AutoML, or automated machine learning, has shown great potential recently and has continued to garner further interest amongst the community. AutoML cuts downtime taken to build systems with just a few lines of code. Even non-technical users can build scalable machine learning models. It is a method inculcated with specific search algorithms that help find the optimal solution for each component present in the ML data pipeline.
Here is a list of some of the most interesting AutoML developments and updates of 2021:
FLAML, or Fast and Lightweight Automated Machine Learning, developed by Microsoft, is a Python package that helps develop accurate machine learning models at a low computational expense. It eradicates the worry of hyperparameter tuning and model selection. Consumption of computational resources while tuning a machine learning model puts a huge burden on the environment. FLAML enables its users to build fast and easy-to-use self-tuning softwares that automatically adjust themselves even when new training data is added. It does so by leveraging the structure of search space, to choose a search order that is optimized for both the cost and model quality. The search then tends to move gradually from cheap trails to rather expensive trials, in turn improving the model accuracy. FLAML can be easily installed into Python notebooks using the code pip install flaml.
The Github repository for FLAML can be accessed here.
Edge Impulse is an automated machine learning development platform that helps create models on devices, targeting the microcontrollers that power consumer-based devices such as smartphones. It combines the power of two techniques to enable the development of AI on microcontrollers, namely AutoML and TinyML. Acquiring data from various sensors attached to the microcontrollers in the device, Edge Impulse facilitates management of the entire data pipeline, from data acquisition to model deployment. Its AutoML tool can be easily run on Windows or macOS, and collected data can be ingested into its cloud-based platform. With a push of a button, labels are identified, and a thoroughly trained model is created that is ready to predict. The tool can further optimize the model and convert it to be deployed on the device. Depending on the data format, the developer can choose an appropriate neural network and optimization technique to train the model.
The official documentation for Edge Impulse can be accessed here.
The Databricks AutoML tool brings forth a “Glass Box” approach to automated machine learning. The AutoML tool allows quick generation of baseline models and notebooks. It provides training code for trail runs to jumpstart the development of models, allowing one to get a quick check on the direction the ML project is heading towards. One can automatically set up MLflow integration for experiment tracking and use the best ML practices that are suggested by the tool itself. Supported algorithms include decision trees, random forests, logistic regression, and many more. The UI is hassle-free and easy to use. It also comes with a Python API enabling its use on Python Notebooks; the only limitation of running on notebooks being – support of only two types of models, classification and regression. Image and Text data types are not yet supported for use on Python.
The Official Documentation for DataBricks AutoML can be accessed here.
JADBio’s AutoML platform provides automation capabilities combining with leading-edge AI tools to build and deploy accurate and interpretable predictive models for biomedical data without the need for coding. The company aims to provide research institutes and data scientists with the power to analyze data automatically. The health segment can use this to improve its business decisions, enabling its users to harness decisive insights from data, regardless of their technical expertise. With just four simple steps, a model with enhanced predictive prowess can be developed and deployed. JADBio can be an essential tool in use cases such as deciphering cancer prediction and stool classification through optimization, visualization, and algorithmic machine learning features.
For further information about its features, access the link here.
Domo For Amazon Redshift
Domo recently announced a new native integration designed in collaboration with Amazon Redshift to leverage their AutoML tool capabilities. In partnership with Amazon SageMaker Autopilot, it now allows augmenting analytics with machine learning, enabling sharing of the generated insights with anyone. The SageMaker Autopilot is a solution tool that automatically trains and tunes ML models based on the data provided by the user. Users can now use data from Domo as an input to the SageMaker Autopilot to create high-performance models and deploy prediction pipelines that adapt easily to new incoming data. Hundreds of training jobs on any dataset present in Domo can be launched to achieve the best performance for the task being performed in a quick time. Model performance can also be evaluated using the DataFlow features.
The Domo for AWS website can be accessed using the link here.
Google Model Search
Google has recently introduced its open-source platform for developing optimal machine learning models through automated machine learning. The Model Search System consists of multiple trainers, search algorithms, and a database to store the various evaluated models. The tool automatically runs both training and evaluation experiments for different evaluated models and collects the knowledge gained by these experiments conducted. Instead of focusing on just a specific domain, Model Search enables finding the appropriate architecture that best fits the problem to be tackled, minimizing coding time and computational resources. It is built upon TensorFlow and can run either on a single machine or in a distributed environment.
The GitHub repository for Model Search can be accessed here.
AutoVIML is an open-source Python library created by Ram Seshadri that aids machine learning in Python notebooks. The library renders data through different machine learning models and finds the model that best fits the problem with the highest accuracy for the provided dataset. It removes the time-consuming process of preprocessing that dataset as it automatically cleans and classifies the data in a single model, therefore saving the sheer amount of effort to create machine learning models. The V in VIML stands for “Variable”, as it tries multiple models with multiple features to find the best performing model. AutoVIML also reduces the number of features tuning the model complexity. It can handle text, date-time, and categorical variables, all in one model process.
The Github repository link for AtoVIML can be accessed here.
Automated machine learning tools are necessary for today’s challenging and dynamically changing technology space where time for the deployment of models is a matter of grave concern. Advancements such as these are helping leverage the artificial intelligence space to a better position each day. One will have to wait and behold what the future presents and how the domain of AI spans even further heights.