accuracy in inventory recognition. The market for AI in retail reaches up to 40 billion images per year, and 85% of customer interactions will be handled by AI, according to Gartner.
Neuromation is also applying synthetic data and dynamic modeling to industrial applications to monitor and predict factory or shipping operations in real time using computer vision. In addition, we’re looking into how computer vision and neural networks can help with biotech diagnostics and drug discovery. Many questions and hypotheses that were too variable or work-intensive to investigate using traditional means can now be answered using computer vision.
Learn more about Neuromation by visiting our website. We’re also available via chat to answer any questions!
