Image
Gebäude

07.01.2021 | News These three trends will characterize the AI year 2021

Transfer learning, cross lingual word embeddings and green AI: IntraFind explains three key AI trends for the coming year.

Artificial intelligence is and remains a hot topic. In 2021, it will once again leave its mark on the IT world. IntraFind, specialist for enterprise search and AI, highlights three trends for text-focused AI and Natural Language Processing that will play an important role.

1.Transfer Learning. After 2020, transfer learning remains a key AI trend this year. Transfer learning is a special method of machine learning and enables neural networks that have already been pre-trained for a specific purpose to be used as a starting point for another task. What has already been learned from a trained network is thus made usable for a new project. This method makes training a neural network much less intensive in terms of computation and time, and the amount of training data required decreases significantly. Transfer learning will further advance the democratization of AI and further accelerate its widespread use in the enterprise world. Until now, AI has only worked particularly well in companies if they have a lot of data - which is rarely the case in the real world - and use customized models.

2. Cross Lingual Word Embeddings. Numerous Natural Language Processing (NLP) applications are only available for the major European languages - often even exclusively for English. One important reason for this is that extending NLP models to new languages usually requires the time-consuming annotation of completely new datasets and is highly CPU-intensive. For less common languages, however, there is often not enough training data available at all. Multilingual models with so-called Cross Lingual Word Embeddings (CLWEs) can help here. These CLWEs take advantage of the fact that many languages have semantic similarities. They capture these similarities and can represent words in multiple languages in a common vector space. 

3. Green AI. Artificial intelligence is becoming more and more widespread in our lives and in the economy. Accordingly, the ecological footprint it leaves behind in the training of algorithms and their use is also growing strongly. Against the backdrop of growing awareness of environmental protection and climate change, green AI is therefore becoming increasingly important. For example, more research is being done on algorithms that require less energy, less memory and less communication bandwidth. The energy supply and efficiency of the data centers used for AI is also coming under increasing scrutiny. Another important aspect of green AI is the use of algorithms to make energy generation, network infrastructure operation and energy use as efficient as possible.

"In the next year, the democratization of AI will continue to accelerate. Transfer learning can simplify many office tasks without the need to collect large amounts of data and develop customized models for each one. We have optimized our products to precisely match operational reality, so that every employee can ultimately improve their performance even without AI experts, with very little time spent on training AI procedures, thus helping the company to save costs and improve its competitiveness," says Franz Kögl, CEO of IntraFind Software AG. "But the topic of sustainability will also gain momentum."