Advances and challenges in Machine Learning Languages
Machine learning languages is an emerging topic developed to efficiently tackle machine learning problems that arise as data-size and model-complexity grow. Early languages such as BUGS have been highly influential in disease mapping, ecology, genetics and many other fields. Over the past few years, several (functional, probabilistic, parallel) programming languages and plug-in libraries have been developed. Expressive probabilistic programming languages such as Anglican, Church, Stan, Infer.NET, Turing, and automatic optimisations tools for deep learning, such as Theano, PyTorch and TensorFlow allow not only existing but, in some cases, an entirely new family of models to be expressed and solved.
The identification of advances and challenges in these machine learning languages is crucial from both theoretical (mathematical foundations) and practical (computational resources -- time and space complexity) point of view. The efficiency of the statistical learning and inference depends on the currently available tools and these tools depend on the theoretical advances such as better optimisation methods, identification of new learning algorithms etc.
To this end, we propose to make a workshop that is going to:
1) make an introductory exploration of the available languages/libraries/tools/software;
2) identify the merits and boundaries of each of them;
3) compare and contrast them from the perspective of several contexts -- e.g. theoretical origins,future directions, the capacity of problems etc.
Details on how to register will be available shortly.