As artificial intelligence (AI) adoption continues to accelerate across a variety of industries, numerous new concepts and techniques related to how we use AI are gaining popularity. Although there are new approaches being created nearly every day, below we’ve gathered the ten hottest AI trends that data scientists should be aware of.

1. MLOps

While not a new concept, MLOps is still a relatively new operations practice that attempts to identify what does and doesn’t work in a model in order to create more reliable ones in the future.

2. Contrastive Learning

This is a machine learning technique that finds similar and dissimilar things in a data set without labels. Contrastive learning is becoming increasingly useful in image databases to help companies find similar images.

3. Transformers

Transformers are a neural network architecture that, like recurrent neural networks (RNNs), handles sequential input data. Transformers are widely used in language models, including language translation and speech-to-text applications.

4. Carbon Footprint

This might not expressly qualify as a “hot trend,” but as organizations utilize more storage and compute for deep learning, they are simultaneously increasing their carbon footprint. Because this directly conflicts with corporate sustainability imperatives, it’s important that data science pros and tech leaders alike are mindful of this AI pitfall.

5. Financial Inefficiency

In a similar vein, another non-traditional trend on the rise is the monetary cost of machine learning. It’s possible to run a neural network for hours before discovering a problem—leading to wasted manpower and storage and compute costs. As a result, there is pressure on engineering teams in many organizations to find ways to avoid these unnecessary costs and build more efficient algorithms.

6. Graphs

Graph neural networks (GNNs) help make sense of graphs, enabling companies to make node or edge predictions. For example, a pharmaceutical company could use GNNs to determine the potential side effects of a new drug.

7. Intuitive, Integrated Toolsets

Data scientists are increasingly clamoring for a single, comprehensive platform in place of multiple platforms and tools. Integrated tool sets would eliminate issues that arise from transporting data and models between tools and increase output for the data science team.

8. Models that Explain Other Models

As pressure increases on companies to ensure AI fairness and equity, expect to see a rise in AI-driven solutions that can help explain other models. This will make it easier for organizations to understand the underlying reasons their models make certain predictions.

9. Contextualized Word Embeddings

Contextualized word embeddings like ELMo and BERT resolve the issue of semantic dependence on a word based on its context—for example, “bank” in the context of “park” has a different meaning than in the context of “money.” These language models reduce training times and increase the performance of state-of-the-art models.

10. Small Data

We’ve written about the importance of small data previously at the APEX of Innovation, and the trend is still driving enterprise value. Because processing small data and designing the right algorithms necessitates a different approach than neural networks or deep learning, doing small data right means going back to old machine learning and data science concepts.

Head over to TechTarget for more on what this entails, along with additional details on the other trends outlined above.