Page 7 - AeM_January2021

RESEARCH, ANALYSIS & TRENDS

AI trends that

will shape

2021

Artificial intelligence is and will remain a hot topic and
continue influencing the IT world. IntraFind, a specialist
in AI, highlights three trends for text-focused AI and nat-
ural language processing (NLP) that will play an im-
portant role.

Transfer learning

Transfer learning is a special method of machine learn-
ing. It enables neural networks that have already been
pre-trained for a specific purpose to be used as a start-
ing point for another task. In other words, what has al-
ready been learned from a trained network can be used
for a new project. With this method, training a neural net-
work is significantly less computationally intensive and
time-consuming, and the amount of training data re-
quired is significantly reduced. Definitely, transfer learn-
ing will advance the democratization of AI and accelerate
its widespread use in the corporate world.

Cross Lingual Word Embeddings

Numerous Natural Language Processing (NLP) applica- therefore getting more attention. For example, there is
tions are only available for the most important languages increasing research into algorithms that require less
- often even exclusively for English. An important reason energy, less memory, and less communication band-
for this is that the expansion of NLP models to new lan- width. The energy supply and efficiency of the data
guages usually requires the time-consuming annotation centers used for AI are also being scrutinized more and
of completely new data records and is very computation- more often. Another important aspect of green AI is the
ally intensive. For lesser-used languages, however, use of algorithms to make energy production, the oper-
there is often not enough training data available. Multilin- ation of the network infrastructure and energy use as
gual models with so-called Cross Lingual Word Embed- efficient as possible.
dings (CLWEs) can help here. These CLWEs take ad-
vantage of the fact that many languages have semantic “The democratization of AI will continue to gain mo-
similarities: they grasp these similarities and can repre- mentum. Transfer learning can simplify many office
sent words in several languages in a common vector activities without having to collect large amounts of da-
space. ta and develop customized models for each of them.
We have optimized our products to reflect operational
Green AI reality so that every employee can ultimately improve
his/her performance with very little expenditure of time
Artificial intelligence penetrates more and more areas of for training the AI processes and thus help the compa-
life and business – leaving an ecological footprint when ny to save costs and increase its competitiveness",
training algorithms and tendency to rise due to growing Franz Kögl, CEO of IntraFind Software AG, states. ◊
use. Against the background of increasing awareness of
environmental protection and climate change, green AI is By MediaBUZZ

January 2021: AI & Automation in Marketing 7