Risk glossary


Machine learning

Machine learning is the current application of the concept of artificial intelligence. Machine learning is a process by which software develops its own models to analyse a given set of data. Models are usually statistical and are used to classify or regress data. They are constructed by adjusting parameters so as to minimise a specified error function.

Machine learning can be split into supervised learning and unsupervised learning. In the first family of techniques, which include Bayesian regression and random forests, a machine is given a set of example data and told what the desired output is, so it can work out a way to derive that solution from the data. In the far less common unsupervised learning, a machine is given complex data and instructed to find a pattern in it, completely unaided. Unsupervised learning might involve determining the distance – or the similarity – between data points according to some metric and grouping them accordingly.

Deep learning is a qualitatively different technique, which can be both supervised and unsupervised. Neural networks, inspired by the workings of the brain, are key deep-learning models. They consist of a system of neurons, which are individual processors, connected by flows of data. Each neuron takes input data and performs a non-linear transformation on that data. It then passes on that transformed data to the next column of neurons. A neuron gives a weight to each set of data being passed to it by a different neuron. These weights are calibrated according to the entire system’s performance – neural connections that increase performance are given greater weight while those having less of an effect are given smaller weights.   

Another technique is reinforcement learning, which tries to train the machine, through a large number of simulations, to choose the best course of action in a particular environment. For example, once trade volume or volatility reach a specified threshold, the machine trained in this way knows to trade certain amounts of an asset so as to achieve a goal, such as minimising transaction costs.

Click here for articles on machine learning. 

  • LinkedIn  
  • Save this article
  • Print this page  

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an individual account here