Monday, March 24, 2014

More Data

Pushing software engineering boundaries can involve many different approaches. There is no overarching way that always leads to success.

If the fields of robotics and Artificial Intelligence (AI) more data and in particular better quality data can trump a lot of work on sophisticated algorithms.

We have a project that needs to grab all of the potential efficiency gains there are going. One obvious candidate is to implement a predictive text feature - you know, where the software predicts what you are going to type next and offers one or more suggestion. However some careful thought showed that this might not be the low hanging fruit it first appeared to be. It will be vital in many instances that any text suggestion be correct - not making a suggestion would be better than offering the wrong one. This is an area where extreme accuracy is a key safety issue with, potentially, people's lives on the line. If there can be many very similar words or phrases but in circumstances where those similar pieces of text have different meanings or might describe different things then we have to face up to the fact that we have to be right 100% of the time.

We think this can be achieved by widening the lexical and semantic nets and supporting the analysis of a great many words and phrases to achieve the objective.

I am sure we will post more on this topic as we develop and refine the techniques we end up applying - in the mean time it is another area we need to carefully research. Others may have trod this path before us. Plus we are going to need a lot of quality data to run trials and tests on.

No comments:

Post a Comment