ANZ Bank Leverages Neural Networks to Avoid Dangers of Deep Learning

ANZ Bank Leverages Neural Networks to Avoid Dangers of Deep Learning
Share this:

ANZ bank has discovered a better way of predicting which of its clients would default after conducting six different proofs-of-concept. However, the banks said that it would not rush the method into production.

Jason Humphrey, the head of retail risk at ANZ Bank, claimed that before the bank progresses with its neural network-based model, there are several health checks that ought to be cleared.

Even though other neural network users can be contented with not knowing the intricate details about the model, the ANZ bank asserted that it requires to know the aspects that are affecting it the most.

“ It’s actually very hard to do, because you’ve got 70-plus models sitting behind your network, so which attributes are the ones that are the most effective given that one could be across 30 models, another could be across 20- – it’s quite difficult, but we managed to do that,” Jason Humphrey said this to ZDNet during a recent Nvidia AI conference held in Sydney.

According to Humphrey, banks in the United States require to explain the most statistically vital attributes regarding a certain decision being made. In fact, he said, “in a deep-learning environment, it becomes very difficult to work out the factors that were the most predictive for this instance, or for this customer.”

Humphrey added that “before we roll out any deep-learning models, we need to solve for that — even though it is not legislated here. I think it is good practice to be able to know why decisions are being made.”

ANZ bank’s revolutionary system was developed in collaboration with Nvidia and Monash University. The motive behind the undertaking was to deliver a more accurate model for the bank’s risk department, which could be applied in areas such as restructuring existing finance, authorizing individual transactions or customers taking on new loans.

Previously, it took three to six months to create a new risk model. However, the neural network-based system and infrastructure consumed only six weeks, with the model requiring five days to build.

Aside from saying that the system requires to satisfy legal requirements, Humphrey stressed that ANZ bank needs to ascertain that no unintentional biases have been introduced.

In its first-half financial results that were released in May, ANZ reported an after-tax profit of AU$3.5 billion. It also lost more than 3, 000 staff members across its enterprise. Furthermore, both Nvidia and Monash University have cultivated a relationship for several years, and have recently been collaborating on GPU-accelerated research.

Back in 2016, Tom Drummond, a Monash professor, asserted that artificial intelligence(AI) systems required the power to handle rich feedback in a bid to enable systems to learn why answers were incorrect as opposed to the normal binary yes or no responses that are currently used in neural network training.

At the time, Drummond said, “rich feedback is important in human education, I think probably we are going to see the rise of machine teaching as an important field–how do we design systems so that they can take rich feedback and we can have a dialogue about what the system has learned? ”

Source Zdnet 

 

Share this:

Leave a Reply

avatar
  Subscribe  
Notify of