Deep learning matters for one simple reason
Performance improves as data increases
I recently attended a presentation on TensorFlow. Towards the end someone asked the speaker, “what is the difference between deep learning and machine learning and why should I care?” The speaker replied with a rambling discussion of function generalization, overfitting, regularization … and the audience was even more confused.
The answer to “why should I care” is actually quite straightforward.
Why does deep learning matter?
I created this diagram based on Andrew Ng’s amazing AI talk.
Humans perform well even with small amounts of data. We can reason and make predictions based on context. Unfortunately our brains are limited by what we can remember and process.
Traditional machine learning techniques like linear regressions and random forests are based on specific algorithms. Engineers pick appropriate algorithms for a given task, pre-process the data and tune parameters. These techniques can solve common business problems but tend to plateau at large data volumes. Stripe is using traditional machine learning techniques to build its new fraud system.
Deep learning is based on neural networks, a machine learning technique (loosely) inspired by our brains. In this context Deep=Big — really big — neural networks. Training big neural networks is more complex than traditional machine learning algorithms. But neural networks can generalize to any linear function and thus scale well with increasing data.
Deep learning uses very big neural networks and super-fast computers (graphical processing units) to solve complex problems using lots of data. Performance gets better as more data is added.
Really, that’s all there is to it.
So where do you start?
My simple explanation begs additional questions.
How much data is “lot”?
Should your team use traditional machine learning or deep learning?
What problems can machine learning solve?
These are the wrong questions for most business applications. A better framework is thinking in terms of supervised learning techniques to create your strategy.
Subscribe to my updates or just email me at kevindewalt@kevindewalt.com if you want to talk about your business.
