Machine learning is one of the fastest growing areas in the technology sphere. It is best described as a branch of artificial intelligence that is capable of identifying patterns and algorithms in large-scale datasets and quantifying this information to predict future patterns. So therefore, the more data that we feed a computer engaging in machine learning, the ‘smarter’ it will become.
Machine learning is revolutionising multiple industries; from the corporate workplace to climate research and health care. A common example of machine learning is the spam filter built into every email inbox. The system is able search through a huge database of emails whilst continuously getting better and better at detecting what is spam and what isn’t. The finance industry is another growth area that has embraced machine learning – with systems in place to analyse markets and make up to millions of trades in one day. This is known as ‘high-frequency trading’ and only possible through updated real time algorithms.
Certain areas of health care and biotech, such as neuroscience, are applying big data solutions to learn more about areas of study. In regards to neuroscience this process is allowing researchers to gain detailed and high level insights into the complexity of the brain. Currently, neuro-imaging provides a way to see what parts of the brain react to specific stimuli at certain times; however, using natural intelligence to decode the complex patterns and activity of the brain may prove inefficient. While the human mind can observe an MRI and observe what parts of the brain are engaged in activity, machine learning may be able to see what areas are working in conjunction with one another and the modes of connectivity. Machine learning’s immense ability to distinguish and decipher complex neuroscientific patterning means that we are now able to begin to predict what people are looking at simply by observing images of their brain.
The Two Outcomes of Machine Learning in Neuroscience:
1) Encoding: looking at a stimulus and predicting what brain activity will occur
2) Decoding: examining patterns in the brain and predicting the stimulus that the participant is looking at
Brice Kuhl from the University of Oregon has been studying MRI images of participants who, whilst involved in a study, where asked to recall faces that had been shown to them previously. Using machine learning they were able to recreate approximations of these faces, looking only at the MRI images. Firstly the relationship between brain activity and images with just basic components of faces, known as ‘eigenfaces’, was mapped. Recreations were then able to be made using this algorithm and just the MRI images of participants. The image below shows face reconstructions using two areas of the brain, the occipitotemporal cortex (OTC) and angular gyrus (ANG).
While it may seem like mind reading, this type of machine learning known as decoding, is actually more of an interpretation of the brain. The main features of the faces that came through in the recreations were skin tone, smile and gender, showing us the sections that we may commit to memory. The study further revealed that ‘memory’ is not a separate part of our brain, but that we actually access the same areas when recalling a memory to those that we used when creating the memory, or seeing the image for the first time. Neuro-scientific research, the study of the brain, is still an area that requires extensive work and understanding. The brain is arguably one of the most important and most complex organs in our body however we cannot predict the outcomes of brain injury with certainty even with neuroimaging diagnostics such as MRI’s.
Machine learning could reveal insights into how the brain works and how head injuries and mental health issues specifically affect functioning within the brain. These insights could aid in more efficient diagnosis and hence provide more definite and specialised treatment plans. The collaborations between neuroscience and machine learning are still in the early stages of development and much like machine learning itself, will no doubt advance exponentially in the coming years. If we consider that machine learning takes the emotional factor out of research and that all humans are consciously as well as unconsciously affected by our emotions, then the AI vs. natural intelligence debate has definitely got a head start.
While ethics, practice standards and other methods exist so that research takes the most objective viewpoint, human error and bias may always be a set back in any area of research. These enquiries raise a plausible question... With the core concept of machine learning being the ability to analyse and predict outcomes, is it then possible that machines will soon be able to understand us more than we understand ourselves? Here at Protogen we are constantly looking at the way technological advancements such as machine learning can transform the world and the way we do things. Check out our success stories and incubation for our projects in the health care field.