We do data science.
What is deep learning?
Deep learning is about taking pure raw data, that has previously been interpreted by humans, and corresponding actions, and training computers to perform that same task, by drawing from examples.
Deep learning systems:
Differ from typical computer programs in that they can learn by example, rather than depending on hand-crafted rules.
Are composed of layers of simple processing units (known as ‘neural networks’) that work together to learn useful representations of data.
Are able to find their own important features in data, and (by intelligent use of GPU hardware) perform those tasks just as well as, and sometimes more effectively, than human experts.
They use a set of examples to tune the network to perform a particular task, without the need for task-specific programming. This computational model is very general, powerful, and can be applied to a wide variety of problems.
It performs particularly well in tasks such as classification or predictive modelling.
Who can benefit from deep learning?
We are only just beginning to discover the myriad of applications for deep learning. It is starting to be applied to medicine (assisting diagnoses), finance (stock market trending), commerce (customer recommendations), automotive (self-driving cars) and many other fields.
Essentially, deep learning could be of use to you if to have:
Raw data that you would like to understand and derive insights from, such as:
- Images taken out in the field in various formats,
- Pieces of geographic information representing assets,
- Timeseries data of particular events.
Concrete actions you’d like to take based on that data, for example:
- Identify objects/words in images to categorise assets,
- Determine where new assets should be placed or built,
- Predict future data points.
Why use deep learning?
Deep learning systems can be used to:
- Generate insights from datasets which were previously inaccessible, such as images and text, and
- Build reliable and scalable systems that perform at the level of human experts.
Traditional machine learning techniques are suitable for use with highly structured datasets, such as databases or spreadsheets. However, many valuable datasets do not fit this mould, including images and natural-language text.
In order to extract insights from these datasets, a new approach that can work with raw, unstructured datasets is needed.
Deep learning can bridge this gap, by enabling computers to learn meaningful representations of raw data on their own. This allows high performance models to be developed without requiring labour-intensive, and often unwieldy, hand-crafted sofware.
While the theory of deep learning has existed for many years in its current form, there has been a resurgence of activity recently as more powerful computer hardware (GPUs) and larger datasets used to train models have become available.
As a result, deep learning methods have recently achieved state-of-the-art (in some cases human-level or better) results in a large and varied number of fields, including image processing, natural language processing, time series modelling and predictive analytics, as well as game-playing and decision making.
How to get started?
If you have a problem you think deep learning may help solve, existing processes that could be more efficient, or simply a lot of data that could be put to better use - contact us (03 9008 5922). We’d love to throw around ideas over a coffee with you.