The growth of deep learning
Deep learning is overtaking more classic machine learning methods according to a recent study of AI industry leaders across Europe, which included Amazon, Google and JP Morgan.
The report, entitled Deep Learning: Opportunities and Best Practice, published by CognitionX and Peltarion, highlighted several factors that are enabling wider uptake, such as increased data availability, access to bigger memory and large-scale GPUs and the flexibility of neural networks.
“These changes did not take place until a few years ago, which is why things are moving so much faster now,” says Björn Brinne, Peltarion’s Chief AI Officer. “It’s opened the door for more organisations to start using deep learning (DL), because using neural networks requires a lot more data, memory and compute power than classical machine learning.”
DL also provides a number of features that other categories of machine learning algorithms don’t. For example, DL algorithms allow for learning from data with minimal pre-processing; they learn to extract relevant features from data – images, audio, text without requiring humans to manually pick inputs notes Joshua Saxe, Chief Scientist at Sophos. “They make it easier to train on massive, internet-scale datasets, and simply work better on classical machine learning problems,” he points out.
The hottest deep learning trends
DL has made several breakthroughs over the past decade and identifying patterns and classifying image data is where we’ve seen the biggest wins to date. This has meant it’s been particularly helpful to the healthcare sector, where it can be used to analyse scans.