Created by Google Brain and initially released to the public in 2015, TensorFlow has become ubiquitous in little time. The open-source library has a bundle of ML and deep learning models and neural networks and uses both Python and JavaScript to help developers and ML engineers build front-end API for applications. These repositories each serve a separate function from running ML models from the cloud to even the smallest microcontroller devices available.
TensorFlow Lite
TensorFlow Lite is an open-source and product-ready deep learning framework that can convert a pre-trained model in TensorFlow into a custom model that can then be optimised for speed or storage. The model can be deployed on edge devices that are light-weight like mobile phones supported by Android or iOS, devices like Raspberry Pi that are based on Linux and even microcontrollers.
The special model is also deployed on the edge device after which the inferences are made on the device ensuring safety concerns around data piracy.
TensorFlow Federated
An open-source ML framework, TFF, or TensorFlow Federated, was developed to help open up research and experimentation around Federated Learning. The library will offer researchers starting pointers and complete examples for research work in different areas under federated learning.
Federated Learning or collaborative learning is an approach in ML that builds one robust single ML model without actually sharing the data which upkeeps data security and distributes data rights evenly. For instance, federated learning has been used to train prediction models for mobile keyboards but without uploading sensitive typing data onto servers.
TensorFlow Fairness Evaluation and Visualisation Toolkit
When considering the impact that AI has had on people, this toolkit serves a valuable purpose. The toolkit analyses the binary and multi-class classifiers in the datasets for fairness and checks for bias. The toolkit first evaluates the distribution of datasets and determines how well the model is performing across defined sets of users and searches for the root causes for the biases while also suggesting the areas of improvement.
The toolkit now also has an evaluation library which is agnostic to the model under evaluation and can be used for non-TensorFlow based models as well.
TensorFlow Rust
With Rust becoming increasingly popular, TensorFlow toolkit can help developers run Rust in production on micro-devices and while deploying neural networks at edge devices. However, the toolkit comes with the caveat that the project is still being developed and does not have the guarantee of a stable API.
The combination of Rust and TensorFlow together is potent because of how easily it can design and train custom models using the bindings available in the toolkit.
TensorFlow Quantum
The TFQ or TensorFlow Quantum toolkit is a quantum ML library for rapid prototyping of hybrid quantum-classical models. Within the TensorFlow toolkit, researchers in quantum algorithms and applications can leverage the quantum computing frameworks.
TFQ focuses on quantum data and integrates these algorithms and logic built in the Python framework ‘Cirq’. It then gives quantum computing primitives that will be compatible with existing TensorFlow APIs and high-performing quantum circuit simulators.
TensorFlow TensorBoard
The TensorBoard toolkit is for visualisation and measurement that developers need in the midst of their workflow. The toolkit tracks experiment metrics like loss and accuracy of the model, visualising the model graph and projecting graph embeddings to a lower dimensional space.
There are separate dashboards in the toolkit for different functions—the Scalars dashboard tells how the loss and metrics change with each epoch and tracks training speed and the learning rate of the models, the Graphs dashboard visualises the model and so on.
TensorFlow tfjs
The TensorFlow.js toolkit, or tfjs, is an open-source JavaScript library that is hardware-accelerated for training and deploying ML models. Tfjs can help developers build ML models in JavaScript and use ML directly either in the browser or in the popular Node.js.
Tfjs can also retrain the pre-existing ML models using sensor data that is connected to the browser or other client-side data.
TensorFlow tfx
TensorFlow Extended is a serving toolkit for ML models that is made for production environments. Tfx helps developers build automated end-to-end pipelines for their ML models.
Tfx saves developers from a manual, tedious process that has no tracking or validation measures.
Protocol Buffers
Protocol Buffers, or protobuf, is a Google toolkit for serialising structured data. Protobuf is language-neutral, platform-neutral and extensible, and it is often used in cases where communication protocols have to be defined or even for data storage. The protocol compiler is written in C++, and it currently supports generated code in Java, Python, Objective-C, and C++.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.