<\/a><\/p>\n Despite all the havoc, 2020 has been a good year for tech and a good year for AI.<\/p>\n We already see the green shoots of recovery at the end of 2020 and 2021 holds much promise for growth and technology<\/p>\n Here are five predictions for 2021 for AI – some of which I have covered here before.<\/p>\n<\/p>\n 1) Could GPT-3 lead to a new way in which AI models are developed?<\/strong> I covered this subject in Dec. GPT-3 was the big story for AI in 2020 but the impact of GPT-3 could stretch beyond NLP. It could offer a new way to develop AI applications with profound impact see<\/p>\n Could GPT-3 Change The Way Future AI Models Are Developed and Deployed ?<\/a><\/p>\n Here, I discuss the wider implications of few shot learning models where we focus only on the forward pass complemented by massive models (like GPT-3),<\/p>\n<\/p>\n 2) Training on Edge devices and distributed training<\/strong> – Both training on edge devices and distributed training could have a profound impact on next-generation AI applications like those in healthcare or using 5G. I discussed this trend in The implications of Huang’s law for the artificial intelligence<\/a>. The acquisition of ARM by Nvidia will fuel this trend.<\/p>\n<\/p>\n 3) Cloud Native development becomes the norm impacting AI<\/strong>: As every company tries to become a data company, a cloud-native architecture driven by MLOps and Kubernetes becomes the norm because such architectures can scale cost effectively. Hence, AI models are built and deployed in an MLOps and Cloud-Native environment. I discussed the significance of Kubernetes in An introduction to cloud native applications and kubernetes<\/a><\/p>\n<\/p>\n 4) ML and DL could be a commodity and it will impact the pay of data scientists at the entry-level as we move to decision science<\/strong><\/p>\n In 2021, everyone will deploy ML or DL in some form. Cloud technologies will make simple ML deployments easier. This means that the demand for data scientists will shift to more complex areas considering the overall Hidden Technical Debt in Machine Learning Systems<\/a><\/p>\n So, we could shift from data science to decision science. The output of data science is a model with a performance metrics (for example accuracy). With decision science, we could take this further. We can suggest actions and even execute these actions or perform simulations with ‘what-if’ criteria. That means algorithms like reinforcement learning could be a part of 2021 and beyond. The idea of decision science is elaborated in a post by the VC Matt Turck<\/a><\/p>\n Another area with rising activity is the world of decision science (optimization, simulation), which is very complementary to data science. For example, in a production system for a food delivery company, a machine learning model would predict demand in a certain area, and then an optimization algorithm would allocate delivery staff to that area in a way that optimizes for revenue maximization across the entire system. Decision science takes a probabilistic outcome (“90% likelihood of increased demand here”) and turns it into a 100% executable software-driven action.<\/em><\/p>\n <\/em><\/p>\n 5) Engineering applications will need a new approach to data science<\/strong> – finally, I see more engineering companies explore data science. The current AI\/ ML\/DL market is heavily skewed towards financial services. As more industries adopt AI, a different approach may be needed which I explored in Why do some traditional engineers not trust data science<\/a><\/p>\n<\/p>\n Comments and thoughts welcome<\/p>\n<\/p>\n Image source:pixabay.<\/a><\/p>\n<\/div>\n Go to Source<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<\/a> <\/p>\n