Exploring artificial intelligence and machine learning

Categories: Machine Learning | Technology |

Last Belatrix attended the XXI CIARP

Last week Belatrix attended the XXI Iberoamerican Congress on Pattern Recognition (CIARP) at the Pontifical Catholic University of Peru, one of the leading Peruvian universities. The event explored artificial intelligence and deep learning, and how they can be applied across different industries.

The event covered a wide range of topics from the challenges of using large data sets, to biometrics and authentication, to healthcare and improving medical diagnostics, to the future of artificial intelligence. I want to share some of my highlights of the event:

  • The event started with Xiaohui Liu, Prof. of Computing, at Brunel University London, examining the history of data analytics. Over the past few years, a gap has emerged between the data generated and our ability to comprehend the data- the so-called “analysis gap”. This gap is not new, but with the explosion in the quantity of data being generated, it is arguably getting worse. Xiaohui Liu also pointed out that there is a real crisis at the moment in science with regard to reproducibility. There is a very significant replication failure rate for experiments, and poor data analysis is one of the top three reasons behind this (the other two being selective reporting and the pressure to publish)
  • The final keynote of the event was presented by Prof. Yann LeCun, Head of AI at Facebook. Prof. LeCun took the audience on a journey through the world of AI, from the very early beginnings in the 1950s and how neuroscience inspired the work on machine learning and AI, to where we can expect to end up in the coming years. The rapid advances we have seen in recent years in AI have all come about in the world of supervised learning – think of this as training a machine to recognize a table or chair. The greater challenge is for the machine to recognize what it has never seen before (unsupervised learning). Beyond that we will look to machines to learn predictive models of the world – Prof LeCun used the example of a pen falling on a table. How do we know which way will the pen fall? There has been some success in using adversarial training where machines are asked to predict the future.

Some of the academic paper highlights at the event provided examples of the wide range of potential real-world use cases that we will see emerging for AI, machine learning, and pattern recognition in the coming years. For example in the world of healthcare, one academic examined deep brain stimulation (DBS) surgery for individuals with Parkinson’s Disease.This surgery involves placing an electrode near the basal ganglia part of the brain, which the medical experts then need to “tune” for it in order to work. In the past it would be tuned using a 3D image that allows the expert to see which brain structures are responding to the stimuli. However this was very time consuming and computationally very intensive. The idea presented at the conference involved using data-driven approximation, to create a method faster and more accurate than existing approaches.

Bringing the possible to real-world implementations

In Belatrix’s recent machine learning survey, 81% of respondents said they thought machine learning would have some or significant impact on their organization within the next 5 years. However at the same time it is evident that while many organizations see the potential, they are struggling to get started. Another survey, by the research company Forrester Research, found that in 2016 among business and technology professionals 58% are already researching AI, but just 12% are using AI systems. It’s our job now as a technology provider, to help organizations in their efforts to implement AI and machine learning, and bridge this gap between what organizations want to do and what they are currently doing.

Leave a comment