We talk about NOMAD™ as currently having the understanding of the world of a smart 13-year old with a near-perfect memory, and making use of that to help your searches. As any good stewards of a teenager, we want NOMAD™ to continually get smarter, and that has been a technology design goal from the beginning.
To that end, in order to have good metrics we have Test and Validation datasets for benchmarks, so we can tell if any change in NOMAD™ increases or decreases our scores. We also have a feedback mechanism for participating sites so we can get human input about the accuracy of results from NOMAD™ searches, enabling us to improve the models.
We also have four big capabilities as part of the NOMAD™ project to ensure increasing value to our users:
- Current events: because we have access to multiple news feeds we can use them to make sure NOMAD™ is up-to-date on recent news, fashion trends, celebrities, etc. This is how NOMAD™ “knows” about COVID, politics, the Olympics and other major events. In a sense, this forces our 13-year old to watch the news and keep up with the world!
- We have developed a “fine tuning” capability for NOMAD™ where the base model can be enhanced, and one way is we take advantage of new publicly available datasets to, as one example, constantly improve NOMAD™’s ability to identify objects. Think of this as taking our 13-year old to museums to learn about a variety of subjects.
- If relevant for a specific user group, we can also use this “fine-tuning” mechanism to educate NOMAD™ on a specific category, like women’s shoes, or car models, or logos, or any other vertical application. This is akin to our teenager taking an interest in a new hobby and getting totally immersed in everything about it!
- Two of the core components of NOMAD™ are a text engine and a visual engine, and because of the modular architecture we used, as better engines in either of those categories become available, we can take advantage of progress in the state-of-the-art! Unfortunately for our 13-year-old, this pretty much amounts to brain surgery: we can remove one section of their brain, replace it with a better performing section, then restore all the experiences our 13 year old has had, so they retain their knowledge but can make better decisions going forward.
This piece was written by David Tenenbaum, CEO of MerlinOne, and an advocate for constant DAM innovation. Connect with him on LinkedIn or email the author directly – firstname.lastname@example.org