Calling All Startups: Applications for Microsoft’s Seattle Accelerator are Open

This post was co-authored with Prashant Sharma, Microsoft Accelerator Seattle. Working with cutting-edge machine learning and data science startups over the past four months has been a tremendous learning experience for us at Microsoft Accelerator Seattle. To roll up our sleeves and work at multiple levels—from solving engineering problems to identifying new business values—has been deeply insightful.

Calling All Startups: Applications for Microsoft’s Seattle Accelerator are Open

 

Calling All Startups: Applications for Microsoft’s Seattle Accelerator are Open

Our proximity to bright entrepreneurs and novel ideas gave us a deeper appreciation of the ever-growing technology space. We plan to put our learnings to good use by extending our support to the next generation of startups focused on machine learning and data science in our upcoming program commencing in early September.

Together with our batch 3 startups, we’ve been able to find answers to some of today’s challenges in the machine learning space, which have pushed us to ask deeper questions for the future of this field.

Below are a few of the scenarios we learned a great deal about and generally saw startups iterating in their existing machine learning models:

  1. “Reusable learning,” where learnings from one task were transferred to other related tasks: How much Learning can be reapplied with minimalistic changes? For example, can virtual medical assistants answer questions related to a particular skin condition in Germany and train the model for a large population size in Columbia? The situation becomes subtle when function transfer is more complex. Most startups we worked with were able to effectively transfer across industries and geographies.
  1. Diminishing cost of supervised learning: Supervised learning, by design, is about estimating functions and their relationships with training samples. Contrary to conventional wisdom, we saw many of our participating startups ‘training’ their supervised learning models with unstructured data to a fairly remarkable level of accuracy.
  1. Strategies of collecting training data: Many of the participating startups found creative ways to find (free) training data from open sources and APIs available to tap into and train their models.

Through working with next generation of startups in our fall 2016 program, we hope to help entrepreneurs further explore and address the following challenges:

  1. Commercially scalable, perpetual learning engines: When and how will we see commercially viable, perpetually learning engines? What will they look like? When will they “outsmart” humans in certain areas/industries?
  1. Cross domain generalization: Humans can generalize (across domains) with ease. Machines can’t. The ability to look at domains horizontally is an amazing human asset, but machines don’t have that ability (yet).
  1. Reinforced learning with structured time: Real world systems have memory and concurrent time scales – fast and slow. We need to learn at multiple time scales with rich structure of events and dimensions. Next generation innovation will attempt to solve RL problems at finer scales.
  1. Self-debugging programs: Every time a programmer fixes a bug, we potentially have a piece of training data. Next generation ML will auto-record the edits, debugging traces, and compiler messages to a database that will contain millions of ‘objects’ – all focusing on debugging.
  1. Mimicking human learning and the emergence of cognitive computing: When will machine learning reach a stage where machines mimic the human brain? The question is no longer if but when.
  1. Computer perception and machine learning: Given the increasing use of machine learning for state-of-the-art computer vision, computer speech recognition, and other forms of computer perception, can we develop a general theory of perception grounded in learning processes?

Calling All Startups: Applications for Microsoft’s Seattle Accelerator are Open

Do you have a technology solution related to some of these challenges? We’re opening applications for startups that are trying to answer these and other questions by leveraging the power of data science, machine learning, analytics, and enterprise IoT.

The application period for our fourth Seattle cohort is now open and will close on July 20, 2016. We are looking forward to reviewing all of your great ideas and working with some of the amazing entrepreneurs behind them – apply now!

Keep up-to-date with accelerator news, program deadline reminders and more by following us on Twitter.