669: Streaming, reactive, real-time machine learning — with Adrian Kosowski

Transformers

Transformers in data processing play a crucial role in transforming data structures efficiently. Adrian Kosowski explains that these transformers differ from those in AI, like GPT models, as they focus on converting one data format to another within data flows 1. This process is essential for streaming and reactive data processing, allowing for seamless data transformation and logic expression.

A transformer is something that transforms one kind of data into another kind of data. In our case, it's a box which transforms tables into other tables.

--- Adrian Kosowski

Kosowski highlights the flexibility of these operations, which can iterate transformations until convergence, making them adaptable for various data engineering tasks 2.

   

Energy Efficiency

Energy efficiency in machine learning is becoming increasingly important, especially in distributed systems. Kosowski emphasizes that these systems optimize energy by performing local computations, reducing the need for extensive communication and computation 3. This approach mirrors natural systems, like ants, which optimize energy use by relying on external storage, such as pheromones, to minimize internal processing.

The incremental way of computation, and also streaming computations, are more energy efficient.

--- Adrian Kosowski

Kosowski suggests that lightweight models with the ability to add or delete data points offer an edge in energy efficiency, presenting a rich area for further research 3.