The PROBLEMATIC Art of Deep Learning

Jobin Reji – King’s College London

Introduction

Artificial Intelligence has two sub-sectors, as we found out in the last article. General AI and Applied AI, machine learning came under the topic of General AI. General AI is the implementation of computational thinking by computers that a human would have programmed. This concept of human thinking is then further studied and applied to Deep Learning.

Deep learning is a subset of machine learning, essentially a neural network with three or more layers. These neural networks attempt to simulate the behaviour of the human brain, allowing it to “learn” from large amounts of data. A single layer neural network can still make approximate predictions; additional layers can help optimize and refine for accuracy. Deep learning drives many AI applications and services that improve automation, performing analytical and physical tasks without the need of humans stepping in.

Deep learning eliminates some of the data pre-processing that is typically involved with machine learning. These algorithms can process unstructured data, like text and images, removing some of the dependency on human experts. For example, let us say that we had a set of photos of different pets, and we wanted to categorise them by a cat or a dog. Deep learning algorithms can determine which features (e.g. ears) are most important to distinguish each animal from another. The features are established manually through humans if this was done via machine learning.

An introduction to deep learning – IBM Developer
Deep Learning is a subset of Machine Learning consisting of many Neural Networks

Real-Life Examples

Deep learning algorithms can analyse and learn from transactional data to identify dangerous patterns that indicate possible fraudulent or criminal activity. Speech recognition and other deep learning applications can improve the efficacy of investigative analysis. This is done by extracting patterns and evidence from sound and video recordings, images, and documents, which helps law enforcement analyse large amounts of data more quickly and accurately.

Institutions in the Financial sector use predictive analytics to drive algorithmic trading of stocks, assess business risks for loan approvals, detect fraud, and investment portfolios for clients.

The healthcare industry has significantly benefited from deep learning capabilities ever since digitising hospital records and images. Image recognition applications have supported medical imaging specialists and radiologists, helping them analyse and assess more images in less time.

Deep learning is also carried out on our phones. If you can unlock your phone using Face ID/ Facial Recognition, you are using the deep learning tools embedded in the phone. The mobile device links a series of parameters on your face. When all the matches are identified, the system recognises what is stored on the phone and unlocks it. However, if any points are distorted, say you pull a face to prevent the phone from opening. The AI will prevent the phone from unlocking as the parameters are not aligned.

What’s Next?

Deep Learning is not new; it has been in discussion since 1958 where Frank Rosenblatt first designed the artificial neural network. Rosenblatt himself knew that the notion of Deep learning was beyond his time, the technology was not available, and his ambitions were outpaced. Years later, the ability to implement additional neural networks was discovered. Additional neural networks increased the number of computations that a computer could do by 10-million-fold that a computer of that time can do in a second. Scientists then went back to Deep Learning in the early 2000s as the components for this process were now viable. However, deep learning has taken a slower approach in development since then. Scientists have advanced dramatically in machine learning with more models and realities, but deep learning has stagnated. Performance under deep learning can be improved and accelerated; however, the costs to implement this are very high. Are costs to blame for Deep Learning’s hindering progress? The large gap between what is practical and what is theory identifies that there are still undiscovered algorithms needed to improve the efficiency of deep learning. Unless the gap between theory and application is down-sized, deep learning is again cornered to advancements and limited to actual development.

Did you learn anything new? Maybe you have something to add. Comment to see more about Artificial Intelligence, its role in Financial Services and beyond…

Leave a Reply

%d bloggers like this: