The evolution of augmented reality

X R D S • W I N T E R 2 0 1 8 • V O L . 2 5 • N O . 2 end-to-end learning, but as a byproduct this has given prominence to an equally complex problem of hyperparameter optimization. Finally, you should focus on understanding the underlying technical implementation rather than continuing to use new technology as though it were a black box. For example, decently complex multi-layered deep learning models can be built in Keras with as little as seven lines of code, which greatly reduces the barrier to entry. However, it is really important to know how to troubleshoot and improve the model’s performance. Gaining this knowledge will be the most challenging part of adapting the new technology and will naturally help you cut through the hype. In conclusion, mining reality from hype is a complicated process with no simple solution. What makes it alarming is when people give in to hype and draw faulty inferences. In the context of machine learning, there has been an increasing interest in exploring new methods that can assist law enforcement agencies with predictive policing, judiciary bodies with sentencing criminals, and the intelligence community with AI assistance. However, the limitation that gets omitted during these applications is the inherent bias in machine learning models that might come from a biased training set, incorrectly optimized hyperparameters, or the lack of explainability. Needless to say, these questions involve ethical issues that might get violated when people start believing in the hype. What do you think? Is the recent interest in machine learning hype or reality? Let us know your views. —Sarang Paraf