Microsoft has recently introduced ML.NET 3.0, the latest iteration of its pioneering machine learning platform that is open-source, versatile and enables integration of machine learning models into .NET applications. Microsoft announced this advanced version on November 27, and developers can easily access it through dotnet.microsoft.com.
The outstanding feature of ML.NET 3.0 is its expanded deep learning capabilities, encompassing object detection, named entity identification, and question answering. These functionalities were empowered by integrations and compatibility with TorchSharp and ONNX models. Additionally, the recent roll-out also enhances the earlier integration with the LightGBM gradient boosting framework.
ML.NET 3.0 strengthens its support for data processing scenarios courtesy of its powerful enhancements and bug fixes to DataFrame. It also introduces new IDataView interoperability features, making tasks related to data loading, inspection, transformation, and visualization more potent than before.
In an update released in May, Microsoft had publicized Object Detection in this ML.NET Model Builder. These features are constructed on top of the TorchSharp-empowered Object Detection APIs, which are a part of the latest ML.NET 3.0. The Object Detection API adopts some novel techniques from Microsoft Research and is backed by a Transformer-based neural network architecture made with TorchSharp. Moreover, this object detection feature is included in the Microsoft ML.TorchSharp 3.0.0 package.
In addition to the aforementioned, ML.NET 3.0 provides notable avenues for natural language processing including robust question answering and named entity recognition frameworks. These scenarios have been unlocked by building top of the existing TorchSharp RoBERTa text classification features introduced in ML.NET 2.0. Besides, the updated version comes with new automated machine learning (AutoML) capabilities with AutoML Sweeper now assisting sentence similarity, question answering, and object detection.
DataFrame has underwent several updates in ML.NET 3.0, amplifying data loading scenarios with data now being importable from and exportable to SQL databases. This is made possible via ADO.NET, which is compatible with SQL-supporting databases. DataFrame has also boosted its arithmetic performance in column cloning and binary comparison scenarios, along with improved null value handling while performing arithmetic operations. This results in fewer steps required in data transformation. Moreover, debugger improvements have been made to ensure a better readable output for grids with extensive column names. It also introduces a new set of APIs to support tensor operations under Tensor Primitives.
Along the same lines as no-code platforms like AppMaster, Microsoft continues to innovate and improve. The tech giant is also concurrently developing plans for.NET 9 and ML.NET 4.0. In the meantime, the company committed that users can expect upgrades for Model Builder and the ML.NET CLI to complement the ML.NET 3.0 release. Microsoft also plans to expand deep learning scenarios and integrations while introducing enhancements to DataFrame. Lastly, the company revealed its intent to expand the APIs in System.Numerics.Tensors and integrate them into ML.NET.
