Training the AI

AI technology has been revolutionizing various industries and has already achieved great success in streamlining work processes in the healthcare industry, reducing man-hours spent in administrative tasks in the education sector and reducing overhead costs in the manufacturing world.

 International Data Corporation (IDC) predicts that the compound annual growth rate for global spending on AI will be 50.1%, reaching $57.6 billion by 2021. This would include investments in retail, banking, healthcare and manufacturing, which will make up over half of the worldwide spending on AI.

 One would be inclined to think that developing the algorithms powering deep learning would be the toughest challenge this technology requires but the actual challenge for most algorithms is not their mathematics, but rather their inputs — collating high-quality data that is well-labelled and allows for the training of these models as quickly and efficiently as possible.

 Most of these datasets which are at scale are often proprietary, expensive or time consuming to manually prepare. To provide a solution to this challenge many tech start-ups have created smart platforms that provide data for effective training of AI algorithms to operate in the real world.

 AI.Reverie provides synthetic data — data created in a virtual world rather than collected from the real world at a fraction of the cost for training AI algorithms. The company builds photorealistic virtual worlds to closely mimic any real location where the AI services are to be used. The diverse images and scenarios to help algorithms generalize well and reduce bias.

 Another company, DefinedCrowd offers high-quality training data to help machine learning products reach the market at improved quality and speed. With expertise in speech and NLP technologies, they support a broad range of use cases, from virtual assistants to a customer review, autonomous vehicles, content categorization, pattern recognition, or even surveillance systems. Their clients include BMW, MasterCard, Nuance, and Yahoo Japan.

 Currently, the autonomous vehicle space is one of the trendiest and most advanced fields applying AI. The self-driving cars are being tested by dozens of auto and mobility companies but they aren’t mainstream yet. This is because the computer needs to be well versed with rules of driving. MightyAI is one such computer vision company that trains AI programs to better see the world around them with their prime focus on data-driven autonomous vehicles.

 And then, in the news recently was the Allen Institute for AI (Ai2) training their AI with playing Pictionary that requires guessing the phrase behind a drawing thus helping the machines to gain some understanding of the way concepts fit together in the real world.

 While there is no doubt that to be properly trained, AI needs data. A lot of data. Currently, it's estimated that 90% of the data generated globally is unstructured and we would need many such smart AI platforms to know our world better.

 If you'd like a daily mailer of our Fintech Newsletter - FinTalk, you can subscribe here: @bankofbaroda https://lnkd.in/ffG2dfG