`Learning Tracks` intro

November 29, 2019

Making the learning of machine learning practical and reproducible

Core machine learning track

The Core machine learning track is NOT yet another DS/ML course. We do not intend to create substitutes for great ML programs from top universities and companies like Udacity, DataCamp, DataQuest.

Rather we created an end-to-end solution for a typical DS/ML project by complementing DataCamp’s online courses with:

  • guided on-site practice-oriented labs by ML practitioners
  • lectures by university professors on the fundamentals of DS/ML theory
  • our R / Python code libraries for recurring data exploration and prototyping tasks – re-usable template code promotes faster learning and gives DS/ML teams a strong headstart
  • TeamHub knowledgebase that centralizes the codebase & documentation and facilitates sharing of knowledge – it runs on Hugo, a popular, fast open-source static site generator.

Deep learning fundamentals

Deep learning is a very dynamic field of ongoing research where the biggest DS/ML changes are happening. New promising libraries and algorithms are rolled out to the community year-round. Albeit still based on experimental techniques and frameworks that might go away in a few years, deep learning is transforming multiple industries.

Our Deep learning fundamentals track builds a solid foundation for computer vision and natural language processing. Trainees will study DL fundamentals, understand modern architectures and learn to build systems in CV and NLP using PyTorch.

Custom training programs

Like in our other learning tracks we adhere to project-based learning approach which means the predictive models that trainees will have built at the end of the custom training programs should be production-ready. For that matter, firstly we take sample datasets with real data, build a solution and then teach it to trainees.

While a custom training program can span 3-8 weeks, its preparation can take a comparable amount of time prior to commencement of the program. Thus, custom training programs are essentially tailored solutions. They will be customized to specific data and business requirements.

The following topics – clustering, association rules, sequence rules, graph/network analysis and customized deep learning traning – are available as custom training programs.

Please contact us for more details.

Why our training programs?

Democratize

We believe that DS/ML should be technically accessible to more people than a few math-savvy data scientists. We also believe that business units can contribute a lot to DS/ML projects upon learning how to perform data exploration and prototyping (which often takes 70-80% of entire project time).

Demystify

Rather than researching scientific papers, practical ML should be more about putting tutorials into practice, re-using template solutions, rigorously testing, carefully calibrating and interpreting the models.

Grow the talent

Talent scarcity in DS/ML is a major issue that will not cease in foreseeable future. It actually will become a bigger issue. Talent is prone to migrate, seeking ever more interesting and challenging opportunities in global ML & AI environment.

Given lack of skilled data scientists worldwide, training people in-house is perhaps the only viable option to outside hiring. From a global perspective, there are only a few thousands of high-class deep learning researchers and perhaps a million data scientists in the world. However, there are circa 21 million developers who are well positioned to learn DS/ML, and at least as many product owners, project managers and business analysts who can contribute their business domain expertise to DS/ML projects.

Due to upcoming proliferation of DS/ML instruments into mainstream software development frameworks, programmers and business analysts will be expected in not-so-distant future to have good command of the basics of machine learning.

Reproducibility

Work reproducibility is critical both for successful teamwork and cost control. Using template solutions from a common codebase, your DS/ML team members won’t have to spend hundreds or thousands of hours on recurring tasks and problems that have been solved before by their colleagues. Thus more projects can be done with available resources. We offer TeamHub as an easy-to-maintain knowledgebase around IT projects that will help to streamline sharing of information across teams.