Machine Learning Advances

Machine Learning Advances 2

 

By: Director of Strategy, Tim Dunn

If, as seems likely, the future of work and technology is about automation, the jobs we used to perform with our arms and legs will be increasingly taken over by autonomous vehicles, smart robots, and various forms of sensors.

 

Then what of the brain? How will tasks that rely on thinking, perception and intuition be replaced and augmented in years to come?

 

Some of these issues were approached by Etienne Bernard, a French statistical physicist and machine learning scientist now with Wolfram Research in Boston.

 

His talk aimed to illustrate some of the fundamentals of machine learning, illustrate how it can be applied in its current state, and provide some practical demonstrations of machine learning at work.

 

Helpfully, he outlined three ways to ‘get going’ in machine learning:

  • the Hard Way: which involves getting a PhD in data science and dedicating your life to science
  • the Regular Way: learning to implement current libraries such as Scikit and Weka
  • the Easy Way: using automatic tools for model and parameter selection, preprocessing and feature extraction

 

To illustrate some of the current and potential uses of machine learning, Etienne brought in some data sets, and wrote simple machine learning commands in real time to perform some quite complex tasks.

 

A CSV of bike hire data was used to predict future demands, and a variety of the data attributes were used to model outcomes with their levels of effectiveness assessed.

 

And in real-time, the program was taught to recognize 4 legendary beasts: the griffin, centaur, dragon and unicorn, and then learnt to tell them apart in a base of hugely varied types of image with a high degree of accuracy.

 

But as well as providing answers, machine learning is equally useful at self-diagnosing, or training its users on why it may not work and need improvement. The simple example of a Confusion Matrix below illustrates areas where the algorithm has most problems making a decision, leaving human users or a more sophisticated algorithm the opportunity to improve the system.

 

Machine Learning Advances

While the science is still developing at an academic level, it’s not hard to see the implications for all areas of digital marketing.

More News

| 11th Dec 2018

What Marketers Can Learn From Burger King’s Geo-Conquesting Strategy Against McDonald’s

Who would've predicted that the physical space that your brand takes up could be up for grabs to competitors?

| 10th Dec 2018

Curriculums Aren’t Just for School Anymore

Selections from the CTO’s library and Isobar's core curriculum.

| 6th Dec 2018

The Move to Focused Markets

IGNITION 2018 - Day 2 Recap