Lyft's self-driving division has developed a framework for machine learning-based solutions in autonomous vehicle prediction, planning, and simulation. The framework includes datasets for training models, the L5Kit library for converting driving scenes into machine learning problems, and examples for visualization and training. The software is open to external contributors and aims to enhance self-driving technology through data-driven approaches.

Dragonfly is an open-source Python framework for scalable and robust Bayesian optimization, developed by researchers from Carnegie Mellon University. The framework is designed to optimize expensive functions with high efficiency and has outperformed many existing systems. It supports Python 2 and 3, is compatible with Mac, Linux, and Windows, and can be installed via PyPI. Dragonfly offers scalability for handling higher dimensional domains and robustness by considering a set of values for acquisition choices and model parameters. It can be used through the command line, Python code, or an ask-tell interface, and includes features for multi-fidelity evaluations, parallel evaluations, and evolutionary algorithms.

AI Fairness 360 is an open-source toolkit developed by IBM researchers to detect and mitigate bias in machine learning models. It provides a comprehensive set of bias metrics, mitigation algorithms, and explanations to help create fairer AI systems. The toolkit, available in Python and R, includes an architecture for dataset representation, algorithms for bias detection, mitigation, explanation, and an interactive web interface. The article also covers a tutorial on mitigating age bias in credit decisions using the German Credit Dataset.

AutoGL, developed by Tsinghua University, is an AutoML framework designed for graph datasets, handling various stages from feature engineering to model ensembling. The article provides a tutorial on installation, dataset import, feature engineering, model selection, training, hyperparameter optimization, and ensemble methods. It also previews upcoming features such as Neural Architecture Search and support for more graph tasks and libraries.

The Reformer is an efficient Transformer model capable of handling context windows of about 1 million words with efficient memory usage. Developed by Google researchers Nikita Kitaev, Łukasz Kaiser, and Anselm Levskaya, it was presented at ICLR 2020. The Reformer introduces techniques such as Locality-Sensitive Hashing Attention, Chunked Feed Forward Layers, and Reversible Residual Layers to optimize memory requirements without significant performance loss. The article provides a step-by-step tutorial on using the Reformer in transformer models, including setting up the environment, installing libraries, and training the model. It concludes that advancements in transformers can lead to state-of-the-art technology in various domains.