The global pandemic has transformed the way we transact. With much of the world moving online, e-commerce, cloud computing, and enhanced cybersecurity measures are just the tip of the iceberg when it comes to assessing current trends in data analysis.
Managing risk and keeping costs low have always been important considerations for businesses. However, having access to the right machine learning technology that can analyze data effectively is becoming crucial for any company wanting a competitive edge.
Our roundup of the top data analysis trends for 2022 and beyond should give our creators a good idea of where the industry is heading.
By keeping on top of trends in data science and adjusting their models to fit current standards, creators can make their work truly invaluable. Whether these data analysis trends inspire you to brainstorm new models or update the existing ones in your toolkit, the choice is entirely yours.
Following the trend in computer gaming, with user-generated content (UGC) monetized as an integral part of gaming platforms, we see similar monetization happening in data science. This starts with simple models, such as classification, regression, and clustering models all repurposed and uploaded to dedicated platforms. These are then made available to a global marketplace of business users who want to automate everyday business data and processes.
This will be quickly followed by deep model artifacts, such as convnets, GAN’s, and autoencoders that are tuned and applied to solve business problems. These are designed to be safe in the hands of commercial analysts, rather than teams of data scientists.
Data scientists selling their skills and experience as consultancy gigs, or uploading models to code repositories is nothing new.. However,2022 will see monetization of these skills through double-sided marketplaces, giving a single model access to a global marketplace.
Whilst most research is understandably focused on pushing the boundaries of complexity, the reality is that training and running complex models can have a big impact on the environment.
It’s predicted that data centres will represent 15% of global CO2 emissions by 2040, and a 2019 research paper “Energy considerations for Deep Learning” found that training a natural language translation model emitted CO2 levels equivalent to four family cars over their lifetime. Clearly, the more training, the more CO2 is released.
With greater understanding of environmental impact, organisations are exploring ways to reduce their carbon footprint. Whilst we can now use AI to make data centres more efficient, the world should expect to see more interest in simple models that perform as well as complex ones for solving specific problems.
Realistically, why should we use a 10 layer convolutional neural network, when a simple bayesian model will perform equally well while using significantly less data, training and compute power? “Model efficiency” will become a by-word for environmental AI, as creators focus on building simple, efficient, and usable models that don't cost the earth.
Not unlike the space tech race of Musk and Bezos, big tech have their own exciting race: who has the biggest deep learning model?
In 3 years the number of parameters in the largest models rose from 94m parameters in 2018 to a staggering 1.6 trillion in 2021, as Google, Facebook, Microsoft, OpenAI, etc push the boundaries of complexity.
Today, these trillions of parameters are language based, allowing data scientists to build models that comprehend language in detail, enabling models to write human level articles, reports and translations. They can even write code, develop recipes and understand sarcasm and irony in context.
In 2021 and beyond, we can expect similar human level performance from vision models which are capable of recognising images without the need for huge data sets.
